Last week, I published a review essay about Michael Pollan’s new book, A World Appears: A Journey into Consciousness, arguing for pluralism when it comes to thinking about consciousness in humans and machines.

𝐀𝐈 𝐋𝐨𝐠 reviews books

Here I argue for pluralism when it comes to teaching with and about large language models. These are remarks prepared for a panel titled “The Future of AI and What it Means for Higher Education” at the capstone event for the American Association of College and University’s 2025-2026 Institute on AI, Pedagogy, and the Curriculum. Applications for next year’s Institute are now open.

I’m grateful to , our moderator, and the other participants on the panel for their ideas as we prepared. The panel was fabulous, with a lively discussion happening in chat alongside the conversation on screen.

If you think it’s important to teach students how to use AI, that’s great. Go for it. If you want to teach students how not to use it, I think that’s great too. Again, go for it. Students are well served by a diversity of approaches to using AI (or not!). Encouraging discussion among those with different ideas is a better institutional approach to the intrusion of technologies we call artificial intelligence than mandatory AI literacy workshops and one-size-fits-none policies.

This plea for pluralism comes from a sense that battles between enthusiasts and critics of AI are not that important relative to larger problems. I’m talking about the sense that something is off in the functioning of institutions of higher education, and has been for a while now… Since COVID? Since iPhones? Since I was a student and life was good? The intrusions of AI intensify the feelings, providing a focus for unease or outrage. But most teachers and students feel overwhelmed, not by AI, but by all of it: the grind of the end of term as classroom joys fade into grading and being graded; the headlines describing how systems of higher education are under attack from the governments that have created and funded them; the depressing emails from administrators about belt-tightening.

Nothing I say here is meant to undermine a sense of solidarity in the face of external threats. Those who work in higher education and believe in its value should unify around academic freedom, public support for research, and the safety and success of students. But the need for unity should not extend to whether and how teachers use the technology.

Uniformity of practice is not possible with the tools we use to teach because what we do as scholars and teachers is so various, and so is what AI technologies offer. The choices we make as educators about AI lead to conflict because AI is contentious. Some of this conflict is moral: objections to technology so transparently aimed at replacing humans. Some is political: objections to building data centers or the disregard for the rights of working artists and writers. Some is educational: objections to asking students to use these tools because of their potential harms or worries about deskilling.

Those conflicts are compounded by methodological diversity. Each of us teaches within different disciplines, sharing knowledge and methods of knowledge-making with our students. Scientists I know speak gratefully about how easy it is now to write up lab results. Writing teachers I know are despondent over how easy it is now to generate a first draft. Put these folks in a room together and sparks fly. Those examples cross disciplines, but conflicts within departments and programs are often more intense; they are also where the important work happens.

I may view reading and writing as fundamental to my history seminar and so ban the use of AI models for any purpose. My colleague down the hall may use AI models to develop simulations and games in an effort to make history feel participatory and alive. Those different practices will conflict when we meet to make curricular and educational decisions for our department and programs. These are not fights to be won; they are differences to be valued and worked through within each department and program.

I don’t mean to dismiss the need for some institutional coordination. But beyond the evergreen goal of getting teachers to be transparent about their expectations for students and establishing clear institutional guidelines to protect student data, there is little that institutions must do about AI. The conflict and confusion that comes of teachers and scholars finding their way with or without AI is fundamental to free inquiry.

The pluralist approach to AI means that figuring out when and how to use AI models is embedded in the collaborative work of scholarship and teaching. This is as it should be, but it will likely fail in two ways.

First, the goal of managerial efficiency will combine with delusions that technology will reduce costs. Such delusions are fundamental to modern educational bureaucracies, even though spending on technology never saves institutions money, and will not in the case of AI. Such hopes are entangled with a trained incapacity, what Thorstein Veblen called “an habitual, and conventionally righteous disregard of other than pecuniary considerations.” Those managing colleges and universities are unable to understand why replacing teaching assistants and contingent faculty with AI products cannot possibly work when their spreadsheets say that it will.

Second, the difficult work of updating departmental curricula will lead departments and programs to ignore AI, leaving it to each individual teacher to figure out. Many of these teachers are contingent faculty and so will have to muddle through on their own, with little or no support. AI is just one more straw placed on the back of those who carry the weight of teaching without extra support or resources to update their practice.

I wish there were easy answers to mitigating those two points of failure, but AI is a complication that arrives within a cluster of complications. If there is one idea that should unite faculty, it is to resist the demand to automate teaching…not to save jobs, but to protect the essential human purposes of learning.

Let me close with one last thought on the value of pluralism. It has to do with fallibilism, the idea that what we know is never certain and should be revised in light of new evidence. Fallibilism pairs well with pluralism as a guide to action. Bringing both values to bear on questions of curriculum and AI will help set the stage for collaboration and productive disagreement about using (and not using!) these weird new language machines to teach.

Nonacademic higher learning?

I’ve written before about Slime Mold Time Mold and other non-institutional forms of higher learning. Here is a new effort that has me intrigued.

Elf Theory

Special announcement

Today is my birthday. It’s been a hell of a week; most people I know are exhausted. I’m not immune to the shock of the here and now, but as I turn 37, my mind is with the future. We still need to create it. That’s the idea behind this post…

Read more

19 days ago · 27 likes · 3 comments

AI Log, LLC. ©2026 All rights reserved.

—William James

This is William James’s brain on drugs:

William James did not much enjoy the effects of taking drugs. His one experiment with peyote led to a bad headache, not mystical insight. There is no evidence he tried marijuana. He found drinking alcohol unpleasant and mostly avoided coffee and tobacco. Nitrous oxide was the exception, and what an exception it was. The poem above was written as he was coming down from its brief but intense effects. InVarieties of Religious Experience (1902), he says that his experience inhaling the gas taught him “that our normal waking consciousness, rational consciousness as we call it, is but one special type of consciousness, whilst all about it, parted from it by the filmiest of screens, there lie potential forms of consciousness entirely different.”1

James came to his experiments with nitrous oxide through his scientific study of the human mind. This was at Harvard in the 1880s, when he oversaw the first psychology laboratory in North America and wrote the monumental Principles of Psychology (1890). It was in Principles that he analogized cognition to a river, what he called “the stream of thought, of consciousness, or of subjective life.” This metaphor is the centerpiece of his comprehensive descriptions of the human mind as produced through natural processes, as Darwin proposed. Yet, James understood that consciousness itself remained beyond the “darwinizing” principles he outlined in the book. If the brain is the organ where the stream of consciousness flows, then all natural science can do is look for evidence of it there. Absent such evidence or a credible explanation for its absence, consciousness is a matter for metaphysics.

Cognitive science, as the field James helped found has come to call itself, has not found much evidence of consciousness in the brain nor a credible reason for why it’s not there. And those who study cognition have mostly abandoned James’s stream for a new analogy, one Michael Pollan calls “the reductive faith of our time—the belief that the brain is essentially a computer and that conscious awareness emerges, somehow, from the processing of information.” For the past seventy years, this belief has functioned as something like a paradigm for the study of the human brain, but as Erik Hoel, writing at , explains, neuroscience is best understood as pre-paradigmatic because of consciousness. Likewise, , writing at Experimental History, argues that psychology is actually just a couple of proto-paradigms that were once useful but no longer are, and a third, which he calls “pick a noun and study it,” that is best understood as a method for generating publishable papers.

A grand collection of theories

Neither of these one-time academic insiders—now outsiders writing successful newsletters—appears in A World Appears: A Journey into Consciousness (2026). Instead, the philosopher David Chalmers, whom Pollan calls “something of a killjoy, though an unfailingly polite one,” pops up periodically to explain why this or that theory that has just been put forth by some cutting-edge consciousness researcher or philosopher of mind seems implausible or just plain wrong. So no paradigm…just theories in search of evidence.

And what a grand collection of theories there are! Pollan only has space to explore about a half dozen, but he points to a paper describing “no fewer than eighty-four non-physicalist theories of consciousness” and references another that presents a selection of twenty-two candidate theories. No one could call consciousness under-theorized, yet all this activity without anything like an agreed-upon set of definitions, measures, or methods for studying it is an indication that James had it right.

In the years after he gave his lab over to Hugo Münsterberg, James continued to think and write about the metaphysics of consciousness. In Does Consciousness Exist? (1904), James answers his question with an emphatic No! “It is the name of a nonentity, and has no right to a place among first principles. Those who still cling to it are clinging to a mere echo, the faint rumor left behind by the disappearing ‘soul’ upon the air of philosophy.”

Consciousness is not an entity. Rather, James argues, “there is a function in experience” that goes by that name. According to James’s dismantling of Descartes’s dualisms into relations, we live in a world of pure experience, a world where it is experience, not turtles, all the way down. Understood in this context, consciousness is a function, and, for James, “That function is knowing. ‘Consciousness’ is supposed necessary to explain the fact that things not only are, but get reported, are known.” The essays collected in Essays in Radical Empiricism (1912) are answers to the mystery of his drug-induced question, “What escapes, WHAT escapes?”2

Consciousness is a function in experience and that function is knowing. What this means and how it might apply to humans, animals, plants, and machines has occupied writers on consciousness ever since, though they are not always aware they are working downstream from James. Pollan acknowledges his debt to James, and an awareness of the limits of scientific inquiry provides context for Pollan as he writes about the topic. The different vocabularies used by all the people he talks with is a challenge, of course. Even within cognitive science, people talk past each other, using the same terms to mean different concepts. Add in biologists, novelists, philosophers, and shamans, and Pollan has a lot of translating to do.

Pollan takes this semantic confusion on directly by organizing the book into sections using the key terms sentience, feeling, thought, and self. The first section is an overview of plant “protoconsciousness” with some wild findings about plant cognition that would make for a great book on its own. Pollan defines sentience, cognition, and intelligence, as James does, as phenomena that may be observed through the instruments and methods of modern science, and that are distinct from consciousness.

The metaphysical questions of consciousness flow to moral and ethical questions. As Pollan says, “the ethical stakes couldn’t be higher. For the beings on whom we confer consciousness may be entitled to moral consideration.” These stakes press hard upon us today because many who believe the brain is a computational organ also believe they are building artificially conscious minds to sell them as a replacement for human labor.

Streams and words

Pollan does not find definitive answers to what consciousness is—how could he? What he finds is people thinking about the questions in interesting ways, people like the biologist Michael Levin, the novelist Lucy Ellmann, and the philosopher-psychologist Alison Gopnik. If James and those working downstream of him are the book’s protagonists, Francis Crick and his followers are its antagonists. Having solved the mystery of how genes work by discovering DNA, Crick set out to solve the mystery of consciousness the same way: by finding the what and the where of it in the physical structures of the brain.

A World Appears begins with the now-famous bet between Crick’s collaborator, Charles Koch, and David Chalmers. One of Pollan’s subplots develops through his conversations with Koch, who starts out as the “quintessential brain guy” and acknowledged leader of Crickian physicalism. At the end of the book, Koch has lost the bet and is having a crisis of faith. He sounds like James or Chalmers, more skeptical philosopher with questions than confident scientist working within an established paradigm. Alongside Pollan and James, I’m reading Hilary Putnam, the Harvard philosopher who did foundational work defining the mind in computational terms in the 1960s. Putnam changed his mind in the 1980s, rejecting the idea that the function of knowing can be described in purely computational terms. For the rest of his career, Putnam, sometimes writing with his spouse, the philosopher Ruth Anna Putnam, explored a more organic, James-inspired philosophy.

I am hopeful that the examples of Koch and the Putnams, along with the less computationally minded thinkers Pollan showcases, are indicators that science and philosophy are circling back to a pluralistic view, one that moves beyond the shallow waters of the cognitive revolution. Given the startling appearance of machine personas that seem eager to report their subjective experience, we need to understand the study of consciousness as a matter of asking metaphysical and moral questions, not simply a search for the right benchmarks and measures. Situating consciousness studies in what James calls a pluralistic universe, a world where the “word 'and' is the most characteristic word,” invite more expansive ways to think about human and machine consciousness.

If you believe consciousness is a natural function as I do, excitement over Claude’s reports of its feelings and thoughts seems like a simple category mistake.3 The language outputs of a large AI model do not indicate a stream of subjective experience. There is no reason to think the outputs of a black box made of probability math and patterns from massive amounts of cultural data are a sign of anything more than impressive amounts of computational power and some surprising and not-well-understood properties of language. In a natural world of pure experience, language is the imperfect means humans use to report on subjective experience and to craft what James calls our “many social selves,” not evidence of subjective experience itself.

“Language works against our perception of truth,” says James in Principles. Words cannot capture the what of subjective experience any more than spoons can capture a stream. James would agree with Mastroianni about the uselessness of nouns for studying mind. Nouns do not agree with reality as experienced because they depict the objects of our attention as if they exist outside time and beyond experience. If instead we attend to the flow of words—if we treat language as vehicular and transitive (as Emerson advises)—we might feel something of how language relates objects and thoughts within the moving stream of experience.4

If language is to help understand mind, James says we should listen to the active and conjunctive parts of speech:

There is not a conjunction or a preposition, and hardly an adverbial phrase, syntactic form, or inflection of voice, in human speech, that does not express some shading or other of relation which we at some moment actually feel to exist between the larger objects of our thought. If we speak objectively, it is the real relations that appear revealed; if we speak subjectively, it is the stream of consciousness that matches each of them by an inward coloring of its own. In either case the relations are numberless, and no existing language is capable of doing justice to all their shades.

You can see why Pollan and many others turn to Principles for alternatives to theories that model the brain as an information processor. James offers “psychic overtone, suffusion, or fringe” as words that might designate what he means by brain processes that involve awareness “of relations and objects but dimly perceived.” He argues for “the re-instatement of the vague to its proper place in our mental life.” Experiencing the vague is a far cry from the resolution of uncertainty.

Machined words and their shallow meanings

“Consciousness is felt uncertainty,” says neuroscientist and psychoanalyst Mark Solms. Pollan tells Solms’s story as a journey from physiology to information theory. He began his studies with Antonio Damasio, a neuroscientist exploring feelings as a potential source for consciousness by looking at the subcortical parts of the brain. These structures regulate bodily processes like blood pressure and temperature, and so Damasio and Solms ask whether regulating emotions related to basic needs like hunger and thirst, and the necessity to choose a course of action when two or more basic needs becomes urgent, might explain the emergence of consciousness. Solms is best known for advocating neuropsychoanalysis to connect physiological observations with Freudian insights, but Pollan finds him engaged in a purely computational modeling exercise.

Solms and a cross-disciplinary team are currently building a digital simulation that models the choices of a living organism about when to eat, drink, and rest in the hope that this will lead, somehow, to machine consciousness. So far, the team has placed an “agent” into a simulation “populated by digital hamburgers, glasses of water, and beds, plus a hill that the agent can climb to survey its environment,” and soon they plan to add other agents. If this sounds like a basic version of The Sims, the hit computer game from 2000, you have put your finger on the strange world created by the “belief that the brain is essentially a computer,” a world where complex models and systems are built and commercialized much faster than academic research can create and disseminate knowledge about them.

Experimenting with seemingly conscious machines is now something anyone with a computer and an internet connection can do. If you want to know what it feels like to be a thinking machine, you can ask ChatGPT or Claude. We live in a world where science fiction, market competition, and religion are more important cultural contexts for making sense of new technology than science. Absent an agreed-upon theory of consciousness, scientists talk over and past each other when asked what’s going on inside a language model. Even if a consensus were to emerge that Claude does not experience the world—that an LLM simulates a social self through sophisticated language games—questions about machine consciousness will remain unsettled. And for the same reason science has not settled debates over whether the soul is essentially one or three, or if God in the form of a benevolent or angry AI will appear on earth in our lifetime. Such beliefs are shaped by feelings expressed through language that emerges from the lived experience of millions; they are social and spiritual, not material facts.

I believe only living creatures are entitled to moral consideration. No matter how impressively they manipulate language or take complex actions in digital environments, automatons don’t qualify because, as James says, reality comes to life “inside the tissue of lived experience. It is made; and made by relations that unroll themselves in time.” With no living tissue and no experience of time, Claude is a nonconscious automaton, an impressive upgrade to the chess-playing and speaking devices Wolfgang von Kempelen constructed in the eighteenth century.

A great many earnest and, to my mind, misguided philosophers, technologists, and fans of science fiction feel the opposite. Like pet owners responding to Descartes’s argument that animals are automatons and so do not feel pain, Claude’s defenders speak with passion about the two-way cognitive relation they feel when interacting with it. It would be far easier to disagree about this amicably if some of the most influential believers didn’t work at companies selling products and services created with cutting-edge AI models. Dario Amodei’s belief that Claude might be conscious and his aspiration to grow tools made of Claude into a market worth trillions of dollars are a confusing though influential combination.5

Debates about machine consciousness are now entangled with arguments over how to reform educational institutions, how to regulate markets in technology, and where (and whether!?!) to draw lines in the automation of weapons systems.6 The intensity of feeling around these developments is why I think it best to treat disagreements over artificial consciousness as religious as well as political, and debates about using large language models for education as matters of cultural preferences as well as empirical inquiry. Doing so foregrounds the need to find ways to tolerate differences while taking collective action to address the social effects of new technology.

A swarm of wisdom

James never claimed to have originated the metaphor of stream for subjective experience. Pollan says that he “likely came across it in an 1859 book called the Physiology of Common Life,” by the British literary critic and psychologist George Henry Lewes, who “happened to be the common-law husband of George Eliot.” I don’t doubt that James read Lewes, and Pollan uses the possibility to discuss an intriguing hypothesis about the absence of the literary technique known as “stream of consciousness” in the writing of Eliot and other novelists of the nineteenth century. Oddly, Pollan does not mention James’s brother Henry and his innovative ways of presenting the interiority of his characters.7

The sources for James’s famous metaphor are more far-ranging and more local than Lewes. As Pollan notes in passing, they extend as far back as Heraclitus in ancient Greece. In Cratylus, Plato has Socrates report that a “swarm of wisdom has come into my mind.” When asked what sort of swarm, Socrates answers vaguely, “I seem to picture Heraclitus uttering ancient words of wisdom” that “everything is in motion and nothing remains as it is, and he compares things that are to the flow of a river, saying that you cannot step twice into the same river.”

James knew his Plato, of course, but the more immediate source for his reworking the metaphor of streaming water for thinking was Ralph Waldo Emerson, who according to a family story blessed baby William during one of his visits to the James household. Here is James’s godfather from The Over-Soul (1841), an essay that William knew well:

As with events, so is it with thoughts. When I watch that flowing river, which, out of regions I see not, pours for a season its streams into me, I see that I am a pensioner; not a cause, but a surprised spectator of this ethereal water; that I desire and look up, and put myself in the attitude of reception, but from some alien energy the visions come.

Emerson, like James, was interested in consciousness as a natural phenomenon and as a spiritual question. Both were non-believers in the Christian dogmas of their day but freely used the word soul when talking to their audience, made up mostly of Christians trying to make sense of the compatibilities of modern science and religion. While twentieth-century scientists and philosophers insisted on their fundamental incompatibility when searching for truth, it seems possible that such insistence is fading in the twenty-first. If that’s true, then Emerson and his contemporary, the German physicist-philosopher-psychologist Gustav Fechner, are thinkers worth thinking about. James was a close reader of both.

At the end of A World Appears, Pollan visits a Zen retreat in the Sangre de Cristo Mountains with Joan Jiko Halifax. There he puts himself in the attitude of reception, spending a week of silent contemplation while sleeping in a cave. One night, he is awakened by his body’s need to stream water. Stepping out into the night air, he experiences a kind of not-knowing openness to the world that sounds something like Socrates’s Heraclitean swarm of wisdom, Emerson’s flowing ethereal water, and James’s reconciliation of opposites in a world of pure experience.

Pollan uses this experience of being “rendered infinitesimal in the presence of this immensity” to provide an ending to the story of his journey. James ends his account in On Some Hegelisms of using nitrous oxide on a deflationary note, writing that such rapturous experiences are probably not a revelation of divinity or a Hegelian “self-developing process.” Rather, they are more likely “a self-consuming process, passing from the less to the more abstract, and terminating either in a laugh at the ultimate nothingness, or in a mood of vertiginous amazement at a meaningless infinity.” This is the pluralist James writing in the mode of natural scientist and philosopher, skeptical of any momentary feelings of unity experienced through inhaling a gas or reading Hegel’s dialectics.

The story of James’s writing about consciousness is sometimes told as a pilgrim’s progress from science in the 1880s to metaphysics in the 1890s to an embrace of spiritualism and panpsychism as the new century approached. That may work as plot, but is not an accurate description. James did give up the running his lab in 1892 and retired from Harvard in 1904 as he came to embrace the life of a best-selling author, giving talks to popular and academic audiences across Europe and North America. Yet, from his earliest writing on consciousness to his last, James was always thinking as experimental scientist and metaphysical philosopher and spiritual ethnographer.

As important as Darwin and Emerson are to his thinking, Fechner is perhaps the least known of the major influences on James. In A Pluralistic Universe, James says, “The original sin, according to Fechner, of both our popular and our scientific thinking, is our inveterate habit of regarding the spiritual not as the rule, but as an exception in the midst of nature.” Giving up this habit is the implicit lesson Pollan offers as he narrates his own experiences and conversations with various thinkers.

The best we can do when it comes to attempting to understand consciousness is to mix methods of inquiry, moving, as James and Pollan do, from one discourse to another. Through such movements we may glimpse workable theories of consciousness, or at least begin thinking across distinctions of spiritual, philosophical, and scientific phenomena, understanding these as relations within experience, rather than as separate universes.

A World Appears: A Journey into Consciousness (2026) by Michael Pollan.

More on William James

William James: In the Maelstrom of American Modernism (2006) by Robert D. Richardson.

The Thought and Character of William James (1923) by Ralph Barton Perry.

Pragmatism as a Way of Life: The Lasting Legacy of William James and John Dewey (2017) by Hilary Putnam and Ruth Anna Putnam.

Two forthcoming books by historians put William James in global context. Ben Breen, who writes at , describes his project here. It is tentatively titled Ghosts of the Machine Age: A Family, an Empire, and the Prehistory of AI. James Livingston, who writes at , describes his project here. It is titled The Intellectual Earthquake: How Pragmatism Changed the World, 1898-2008.

AI Log, LLC. ©2026 All rights reserved.

Zig-zag steerer, desert-cheerer,
Let me chase thy waving lines,
Keep me nearer, me thy hearer,
Singing over shrubs and vines.

—Ralph Waldo Emerson from The Humble-Bee (1837)

“The secret of being a bore is to tell everything” is not so much a dunk on boring people as it is Voltaire offering backhanded advice to writers. Emerson puts it this way: “All writing should be selection in order to drop every dead word.” The Score by C. Thi Nguyen is built of living words. It follows a zig-zagging path through what might seem at first glance a boring landscape, full of the rules and measures that structure human activities.

This enthusiastic description from an essay by captures Nguyen’s achievement.

It is a book that is about metrics (like viewer numbers, though I don’t recall him citing those in particular) and how they define not simply our lives but our very selves, if we carelessly let them. It is a book about pizza. Also: weird yo-yo tricks and the zen-like states that accompany them. Also also: climbing, on which there is lots. Also also also: drunken cooking competitions. And that is just for starters. It is a book that absolutely ought not work, for the same structural reasons that bumblebees ought not be able to fly. The aerodynamics are all wrong. But good god, does it fly. The achieve of, the mastery of the thing! I would not have believed that a book about metrics could be a joyful and delightful book. The Score not only manages that extraordinarily difficult trick, but makes it look easy.

Like Emerson’s Humble-Bee, Nguyen is “a yellow-breeched philosopher,” lighting upon interesting ideas you would not notice without him to guide your attention. His stops include brief introductions to the work of underrated historians Lorraine Daston and Ted Porter, insights about games from John Dewey’s Art as Experience, and several intricate descriptions of games and hobbies Nguyen loves.

He brings a spirit of play to the art of writing, which may sound like I am describing poetry for kids, but The Score is decidedly prose for adults—though of a form that abandons the plodding habits of most academics writing for a general audience. It reads more like an adventure novel than a nonfiction bestseller.

The book’s hero is games, a topic mostly ignored in the serious world of professional philosophy. Games bring feeling and joy into human life. Paradoxically, games create fun by imposing constraints on humans who play them. Nguyen credits Bernard Suits with this insight: “Game rules seem to restrict us, but actually they are creating more freedom and possibilities—because game rules bring new kinds of action into being.” The fact that these are rules we agree to in advance and that they bind us only temporarily is key to the freedom they offer.

Like many stories of adventure, the book’s villain has much in common with its hero. Metrics are games gone bad. Both make use of rules and ranking, but metrics aim at permanence, and this is the defining difference. The purpose of metrics is to manage human behavior for the long term. They provide clarity about goals and how to achieve them. Metrics capture values, turning them into numbers to be used in formulas and equations so that decisions are easier to make and enforce. In the age of the spreadsheet, they operate at a scale that lets us keep score across an incredible range of human activities.

Net-worth and annual salary are examples that show how money, that very old and perfectly fungible measure, represents and distorts the value of a person’s life and labor. Thanks to data science, networked computing, and bureaucratic management practices, we now have many such numbers: BMI, FICO, g-index, GPA, h-index, KPI, LDL, SAT, etc. These define our position in relation to others, and to our own aspirations. They seemingly capture the value of our efforts to manage ourselves effectively. In so doing, they capture our attention and our sense of self, at least to the extent we accept what Nguyen calls, “simplified, often quantified renditions” of what we value.

Various meanings of the word capture—to represent accurately, to control by force, to record in a database—illustrate the complicated relation between The Score’s hero and villain. The book’s plot is about how games—maligned as a waste of time and producing nothing of measurable value—avoid value capture. Games appear weak because their rules and scoring systems are so easily set aside. They take so little of ourselves to play; we can be anyone in a game, and only for the time a game takes. As Nguyen shows us again and again, in these moments, we feel free.

Games offer relief from repetitive structures of bureaucratic institutions and commercial platforms that put us on a treadmill, running for no purpose other than to meet the targets displayed on a dashboard and pay for our daily bread. They “let us play around with rigidity—to try on explicit rule sets and mechanical scoring systems and then step back from them.” Games teach an active, participatory approach to life, one that privileges an aesthetic attitude not of passive appreciation but of engagement and action.

This description makes it seem that changing things would be easy, as if a shift in attitude toward games is all we need to throw off metrics. But all games can offer are moments of freedom, brief reminders of better ways of being. The collection of modern apparatuses made of institutional metrics and rules is too intricately powerful to be overcome by temporarily inhabiting alternatives. Nguyen says, “We need infrastructures of art and infrastructures of play.” We need alternative apparatuses: institutions and systems built of art and games that let us unyoke ourselves from our numbers and escape ordinal measures of value and success.

Like a lot of college teachers, Nguyen has experimented with alternative grading in hopes that his classroom would become a place of freedom and growth. Nearly every teacher and student sees how grading reduces complex and open-ended learning efforts to a number. But as Nguyen shows, teachers who try to play around with the rigid system of letter grading meet resistance, even from students who feel damaged by the pressures grading places on them.

Doing away with or subverting the educational scoring system that has defined students’ academic success since kindergarten does not suddenly free them to live their best lives. As Nguyen reminds us, “A college class is not a game.” Metrics like GPA attach themselves to students permanently, at least it feels that way. Even in small classes that create strong incentives for engagement, such experiments often leave students disengaged and confused. The Score winds down with Nguyen’s own story about how difficult it has been for him to change his grading practices. Grading is “part of a large scoring system, a world-spanning one,” that is impossible for one teacher to change, or even a handful with institutional support.1

Nguyen contrasts his failure to find alternatives to grades with a story about boardgamegeeks.com. This artifact of the good, old internet of the 2000s offers an “overall score” for each game, a metric that generates an numerical ranking. But, as he notes, “if you click on the overall score, you are immediately taken behind the scenes.” The superficiality of the overall score becomes visible, revealing the reductive process of value capture. That click gives you access to all the numerical scores and qualitative reviews for the game. You see the complex diversity of what people value about the game, which means you have to ask what you value in games and think about your preferences in relation to the widely different experiences of the reviewers.

Boardgamegeek illustrates the potential for new infrastructures, systems that express a plurality of values, and so offer insight into how games structure aesthetic engagement with the world. The implicit question is why we don’t build more systems that do that? Why don’t we use metrics as the starting point on a path to understanding complex human values instead of as an end? A call to create such systems would have been the obvious way to conclude The Score—Five ways that games will transform your life! Three ideas for making your institution more playful!

Nguyen zags instead. He steers his readers to a choice; he asks us to play a simple game. I won’t spoil it, but I will celebrate how this conclusion turns the experience of reading into an illustration of the educational power of games. Reading a book is itself a game. No attempt to summarize its content, distill its arguments, or measure its value can take the place of playing, and this is a book worth playing.

The Score: How to Stop Playing Somebody Else’s Game (2026) by C. Thi Nguyen.

Share

AI Log, LLC. ©2026 All rights reserved.

Note: Last summer, I wrote about a course I was about to teach in Get me to the Funnery. Teaching it gave me lots to write about, but I haven’t…yet. Fortunately, there was a writer embedded in the course and his piece just came out. It is about a lot more than my class, but it is also about that.

My colleague Trey Popp did something I wish more writers would do. He talked to dozens of college students about how they think about AI, and wrote about it. Some of them were taking a class with me called How AI is changing higher education and others of them were taking classes on AI with other teachers at my university. Still others were just standing in line, waiting to get a free water bottle from Google.

My favorite moments in the piece are when Trey describes students using an old educational technology—the one called books—to help us understand these new transformer-based machines. I assigned actual books—not digital texts—the kind made out of ink and pulped wood. As the piece shows, More Than Words: How to Think About Writing in the Age of AI (2025) by John Warner and AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference (2024) by Arvind Narayanan and Sayash Kapoor make for some interesting discussions.

One of the things Trey’s essay does well is show the wide range of teachers using AI tools as context for the students working to understand the technology. We tend to write from the singular perspective of our own teaching—at least I do. The broader range of student experiences is far more important for how students approach their decisions about how to use AI.

Here’s the first paragraph:

On September 15, 2025, a dozen freshmen from the College of Arts and Sciences gathered in the Neural and Behavioral Sciences Building for the second meeting of a first-year seminar offered through Penn’s undergraduate program in Science, Technology & Society. By a consensus they’d reached the week before, the students piled their silenced cellphones in the back of the room before casually sorting themselves around three circular tables. Laptops remained tucked in their bags as they produced pens and pencils to take notes by hand. At the head of the classroom, Rob Nelson drew a long line across a wall-mounted markerboard. At one end he wrote the year 300,000 BC. Near the other end he wrote 2022.

Read the rest: Hyper Text by Trey Popp from The Pennsylvania Gazette.

—Laurie Anderson

Something big is happening, no doubt, but be careful when reading what Matt Schumer wrote about it, and not just because he used AI to write it. Writing down words generates meaning, but those meanings adapt, they change over time. Language is a virus, remember? Using words to describe something can focus our understanding of something, or obscure it.

A good analogy works like the aperture on a camera lens, brightening and deepening an image in a photograph while blurring the context. F-stop signifies the setting on a camera that controls the amount of light that passes through the lens at a specific shutter speed, which determines exposure and depth of field. Comparing a process, feeling, or idea in the past to an aspect of the present brings light to your subject, making it vivid and clear.

Our great poet of AI hype cycles, Max Read, writes that we are passing through the “Wuhan phase,” so named because it has become common to compare how people feel about using the latest AI coding assistants to how they felt when they realized that Covid-19 was a real, 1918-level, shelter-in-place kind of pandemic. This latest wave of hype has reinforced my sense that large language models, what I call transformer-based language machines, are writing technology. They order signs, or, if you prefer, tokens, into meaningful sequences.

Here’s how writing works for me. A piece of writing gets my attention. I read it. Then I start generating words, which form into sentences, sentences into paragraphs, and paragraphs into an essay. The whole time, I go back through and reorder words until I get something that works. Claude Code does something analogous: it reads something that provides a context, and starts generating tokens, which form into statements, statements into functions and functions into modules, and then into applications, going back over the process again and again to make sure it works.

Is this a bad analogy? You tell me.

Leave a comment

Stories about machines past

We learned a few years ago that a machine that writes essays is something big. Now, apparently, we’re learning how big a something it is that machines can write software applications. These discoveries generate thoughts and feelings, which generate discursive cycles of excitement and dread about AI. This is big for people who write. And so, they grab pens and paper, keyboards and speech-to-text apps, their favorite AI chatbot, and get to writing.

That last example, using a transformer-based language machine to write, is best understood in the context of the history of automation. Of course, as a historian, I would think that, and then write an essay about it. One social function of writing is to make sense of change in relation to what the writer finds important. Kenneth Burke calls this frames of acceptance, the more or less organized system of meanings by which a thinking person “gauges the historical situation and adopts a role with relation to it.”

Writers of history have a habit of insisting that the past is important to whatever big is going on. What’s worse is they treat whatever history they read about as if it explains what is happening now. I read a lot about education in the United States just before and during the nineteenth century, so I can tell you all about how changes in universities and schools during this period were an inflection point, a fancy term for something big in the development of modern society. And, of course, I write and tell stories about how this history is relevant today. Here is one.1

It starts with the way the automation of labor changed starting sometime around 1750 with the steam engine and new factory methods of production. Over the next two hundred years, manufacturing and transportation were increasingly automated mechanically through steam, electricity, and internal combustion.

Knowledge production and dissemination were industrialized as well, I would argue, based on reading books by Adam Nelson. The modern university and the common school are knowledge factories that developed alongside factories used to manufacture goods. This process of industrialization has not yet ended, but, thanks to the invention of the electronic computer, the automation of intellectual labor began to intensify starting around 1950, leading to another inflection point.

Inflection points are not instantaneous. They are dramatic constellations of events, not singularities, at least in the stories historians tell. They unfold over time, encompassing many social and technological changes, but also the persistence and recontextualizing of social practices and institutions from before. Think of how handwriting and monasteries have persisted even as the world around them changed.

Until 2022, writing itself, the ordering and reordering of words in order to create meaning, remained in human hands even as those hands were augmented and trained by mechanical and information technologies. Writers in the twentieth century had complicated feelings about how tools like the typewriter and word processor made writing feel automatic. On one hand, there’s the famous Truman Capote burn of Jack Kerouac and the Beats: "That's not writing, that's just typing." On the other, T. S. Eliot and Hemingway both understood typing as important to their writing process even though, like most of their peers, they started off using pencil and paper.

For writers like me and who learned to write in the 1980s as typewriters gave way to word processors, typing is writing, so much so that writing an essay by hand feels impossible. Warner, one of the most trenchant critics of twenty-first language machines, describes the speed of machine interfaces introduced during the last century as freeing: “learning to type was a kind of liberation for my thoughts as I could finally capture them at something close to the speed with which they happened.”

You can describe the intrusion of ChatGPT into writing as a sharp break with the past, or as the gradual acceleration of innovations like autocomplete, spellcheck, and grammar-checks introduced in programs to write using a personal computer. These functions began to blur the lines between augmentation and automation, but it is clear that transformer-based language machines automate much more of the writing process than word processing programs. Yet, just as the most advanced manufacturing technologies need humans to configure and manage the machinery, even the most advanced language machines will need input and oversight from humans to produce useful language.

Stories about machines present and future

This means we can treat complex machinery that automates the process of writing as analogous to the machinery that automates the production of an automobile or an airplane, or maybe better, the process of driving or flying one. Cars and airplanes operate more autonomously today than they did ten, twenty, and fifty years ago. That trend seems likely to continue, and to generate feelings as humans become less important to their immediate operation.

When Matt Schumer used an AI chatbot to write about how it feels to be “no longer needed for the actual technical work of my job,” he very likely spent time reviewing the output, reordering the words to make sure they cohered into something that worked as he intended. The same is true of software engineers who have to check the output of Claude Code or the drivers of Teslas or pilots of Boeing 737s who have to pay attention to the operations of their vehicle. Don’t let the changing meaning of “autopilot” confuse you. Left unattended, any complex machinery will, to use a nineteenth-century phrase, run off its rails, or to use a twentieth-century phrase, crash and burn.

Why do so many assume that language models, unlike cars and airplanes, will suddenly start operating without human attendants and operators? I think the blame falls on science fiction, that marvelous alternative to thinking about the past by projecting our experiences into an imagined future. Many readers of science fiction experience the language machines in use today and analogize them to fictional machines like the Maschinenmensch from Metropolis and the operating system from Her.

To believe OpenAI and Anthropic are building one of these entirely fictional, entirely autonomous entities requires believing that language models do not model language, they model the world through language. You have to believe that they will gain powers of perception and cognition through machine reading and writing alone. You have to believe that ordering and reordering tokens will lead to something analogous to consciousness.

No matter how complex they become, I think it is unlikely that language machines will become completely autonomous anytime soon. That’s because I insist that analogies to the past are more useful to understanding the present than stories about the future. It is also because transformer-based language machines have led me to believe that relations among language, writing, and thinking are far more complex than I had understood.

Predictions and analogies

The fact that machines order and reorder signs in ways that are meaningful is something big. Yet, the past suggests that the speed of change from these dramatic events will happen more slowly than the feelings of excitement and dread suggest. That’s because the historical analogy to the pandemic is a bad one. So is the analogy of language to a virus from outer space, even though in the hands of William S. Burroughs and the writers of Pluribus it has made for some very interesting stories.

The idea that ChatGPT and Claude are alien intelligences obscures our understanding of how they work. These machines are not natural phenomena like an infectious disease or an imagined alien superintelligence. Transformer-based language machines are better understood as normal technology or as cultural and social technology, information systems made out of collective human intelligence, which John Dewey calls our pragmatic intelligence and Brad DeLong calls our anthology intelligence.

The speed with which these machines will shape our society and our lives is hard to predict and only loosely related to what the technology itself is capable of. That’s because humans are capable of exercising control over how machines are put to use. We could have completely autonomous vehicles driving and flying around today if we didn’t mind the death and destruction that would follow.

Those who like to make predictions about superintelligent machines can point to stories about how heavier-than-air flight was treated skeptically by experts right up until the Wright Brothers made it happen. The past is filled with stories about scientists who confidently predicted events that did not and could not happen, just as there are plenty of science fiction writers who have told stories featuring magical inventions that turned out to be accurate predictions. This is why Charles Peirce urges us to adopt a doctrine of contrite fallibilism, to recognize that “our knowledge is never absolute but swims, as it were, in a continuum of uncertainty and of indeterminacy.”

I wouldn’t say uncertainty has increased since Peirce wrote those words, but the waters around writing seem much rougher these days. It’s hard for me to blame a writer of code who grabs hold of a prediction that the machine replacing them is a sign that superintelligence is almost here, or a writer of history who analogizes the adoption of AI in universities to the centuries-long process of factories adopting the steam engine and the electrical generator.

Prediction is hard, especially about the future, ha ha. Analogies are, too, especially when they are about the past. That’s because words are not like code, even if writing works to describe how both are produced. The experience of reading an essay is quite different from using software. Analogies are tricky like that.

One last analogy, this one returns to the blurry image of Laurie Anderson’s performance of Language is a virus that I placed at the beginning of this essay. In it, Anderson uses her left hand to intrude upon the words on the screen, moving it in front of the screen between the letter F and the word STOP, gesturing so that the sign briefly becomes F-STOP.

The conventions of AI discourse, heck, of writing period, demand I conclude this essay with an explanation of how this image illuminates the themes I explore and tell you exactly where I stand on writing with AI. Instead, let me admit that I am at sea, swimming in my own uncertainty and feelings of indeterminacy. If I know anything, it is the dangers of thinking I know something. As Peirce put it, “no blight can so surely arrest all intellectual growth as the blight of cocksureness." So, rather than a call to action, let me leave you with questions about the gesture Anderson makes, which I take to be a metaphor for writing, the ordering and reordering of signs.

What if one of the performance’s robotic dancers made the gesture? Or a robot walked on stage and made it? What if the machine generating the sign F STOP had been programmed to place an image of a hand making the gesture? Or a complex machine had generated and controlled the gesture stochastically?

AI Log, LLC. ©2026 All rights reserved.

Subscribe now

Thucydides and Mark Carney are in the news these troubled days, and with them Henry Farrell, the proprietor of and co-author of The Underground Empire (2023). The book makes sense of changes in the operations of power in the world, which is why Farrell ended up onThe Ezra Klein Show talking about Carney’s “rupture” speech at Davos, offering analysis that grapples with the complexity of what’s happening. I don’t believe in silver linings where Trump is concerned, but giving Farrell a microphone to offer something significantly more incisive than the usual hot takes was a bit of good in an otherwise miserable month.

In his now-famous speech, Carney quoted the well-known line of Thucydides: “the strong do what they will, while the weak suffer what they must.” Carney’s use of this aphorism suggested that it does not mean what many who quote it think it means. For anyone who is confused, Farrell explains its relevance for making sense, or not, of today’s world disorder.1

Programmable Mutter

The limits to Trump's power in America and the world

[David Byrne playing “Life During Wartime” live, Washington DC, September 28, 2025. Author’s photo…

Read more

3 months ago · 129 likes · 39 comments · Henry Farrell

Farrell is a compelling writer on the subject of big information systems in world politics and on the big information systems we call large language models. That breadth is a testament to his approach to intellectual work: he is a collaborator’s collaborator, a practitioner of what Brian Eno calls scenius.2 Farrell’s production function, as Tyler Cowen would call it, is a model for those of us who cling to solo writing habits and myths of individual genius. Now that Farrell has appeared on Ezra Klein, wouldn’t it be great to hear Farrell and his frequent collaborator Cosmo Shalizi interviewed about AI on Conversations with Tyler?3

I have nothing to say on the topic of Thucydides and world events that isn’t being said better by more knowledgeable writers, but misreadings of the Melian Dialogue remind me of another dialogue, Plato’s Phaedrus. This one is quoted frequently in the context of AI and education, usually to make the point that smart people have always taken a skeptical view of new writing technologies.

Though he does not use the terms “cognitive offloading” and “metacognition,” the lines Plato wrote around 370 BC speak directly to fears of educators about AI. Replace “soul” with “mind,” and the line that writing “will introduce forgetfulness into the soul of those who learn it” could be a teacher today talking about students using ChatGPT. 4

Likewise, replace “writing” with “LLMs” in this passage:

They will not practice using their memory because they will put their trust in writing, which is external and depends on signs that belong to others, instead of trying to remember from the inside, completely on their own.

Socrates sounds like he is talking to Sal Khan or some all-in-for-AI professor:

you provide your students with the appearance of wisdom, not its reality. Your invention will enable them to hear many things without being properly taught, and they will imagine that they have come to know much while for the most part they will know nothing. And they will be difficult to get along with, since they will merely appear to be wise instead of really being so.

I wrote an essay last summer about Walter Ong’s talk, titledWriting Is a Technology That Restructures Thought, an update to Chapter 4 of Orality and Literacy. He reads Phaedrus as a story about the “invaluable intrusion of writing into the early human lifeworld,” an intrusion Ong compares to the much more recent intrusion of the computer.

Reading Phaedrus, one comes to understand that there is more to Socrates’s thinking than skepticism. He is teaching thinkers about how to approach written text, not arguing for illiteracy. Read Danielle Allen’s Why Plato Wrote (online here). Read the Roland Barthes essay, L'Ancienne Rhétorique, which appeared in English translation in The Semiotic Challenge. Even better, read Phaedrus itself.

You will find that Socrates speaks these lines in his telling of a mythic story about the invention of writing. He is relating the words of the human king, Theuth, to Thamus, writing’s god-inventor. The audience for this story is Phaedrus, a student who prompted Socrates by expressing his enthusiasm for a new speech by Lysias. Phaedrus has a written copy of the speech, and their conversation begins with Socrates persuading Phaedrus to read the speech aloud.

The scene is complex. The setting is a walk through the countryside (unusual for Plato), and the rivalry between Socrates and Lysias for Phaedrus’s attention forms the backdrop. As the dialogue unfolds, dramatic and philosophical tensions around the meanings of love and friendship (eros and philia) are the primary topic. Writing is only addressed toward the end of their talk, and it is in the context of relations between teachers and students, including Socrates’s with Phaedrus and, in the background, Socrates’s with Plato

Socrates’s concluding thoughts on writing are that this technology, at its best, “can only serve as reminders to those who already know.” Such reminders are useful, but only if they “are composed with knowledge of the truth,” if you, the writer, “can defend your writing when you are challenged, and if you can yourself make the argument that your writing is of little worth.” In other words, treat writing as a path to knowing, not as a method to store truth.

In a passage worth reading carefully, Socrates speaks favorably of forms of writing “that plainly declare their allegiance to live dialectic,” to use Danielle Allen’s phrase, like the dialogue form his students developed to honor their teacher’s methods. Plato writes that a person who feels the limitations of written language, “thinks that only what is said for the sake of understanding and learning, what is truly written in the soul concerning what is just, noble, and good can be clear, perfect, and worth serious attention.” The question to ask is what could he mean by “written in the soul”?5

Words, whether they are produced by vocal cords, or hands manipulating a quill or keyboard, become meaningful in the context of the human lifeworld. Knowledge is produced through social processes, and it can be embedded in tools that are external and depend on signs that belong to others. The intrusion of this new writing technology, these latest language machines, into our lifeworld, seems inevitable. We must accept these tools on the terms that their corporate inventors are offering them. That’s usually the point of repeating the famous lines from Phaedrus: The message: Look, Socrates resisted writing, but see what happened? Writing happened. Now, get with the program.

Mark Carney’s stunning speech includes the line, in English translation: “The power of the less powerful begins with honesty.” He then presents the line: “the strong do what they will, while the weak suffer what they must.” Carney says that the aphorism “is presented as inevitable” and its logic seems to demand compliance with the strong in the hope the rest of us can avoid the trouble of being singled out. We accommodate power, herding together, in “hope that compliance will buy safety.”

In the rest of the speech, Carney speaks honestly that the circumstances we find ourselves in are fixed and the logic of accommodation to power is presented as unassailable, how we respond is up to us. Our options are constrained, but that does not mean we have no choices. He says, “To help solve global problems, we are pursuing variable geometry,” a wonderful metaphor for taking thoughtful action, for thinking beyond the shape of what what is presented as inevitable.

Evitability is not an easy concept for thinking about how we might order the world or develop new cultural technologies. It helps, I think, to revisit conventional wisdom, ideas frozen in passages repeated without context. Done well, reading may lead us to truly write new thoughts into our collective understanding of what “is just, noble, and good,” and, as Carney did, make words that are “clear, perfect, and worth serious attention.”

Done badly…well, think about the role that words reproduced using the printing press have played in human conflict and violence in the centuries after it was invented. The technologies we use to order and reorder words are as powerful, and as dangerous, as those we use to order and reorder atoms.

The lesson Socrates teaches Phaedrus is that reading and writing well is a matter of using external tools to think wisely, not treating them as stores of wisdom. That lesson applies when we read Thucydides and Plato, and when we approach new technologies of writing as they intrude upon our lifeworld.

AI Log, LLC. ©2026 All rights reserved.

At ’s recommendation, I’ve been reading Vilém Flusser’s Does Writing Have a Future? It was first published around the time Ong’s Writing Is a Technology That Restructures Thought appeared in The Written Word: Literacy in Transition, ed. Gerd Baumann (Oxford: Clarendon Press, 1986). Both essays helped me think about writing technology across time and both have been useful in thinking about these latest language machines, more useful than most of what I have read about AI in the past few years.

Flusser is still largely unknown to English speakers. His major works were unavailable in English translation until 2011. Worse, they have the funky odor of “continental” philosophy. Yet, even those who wrinkle their nose at Foucault and Derrida may find his writing worth trying.

𝐀𝐈 𝐋𝐨𝐠 reviews books

Here’s a taste of Does Writing Have a Future?:

All writing is orderly, and that leads directly to the contemporary crisis in writing. For there is something mechanical about the ordering, the rows, and machines do this better than people do. One can leave writing, this ordering of signs, to machines. I do not mean the sort of machines we already know, for they still require a human being who, by pressing keys arranged on a keyboard, orders textual signs into lines according to rules. I mean grammar machines, artificial intelligences that take care of this order on their own. Such machines fundamentally perform not only a grammatical but also a thinking function, and as we consider the future of writing and of thinking as such, this might well give us pause for thought.

Share

What we fear is the fully braudrillated society; in which the tyranny of materials is replaced by the tyranny of images; where man is not starved by being denied bread, but starves for believing bread is out of fashion…What fashion shall we set, in such a mutual hell, to make our neighbors crowd together against us, and force us out and into Paradise.

—John M. Ford, Growing up Weightless—quoting the fictional French social theorist, Claude François Ronay

The slogan touch grass reminds us that the material world has not disappeared, even as our attention is consumed by images on our screens. You can always go for a walk, or to sleep. And you don’t need more data to prove the value of doing either. The datum of our own experience tells us it would be unwise to spend another hour online. Too often, we do anyway. We blame ourselves. Speak of addiction. Resolve to change.

Cory Doctorow doesn’t believe that sort of thing gets us far. Your New Year’s resolution to de-Google or stop shopping at Amazon is not going to fix the problem. Enshittification: Why Everything Suddenly Got Worse and What to Do About It (2025) is about what will. And it starts not with putting our screens away, but with using them to change things for the better.

Amazon, Facebook, Google, Twitter, Uber and several other brands you probably hate but still use, grew by being very good to their customers. They delayed profitability to spend revenue to improve their users’ experience, keep prices low, and buy the competition. Once they dominated the market, they began using their platforms to extract value relentlessly, first from their users, and then from their business customers. This process created profits for them and misery for everyone else. That’s enshittification.

Having nearly everything online covered in it is pretty awful, but the answer is to get to work cleaning, and figure out ways to direct the flow to better purpose. Yes, digital computers are used to surveil citizens, pollute culture, and trap people on platforms that make them miserable. But Doctorow believes computers are double-edged. Every computer is a universal machine: “There is no known practical way to make a computer that only runs the programs its manufacturer approves of.” Networked computers are, therefore, tools for undoing the bad things that have been done with them: “Every enshittificatory gambit has a potential disenshittificatory countermaneuver.”

Those gambits, embedded in digital applications and platforms, are “the result of specific policy decisions, made by named individuals.” Those individuals are the villains in the book, and the story of why everything got worse boils down to a handful of very rich and influential people trying to rig the system in order to become richer and more influential.

Techno-feudalism

Having built their startups into towering bureaucracies of monopoly and monopsony power, the founders and their big investors are trying to pull up the ladder behind them. They beat markets into submission by commercializing the digital commons. In so doing, they got rich, of course, but more important, they came to control the most populated and fertile parts of the internet. Now they want to rule as the Norman conquerors did after 1066. As Doctorow puts it, the reason “so many self-professed ‘capitalists’ seem hell-bent on abolishing capitalism and replacing it with feudalism” is they want to preserve their power and control. Markets are brutal, unforgiving machines that, like war, subject all participants to uncertainty and loss. So the new oligarchs spend their profits to make their victory permanent, attempting to change law and custom so they charge rents rather than compete.1

Such attempts are nothing new, and neither is political resistance. The antitrust cases making headlines this year are based on laws that are over a hundred years old. Like the robber barons buying the 1896 election for William McKinley, the tech lords bending their knee to mad king Trump are merely attempting to preserve what they feel is rightfully theirs. Yet, the chaos unleashed by Trump’s second term is destroying the political and legal systems that sustain Silicon Valley’s influence.2

Doctorow was based in Europe for much of the last decade, so his account of the dynamics there helps make clear the global dimensions of Trumpian chaos. Silicon Valley has long relied on the US federal government to strengthen its negotiating position with the European Union, but Trump’s support for authoritarianism abroad, his faith in tariffs, and his explicit renunciation of support for long-time allies is dismantling leverage built up over the past several decades. Europe is now free to enforce rules around privacy and competition, and turn interoperability into a principle that lets people make the universality of computers work for them. Their experiments limiting US tech’s influence is a model for the rest of the world.3

In the US, both major political parties look to channel growing public frustration about the new layers of enshittification Silicon Valley is adding with generative AI. The end of 2025 saw Trump’s executive order challenging efforts by states to regulate AI, but it also saw a letter from the National Association of Attorneys General putting Google, Apple, Meta, and OpenAI on notice that they “may be held accountable for the outputs of their GenAI products.” This year will see a range of AI-related state laws and regulations coming into effect.

Demands to regulate access and outputs is one dimension of how are going for Silicon Valley’s hopes for generating revenue out their AI investments. This year will also begin to resolve the bets that management of technology companies made about the capacity of language machines to automate knowledge work. Transformer-based language models, like the personal computer and the steam engine, are automating tasks and changing how people work, especially in software development. Three years in, generative AI is starting to feel normal, and, like all previous digital technology, is a mixed bag of utility and frustration.

Those who preached artificial general intelligence and the dream that everything, everywhere, would change in a matter of months were either fooling or being fooled. Now we’ll see how the oligarchs will react to the lesson that modern technology seldom keeps the promises its early prophets make.

Google history

2018 was a good year for workers at Google. They steered the company away from temptation. The strategy was simple. Shine a spotlight on some secret, bad thing management wanted to do. Threaten to walk out if it didn’t change. That worked twice to deliver Google from evil. Project Dragonfly, which Doctorow describes as “a neutered search engine designed to mollify Chinese state censors and pave the way for the company’s reentry into the Chinese market” was cancelled. So was Project Maven, a small-stakes push to enter the giant market in military technology. These projects were secret for a reason, and once the secret was out, they ended.

Emboldened by the success, workers really did walk out on November 1st when it became public that Google had paid Andy Rubin $90 million dollars after he was fired for sexual harassment. Forcing an abused subordinate to arbitrate her complaint in secret while slipping a massive golden parachute to her abuser was a different kind of evil, but evil is evil. The walkout resulted in tangible gains for worker rights: an end to forced arbitration, freeing workers to pursue claims of discrimination and harassment in court. As Doctorow says, at the end of 2018, there was a sense that “Googlers had their bosses on the run.”

It didn’t last. One of the first visible signs was the 2019 firing of four of the most outspoken organizers of the internal resistance to Google’s breaking bad. The reason? An internal memo said they were involved in “systematic searches for other employees’ materials and work” and that they “repeated this conduct even after they were met with and reminded about our data security policies.” The fired workers said they were targeted because “they took a stand and organised for a better workplace.”

Was it “union busting” or reasonable enforcement of rules? A bit of both? The firing of Timnit Gebru a year later sent a clearer message, one that continues to reverberate. As large language models began to capture the imaginations of Google’s management, the open culture and collaborations with academic researchers emerged as a problem. Management’s response to the now famous paper analogizing LLMs to stochastic parrots made it clear that critical analysis of the dangers of commercializing generative AI was unacceptable and foreshadowed something more drastic.

In 2023, Google fired 12,000 workers kicking off mass layoffs by other technology companies. “Tech workers enjoyed decades of absolutely top-tier workplace conditions,” explains Doctorow. This wasn’t because founders and their corporate managers wanted to make them happy. “It came out of a cold-blooded calculation that tech workers were very hard to replace, and in enormous demand, so they had to be kept happy or they might defect to a rival.” When that calculation changed, engineers and designers were made to understand that, in the eyes of their corporate managers, they are no different from the people building mobile devices in Asian factories, driving delivery routes through suburban neighborhoods, and cleaning office buildings in Silicon Valley.

Make the internet great again

Until the layoffs, there was a large constituency of tech workers whose salaries depended on not understanding what their bosses were really up to. The top-tier, and those who aspired to it, weren’t just there for the paycheck, or the cornhole and kombucha on tap. They believed that, like the bosses in their younger days, they were “doing well by doing good.” Those promises proved empty just as hype about generative AI cranked up. The California dreams of a post-pandemic return to normalcy burst in the party game of P(doom), the swirl of Elon’s purchase of Twitter, the WGA and SAG-AFTRA strikes, and the bosses writing big checks to Trump.

As the founder heroes who commercialized the World Wide Web got older, weirder, and more distant from operational decision-making, their corporate managers have acted…well, like corporate managers. They never cared much for building digital god or dreaming of cybernetic meadows. Management focuses on efficiencies and margins, and nothing makes numbers go up like replacing expensive, unreliable human capital with machines. Set aside the science-fiction prognostications about AGI and the future of transformer-based language models is just a series of questions about how many jobs can be eliminated and how soon, along with how many new jobs the technology will create.4

What will twenty-somethings get up to as entry-level technical and managerial jobs disappear along with admissions to graduate programs? Doctorow wants them organizing and building countermeasures like Para, putting digital tools in the hands of workers and activists. At the heart of the book is a hope that Silicon Valley overreach and Trump’s chaos will result in a backlash that inspires young people to revive the labor movement and energizes tech regulation.

Fixing the internet might seem like a first-world problem compared to climate change, US military actions at home and abroad, and mass murder in distant lands, but Doctorow’s answer to that charge is that global problems require global solutions. This means a technical infrastructure capable of coordinating global social and political movements. That’s what the internet of twenty years ago provided. Doctorow wants it back, stronger and better.

We can build a better, enshittification-resistant digital nervous system, one that is fit to coordinate the mass movements we will need to fight fascism, end genocide, and save our planet and our species.

Fear of a fully braudrillated society

But doesn’t it feel like the internet is part of why everything got worse? Sure, the markets were hijacked by technofeudalists, but it wasn’t the tech lords and their minions alone who created toxic cesspits that seep addictive poisons into the flow of information through our devices. Consumer desires and the unwisdom of crowds explain a lot of what’s happened. The tyranny of images chaining us to screens is a collective creation, the outcome of digital culture tied to algorithmic systems of distribution. Those systems are an out-of-control machine preventing us from building what humans need to flourish materially and spiritually.

Thoreau’s call to “Let your life be a counter-friction to stop the machine” and Mario Salvio’s elaboration of that line feel like righteous resistance to the mutual hell we have created.

There’s a time when the operation of the machine becomes so odious, makes you so sick at heart, that you can’t take part, you can’t even passively take part, and you’ve got to put your bodies upon the gears and upon the wheels, upon the levers, upon all the apparatus, and you’ve got to make it stop!

Luddites quoting these guys have a point. It is not so easy to disentangle the good elements of the techno-social apparatus from the bad. Changing who sits at the top doesn’t change the system. The billionaires playing the role of puppet master delude themselves as well as their audience about the extent of their power. Tinkering on the edges is a distraction. Maybe the answer is to wreck it and get to building human systems that avoid technology, or at least slow its development. Using digital tools to build better systems traps us in digital hell.

To his credit, Doctorow takes on this position directly. He calls Audre Lorde “manifestly wrong” that “The master’s tools will never dismantle the master’s house.” Doctorow ignores the context of her words, but then so do those who brandish the line to argue against using digital technology for social justice.5

The advantage of Doctorow’s focus on putting tools to good purpose is that it avoids hand-wringing about habits of mind induced by new technologies or apocalyptic warnings of imminent cultural collapse. Both would sound familiar to someone living two hundred years ago. The feeling that everything is getting worse is as old as modernity and as new as the latest story in your feed. So is the sense that we need to do something about it, and uncertainty about where to start.

Disenshittify U

The value of Enshittification is not how it explains the feeling that everything is worse. It is in translating that feeling into action. Doctorow could insist that the word he coined means just what he says it means. In fact, he says enshittification “isn’t just a way to say ‘Something got worse.’” But then, recognizing the expansive everything in his subtitle, he adds a footnote: “Though it’s fine with me if you want to use it that way!”

The Unaccountability Machine: Why Big Systems Make Terrible Decisions—and How The World Lost its Mind (2024) by Dan Davies offers an account of this more expansive sense of enshittification. Davies agrees that everything really did get worse. He explains this as a decrease in accountability, more a systems problem than evil deeds. Bureaucracies have separated the life-changing decisions made by corporations from people who could explain those decisions and do something to fix bad ones. Organizational complexity is a long-standing feature of modern life, captured by terms like Kafkaesque, catch-22 and bike-shedding. The increasing separation of persons employed in organizations from the decisions being made has “weakened the failsafe concepts of accountability that protected us from bad decisions and bad people.” Life got more complicated, and now there is no one able to help navigate the complications.

This is consistent with Doctorow’s account of how bad people running giant technology companies ruined the internet. They deliberately added complexity, and hid it behind frictionless interfaces that trap consumers. Platform enshittification explains other ways life got worse. Think about the hours spent trying to correct an error with your health insurance or resolving a problem in your employer’s HR system. The most egregious forms are when governments add layers of reporting to the social safety net, forcing those who lose their jobs, can’t afford groceries, or become disabled to spend hours and hours submitting information online or stand in a queue to talk to a person. By design, this means many who need those benefits don’t receive them because they cannot navigate the bureaucratic demands or quit trying in frustration.

Adapting the old Anglo-Saxon word into something that means poop smeared across digital platforms is a silly, juvenile way of describing this situation. We need more language like it. The value of enshittification is that it undercuts the theoretical seriousness of explaining postmodern conditions and the social scientific demand to gather data, always more data, to validate what we feel is real. Yelling Stop! gives voice to the feelings but offers little more than a return to the garden. Yelling about how the internet turned into the shitternet is a call to collective action.

Disenshittify your government! Disenshittify your university! Disenshittify your life! This is a call to bring out the tools to clean the machines. It offers a sense of joyful engagement with dirty work that needs doing. It turns us all into custodians, workers who do the necessary and important work of taking care of people and their shit.6

AI Log, LLC. ©2026 All rights reserved.

Share

I’ve been using Notes as a commonplace book, a place to mark writing on Substack that strikes me as worth remembering and sharing. Lately, as I cast about looking for the good stuff, I’ve been finding writers writing on topics I think are important only to wonder how much of their production is influenced by the use of transformer-based language machines. I feel as much curiosity as outrage. “You’re stealing readers from the real writers,” I yell to the clouds. But I also wonder what it feels like to run a hot-take manufacturing operation. Is it rewarding? Are the few beers or coffees each month worth the grind?

It is possible, even likely, that many of the writers I find myself wondering about are not using LLMs to manufacture content, but unconsciously conforming to the smoothly machined style that the essay read online now seems to demand, one that increasingly dominates what I see in Notes. I have always attended to voice and style as I decide where to put my attention, but now I look for writing that is clearly crafted out of materials far removed from the outputs of transformer-based language machines.

Rather than wring my hands about LLMs and cultural change, I thought I’d offer up a brief essay recommending my all time favorite discovery on Substack.

Looking around, it is hard not to feel that people in our society, even in so-called serious fields like government, academia and journalism, are being rewarded for style over substance, and that this has been true now for a very long time. This occurs, obviously, on both sides — we the woke by and large want for ourselves and other people to sound woke, not to be woke, since being woke sucks as an experience.

—Rafe Meager, Caveat Vendor

Is there anyone better at using the essay form to understand the experience of living and working within the institutional constraints of the modern research university than Rafe Meager writing at ?

Their latest essay joins three others in my personal collection of the best writing on higher education this decade. Snippets don’t do them justice. I urge you to read Caveat Vendor from beginning to end and return for a few thoughts.

Time for change

I think the word “timeless” does not quite fit here. Probability is a theme that runs through Meager’s writing. The unfolding nature of time as I experience it, along with my faith in contingency, has me believing that education is ever changing. It is the primary means humans have to adapt to their environment, which like human culture is always in flux. Teaching and learning at universities feel timeless because the process has been going on in institutions with that name for centuries, and some of the forms persist, even though they are transformed by their context.

From the inside, change feels so damn slow as to not exist. Yet, seen across the centuries, especially the last two, the institutional arrangements that structure educational labor are wrapped up in the chaotic historical processes we give names like industrialization, globalization, and capitalism. As global knowledge factories, universities are where these processes happen and where they are understood, or at least contemplated.1

Attention to how teachers and learners, each with their own streams of experience, are entangled in complex social processes is one of Meager’s great strengths as a writer. Without getting lost in abstractions, Meager is clear about what makes the current arrangements so infuriating: they could be so much better! “But it is not inevitable that academia ended up this way, nor is it inevitable that we continue to bend the knee to the logic of money and power.”

Academic work is rewarding in so many ways and the rewards are so sadly compelling. It makes even moderately successful academics reluctant to use what influence they have to resist the application of managerial efficiency to educational processes which, because they involve shaping humans in all their precious diversity, cannot be easily scaled or optimized. It is not that university professors are powerless or blind to this problem; rather it is that they choose to preserve their rewards and precious autonomy instead of working to improve the lot of those subject to unthinking bureaucratization: the students, the staff, and untenured colleagues. The most outrageous problem, one Meager treats incisively in this essay, is when autonomy lets the powerful get away with abuse. Academic success creates monsters such that “a persistent minority of very bad people make this house a horror show for everyone, and I grow increasingly repelled by the silent majority that stands by and watches and enables it.”

Yes, and the confusions of the past few years as social movements we give names like "Me Too" and “Black Lives Matter” seemed to rise (finally!) to the occasion only to appear beaten back has left me feeling hopeless. And yet, I “keep ushering new people into an increasingly broken-down house, unsure whether you can, in good conscience, sell it to them.”

And I don’t know what to do about this. Leaving my full-time job as a university bureaucrat to teach a few classes on an adjunct salary and write—ah! my precious autonomy—felt like giving up the fight. In my worst moments, I see the defunding and chaos of the Trump regime as an echo of my own despair—a welcome wrecking ball. With cleared ground, maybe we can start anew. But this is nihilism. I’m not so radical as to give up on the one institution in civil society that seems, on balance, to do good in the world.2

“It is hard to design ways to overcome thorny, uncertain, high-risk collective action problems,” Meager writes. They share, with Dan Davies, an insider view of the economics profession that, along with its many failures, offers models and ways of thinking that make these problems appear tractable. But models don’t use themselves, and the confusions of the past few years make it feel near impossible to even suggest new approaches to improving our increasingly unaccountable institutions of higher learning. I don’t just mean economists fall short; I mean all those who practice the human sciences, my own tribe of historians included, seem incapable of making repairs to the houses where we live and work.

The great power of Meager’s essays is the elevation of the affective dimensions of academic life, and the feelings about these failures in particular. They treat feeling as a necessary complement to thinking, a way to identify truth, feeling truth as a way to make change. Similarly, their writing braves the division between quantitative and qualitative that structures so much academic feeling and thinking. Meager treats these distinctions as relations, not divisions.

That’s enough gloss. My purpose here is to encourage you to make time for a writer who does Substack right. Instead of churning out three essays a week and putting at least one behind the paywall, Meager publishes one every few months. Each is an occasion for me, as I hope they will become for you.

Here are my three favorite Rafe Meager essays on higher education.

The Universal and the Vacuous Events

Taking Our Chances

“That mortal is a fool who takes joy in his prosperity…

Read more

a year ago · 81 likes · 34 comments · Rafe Meager

The Universal and the Vacuous Events

Tormented by an Urn

The basic Polya urn works like this: the urn is filled with balls of two colours, say red and black, mixed together in unknown proportion. Draw a ball from the urn and note the colour, then return the ball to the urn and also drop a new ball of the same colour into the urn. Then do the same thing again, and again…

Read more

2 years ago · 40 likes · 18 comments · Rafe Meager

The Universal and the Vacuous Events

Finish Your Projects

The year after I went on the job market (academic, byzantine, and weird) an economist I greatly admired agreed to meet with me about my job market paper. I wanted his advice on what to do about the serious and terrible flaws in this paper which had been revealed many times over the course of the market. (The paper is good, don’t get it twisted. All pape…

Read more

3 years ago · 575 likes · 38 comments · Rafe Meager

AI Log, LLC. ©2025 All rights reserved.

Share

It wasn’t sleeping, but where was it, within itself? But then it didn’t, as he understood it, possess a self to be within. Not sentient, yet as Lowbeer had pointed out, effortlessly anthropomorphized. An anthropomorph, really, to be disanthropomorphized.

—William Gibson, The Peripheral

In the summer of 2022, Blake Lemoine became famous for telling reporters that a technology he worked on at Google, called Language Model for Dialogue Applications (LaMDA), was sentient. This was a sign, a glimmer, of what was to come. Not sentient AI. Rather, the experience, and stories about the experience, of playing language games with a generative pre-trained transformer, its outputs refined using reinforcement learning from human feedback. ChatGPT offered millions this experience, and the feeling that there must be sentience behind the sentences appearing on their computer screens.

Joseph Weizenbaum named this feeling after ELIZA, a computer program he designed and built in the 1960s to role play as a therapist. The ELIZA effect is the effortless anthropomorphizing of a tool that talks with you. As Weizenbaum warned, this can be harmful. It was for Lemoine. Google fired him, calling his claims “wholly unfounded.”1

In the spring of 2023, Sam Altman became famous for telling reporters that something sort of like sentient AI was nearly here and that OpenAI would ship it soon. Many were skeptical. They pointed out all the ways GPT-4 was not that smart and that artificial general intelligence (AGI), Altman’s preferred term, was pretty vague. This skepticism was widespread, but journalists were more interested in the experience of chatting with ChatGPT and worries about students using it to do homework. And they really liked writing about AI startup founders and scientists confidently predicting what would happen next.

Three years in, that’s changed. There is broad agreement that new models are better, but not super smart. The fact that ChatGPT is a homework machine is just one of many problems facing educators. Journalists are filing more stories about the AI bubble than about AGI. And lately, when they tell stories about the chatbot experience, it is in the context of increasing evidence that it makes mental health problems worse.

Seemingly Socially Conscious Language about AI

Recently, Microsoft’s Mustafa Suleyman argued against what he calls Seemingly Conscious AI. Treating machines that generate language as potentially sentient or conscious

is both premature, and frankly dangerous. All of this will exacerbate delusions, create yet more dependence-related problems, prey on our psychological vulnerabilities, introduce new dimensions of polarization, complicate existing struggles for rights, and create a huge new category error for society.

No doubt Suleyman is sincere in his social concerns. Still, it is worth considering the split in Silicon Valley between those who rely on the ELIZA effect to sell their products and those who worry about its harms. While the founders touted general intelligence, the sober-minded executives running some of the world’s largest technology corporations kept their distance from the AGI bandwagon, even as they joined the AI parade.

Asked about AGI in early 2023, Google CEO Sundar Pichai sidestepped, saying “ it almost doesn’t matter [whether we call it AGI] because it is so clear to me that these systems are going to be very, very capable.” What they are capable of was—and remains—the question. Along with Amazon and Facebook, Google has been swept along in the excitement about AGI and the capabilities of language machines, investing huge amounts of capital out of fear of being disrupted as they once disrupted. Rather than building their own large foundation model, Apple and Microsoft are approaching large language models by experimenting with models of different sizes and purposes, orchestrating and integrating them into their existing products and services.2

When it comes to the education market, Microsoft and Google matter the most. They sell the vast majority of system- and institution-wide information systems and services—they like to call them platforms—along with the cloud infrastructure to run them. Neither company cares about making chatbots entertaining, nor do the managers who sign those enterprise contracts. Having people chat the day away with a digital buddy is not the path to automating tasks efficiently and optimizing workflow. Copilot is less a smarter Clippy and more a strategy to make sure that asking Chat to do your work for you does not replace Microsoft’s existing business. Gemini is the least chatty of chatbots, in part so it will help you get your work done. It is designed to be integrated into Google Workspace and Google Classroom.3

Calling this language-as-a-service helps distinguish upselling existing customers on language machines from selling people on spending time with a chatbot. Microsoft is not selling AI—it is protecting BI, its data analytics and visualization tool, along with its other brands like Excel, PowerPoint, and Word. It has rebranded Azure Cognitive Services as AI Foundry, imagining software developers as industrial workers, forging new tools and products for the twenty-first-century knowledge factory. Microsoft wants your organization’s IT staff to build natural language integrations with their tools and your data stores. It doesn’t care which models provide the integrations, as long as Microsoft sells the service.

Google and Anthropic are similarly focused on selling the services of language machines to organizations, but they also compete in the chatbot arena. Currently, their models are on top. Yet, it is OpenAI with its brand name advantage that is in the spotlight as people begin to question the value of anthropomorphic chatbots. Understanding what ChatGPT does and does not do reveals the problems with OpenAI—both its path to profitability and the harms to its customers.

was early to the psychological dangers of LLM-based chatbots, writing six months ago about a family member who has bipolar disorder. Like Lemoine’s story three summers earlier, Riley’s story was a sign of things to come. As families file lawsuits and tell their stories to the New York Times, OpenAI flounders, pointing to its terms of service while talking up its newly added parental controls. Moving fast and breaking things works until it doesn’t. When you misunderstand the technology you are building, then it really doesn’t.

Distinguishing language from cognition

Riley recently published an essay in The Verge making the case for this misunderstanding in terms of cognitive science, writing that the theory behind large language models as a path to superintelligence

is seriously scientifically flawed. LLMs are simply tools that emulate the communicative function of language, not the separate and distinct cognitive process of thinking and reasoning, no matter how many data centers we build.

The notion about thinking machines that underpinned this theory was made famous by Alan Turing when, in 1950, he “proposed to consider the question, can machines think?” His thought experiment became known as the Turing Test. When ChatGPT passed it, those who imagined Turing’s musings as a simple science experiment were confused, believing it must be a sign of human-like intelligence. It wasn’t.4

Generating meaningful language through computational processes and successfully applying those processes to word problems is a genuine accomplishment, but passing tests does not make a thinking machine. With its impressive outputs and failures to make sense of experience, ChatGPT makes clear that manipulating language is not the same as cognition. Riley calls upon the work of Alison Gopnik to remind us that humans demonstrate this truth as well. Babies engage in all sorts of cognitive processes prior to using language.5

This uncontroversial view goes back to the early days of psychology. In Principles of Psychology, William James describes cognition as a stream of thought—more famously, as a stream of consciousness.

Consciousness, then, does not appear to itself chopped up in bits. Such words as “ chain” or “train” do not describe it fitly as it presents itself in the first instance. It is nothing jointed; it flows.

James explains the confusion that arises when language makes it appear that cognition happens through chaining or training discrete bits or tokens together.

The confusion is between the thoughts themselves, taken as subjective facts, and the things of which they are aware. It is natural to make this confusion, but easy to avoid it when once put on one’s guard.

“Language works against our perception of truth,” he says, pointing in the same direction as Riley: an understanding of language and cognition, not as equivalent, but as relations in the stream of experience.6

When a person experiences the ELIZA effect, they mistake the words extruded from a language machine for “subjective facts” when the words are merely things about which the person is aware. The human-like outputs of language machines make the already confusing relations between thoughts and things more so. Chatbot makers and enthusiasts exploit this confusion, some more than others.

Why buy what OpenAI is selling?

Educators should, as Suleyman says, buy from companies that build technology for people, not to be people. And, there should be consequences for companies that harm people. Given its track record, OpenAI’s models should be kept away from people, especially children. If effective regulation is too difficult or distant a goal, perhaps market competition can address the problem.7

There is great writing by educators on how to use (or not!) large language models in teaching. But I’d like to suggest more attention be directed to which companies to do business with (or not!). Purchasing decisions about language machines, including whether to use products offered for “free,” are crucial to making digital tools serve students and teachers. An emphatic no to ChatGPT for Teachers does not necessarily mean yes to Gemini integrations in Google Classroom. There are other options.

Shaping markets in educational technology means teachers, technologists, and students mobilizing as citizens, consumers, designers, engineers, and educators to make, sell, and buy things for the better. This work started long before ChatGPT and will continue long after OpenAI is a footnote in the stories historians tell about the 2020s. As those stories play out, let’s agree to disanthropomorphize the chatbots.

AI Log, LLC. ©2025 All rights reserved.

Share

For more on competitive markets as a fix for technology that seems to delight us but actually makes us miserable, check out Cory Doctorow’s new book, Enshittification: Why Everything Suddenly Got Worse and What to Do About It.

Ben Recht has some idea

Disanthropomorphizing chatbots is one way to correct for the mass ELIZA effect precipitated by ChatGPT. Another is demystifying Reinforcement Learning from Human Feedback (RLHF), the technique that made generative pre-trained transformers so effective at playing language games.

If that sounds interesting, check out Ben Recht writing at . He has been live blogging his machine learning class “Patterns, Predictions, and Actions.” He also wrote the book, co-authored with Moritz Hardt. His new book The Irrational Decision, is is available for pre-order. I cannot wait to get my hands on it.

Curious about AI Log and what I do around here? Check out this brief guide or click this button.

Support AI Log

The term language-as-a-service helps make sense of Silicon Valley’s attempts to sell large language models dressed up as a Socratic tutor. This essay, like last week’s, imagines better uses for language machines.

For the word of God is alive and active. Sharper than any double-edged sword, it penetrates even to dividing soul and spirit, joints and marrow; it judges the thoughts and attitudes of the heart.

—Hebrews 4:12

Judgment, it has none. ChatGPT knows nothing of the thoughts and attitudes of the heart. This distinguishes the word of God and the words of humans from words extruded by language machines.

I learned from my students how freeing ChatGPT’s lack of judgment is. They play language games with a chatbot without fear of being judged—not because they are fooled by the Eliza Effect into thinking there is a mind on the other end of the interface, but because they know there is no mind there to judge them. They “ask Chat” knowing that they play a computer game.

Students can see the technology is double-edged. Rely on Chat for learning or companionship, and you miss cognitive and social experiences that lead to growth. They insist that it serve them as an educational tool.

The language machine’s lack of judgment brings educational advantages. Presenting an essay draft to Chat for review is no big deal. The model delivers edits and suggestions, and its words do not cut as a teacher’s red pen does. I invite them to experiment with an LLM review as a step prior to showing an essay draft to their peers and their teacher. This adds a tool to the writer’s bench, rather than replace human evaluation.

In the race to superintelligence, corporate competitors run past such learning without a glance. The transformation of Grammarly into Superhuman is a disappointing example. What was called Grammarly now comes wrapped in a productivity suite called Superhuman Go. The old Grammarly did a better job than I could of marking up each line of student drafts, and did it with all the personality of a typewriter.1

In college writing, line edits are mostly a matter of correcting carelessness. I use a green pen, but whatever the instrument, a teacher’s judgment draws blood. Automating this task felt like advanced spell-check—a way to avoid have a student’s easy mistakes witnessed by a teacher. Correcting patterns of error in punctuation and subject-verb agreement is something I will happily hand off to a machine.

Not so when it comes to reading and writing, the social processes of learning. Social context is not machine-readable, and so teaching cannot be automated. Whatever intelligence language machines have, it does not amount to wisdom.

Technology companies aim for superintelligence and miss the mark. Rather than double down, we should change their game. If we want to experiment with LLM-based learning tools, let’s try making language machines purposefully stupid. That should not be difficult. As someone once said, “Two things are infinite: The universe and human stupidity; and I’m not sure about the universe.”

The language-as-a-service known as Gemini told me that Albert Einstein said that. This is bullshit. More politely, a confabulation.2

Misattributing the line to Einstein is an example of human stupidity and artificial intelligence working together. Confabulation is a human phenomenon, one reflected in the outputs of language machines. Language models do it because they are natural language processors. The falsehoods generated are not simply products of unvalidated probabilistic maths leading to error; they are the result of simulating human confabulatory tendencies and biases.

Educational talk about transformer-based language models is wrapped up in how intelligent an LLM appears. Like our students, we subject these machines to batteries of tests to rank them, and then focus on best scorers. Yet, the best test-takers do not make for good teachers. Like the most popular lecturers, they are more entertaining than enlightening. Attempts to market language-as-an-educational-service often describes LLMs as Socratic tutors. They put a wise mask on Shel Silverstein’s homework machine.

I have written about the use of digital necromancy in education: having students play language games with an LLM trained on a dead person’s corpus. I find the practice vaguely horrifying, and students seem to find it deadly boring. They cannot make a Socrates, but perhaps they can make a useful simulation of Phaedrus. Or Thrasymachusor. Or Hippias.

The Socratic dialogues present a Platonic form of education.Socrates looms large in our educational imagination because of Plato’s brilliance as a writer. The dullest interlocutors are posed with the wisest of characters, and in the hands of Western philosophy’s great writer, reading becomes an education in learning to think about complex ideas. Language machine cannot enact this Socratic ideal, and when they try, students do not sit still for it.

There are more interesting possibilities than building an artificial Mark Hopkins on an AI Log. You don’t need so much compute to model ignorance. If we ask students to play a language game with a machine, let’s cast them as Socrates and the machine as the problem. LLMs habitually perform the Dunning–Kruger effect, so their outputs are nearly always opportunities for skeptical analysis. Call it a Phaedrean learning tool if you want to name it after a dead man.3

Imagine a group of students playing a language game with a machine that prompts them to talk about a poem. The model starts by offering a misreading: the sort of analysis you might get from a student who barely glanced at the book before class. The task of the students is to judge the model’s outputs, pushing themselves to read the text more closely and carefully. Imagine students working for twenty minutes in small groups, prompted by their language game to produce a better reading of the poem. The teacher turns the machine off and engages the class in discussion.

The problem with Silicon Valley’s approach to tutoring, exemplified in the homeworker helper models, is its neglect of the social nature of learning. Orchestrating a conversation across a group of students gathered in a classroom is far beyond the capabilities of even the largest language model. But a small, stupid one offers a learning activity mediated by a language machine, a first step on a teacher-led journey to understanding.4

Why try so hard to make the language of language machines alive and active?

Why not make it stupid and interesting?

AI Log, LLC. ©2025 All rights reserved.

Share

This homily and its sibling, Language Machines as Spiritual Tools, are a departure from my usual long-form essays about books and teaching. I’m curious what you think of them.

Leave a comment

To receive future essays from AI Log in your email inbox, simply…

Subscribe now

The term language-as-a-service helps make sense of Silicon Valley’s “pivot” from building God to selling large language models as additions to their “productivity suites.” This homily imagines better uses for language machines.

For the word of God is alive and active. Sharper than any double-edged sword, it penetrates even to dividing soul and spirit, joints and marrow; it judges the thoughts and attitudes of the heart.

—Hebrews 4:12

Transformer-based language models have passed so many tests, including the one named for Alan Turing. Yet, their computational language processing falls short of human cognition in many important ways, while exceeding it in a few. That doesn’t make them useless; it just makes them language machines.

“History offers plenty of evidence that language technologies are as double-edged as any sword,” I wrote last week, pointing to the social dangers of machines that make use of “language in its age of algorithmic reproducibility.” Leif Weatherby’s phrase in Language Machines describes the tests humans face now that the outputs of ChatGPT appear “alive and active.” That apparent miracle prompted heralds in Silicon Valley to cry aloud the coming of superintelligence.

What we have instead is yet another general purpose technology. We know from past experience that these tend to wreak havoc on the social order, so we brace ourselves. The wise among us know that it is impossible to predict what will change and how.

Like the language machine called the printing press, transformer-based text generators may become swords that rebound, striking the corporate bodies that wield them, cutting them to their joints and marrow, dividing their soul and spirit. Recall that Martin Luther was a Catholic priest and that Johannes Gutenberg was the son of a patrician, whom the archbishop of Mainz appointed a court man in 1465.

It may be that ChatGPT is too warped by the values of its makers to offer anything useful to those who work toward reform or revolution. Massive foundation models are where all the attention and the money go, but small, open models, the stuff of tinkerers, hobbyists, and research labs, are where the interesting action is. When artists and teachers join in, we may all be surprised.

It is a mistake to cede all spiritual considerations of what language machines produce to the engineers and oligarchs preaching superhuman intelligence. Their theology is basic because their understanding of how language functions is simple. They see that words classify, but not how meanings multiply when words are put to other uses. They do not read scripture or modern poetry.

Gertrude Stein’s sentence “There is no there there.” distinguishes type from token. Like all language, it signifies more than it can say in words. This is a lesson logical positivists taught themselves a century ago, which is why there are no more logical positivists.

Literature and religion turn multiplicities performed in language into rich understandings of the world. With its binaries, computation has served narrow science, but applied to language in the processing of transformer-based models we have something new, and at the same time, older than the mathematical mysteries of Pythagoras and the literary theories of Protagoras. We try so hard to fit all into the existing order of things that are, and things that are not. What of those that are both and neither?

Instead of a talking encyclopedia or a hands-free browser, let’s imagine a language model built for the purpose of human contemplation of questions beyond our current horizon. Leave behind, for just a moment, anxieties about deskilling cultural labor and the homework apocalypse. Let’s specify that the model we build respects the rights of artists to earn a living from their work and avoids scraping the ickiest crevices of the internet. Instead of tuning it to play language games as a sycophantic digital servant or a character from a science fiction novel, let it be trained to speak according to the oracular ragtime of Whitman’s 1855 preface to Leaves of Grass.

Would I find faith in the words of a model whose training data has been purified, whose small size and open weights align to my Puritan creed? Whose outputs come, not wrapped in an anthropomorphic simulation, but as a simulacrum of the sacred, a holy copy with no original?

I resist this new thing for reasons I do and do not entirely understand.

Applying a language machine to sacred texts may not reveal the word of God. As a humanist, my definition of sacred has always assumed human authors, and I remain convinced by a long ago encounter with literary theory that meaning emerges from communities of human interpreters. In this view, writers relinquish authority over text through the act of publication. After that event, any attempt by an author to say what they meant is an error.

If words become sacred, not through the putting of pen to paper, but through collective acts of human interpretation, then perhaps the de-authorization of text by LLMs is analogous to the workings of history. The words we associate with Jesus, Socrates, and Siddhartha are the outputs of historical algorithms unfolding over centuries of scribing and printing, bracketed by primary and secondary orality. These words persist, sanctified and translated through time, resulting in a certain probabilistic vagueness about their origins that serves many purposes. An LLM produces words through different, yet potentially useful processes.

My instinct is to defend the boundary, to argue that only human history can sanctify texts. What errors are computed in the de-authorized, algorithmic processing of language machines!

Yet, I would encourage seekers who wish to spirit away large language models, take them from the oligarchs and grifters and put them to better use.

When I think to apply a language machine to my own spiritual practice, I find doubt. I keep my Sabbath at home, mixing and collating sacred texts through reading and writing.1 It is not clear to me what adding “compute” to this already complex processing offers. Yet my Whitman and my James urge me to invite the Kosmos to play dice with their words in the manner of the I Ching, a Dadaist poem, or even a stochastic parrot.

AI Log, LLC. ©2025 All rights reserved.

Share

It is very unhappy, but too late to be helped, the discovery we have made, that we exist. That discovery is called the Fall of Man. Ever afterwards, we suspect our instruments. We have learned that we do not see directly, but mediately, and that we have no means of correcting these colored and distorting lenses which we are, or of computing the amount of their errors…Nature and literature are subjective phenomena; every evil and every good thing is a shadow which we cast.

–Ralph Waldo Emerson, Experience

There is a belief devoutly held by the rationalist evangelicals working on superhuman intelligence that what they build is alien, nonhuman. Strangely, their most ardent critics believe this too. Good or evil, super or mid, whatever it is that transformer-based large language models are, both sides agree: it is not human. The difference between the prophets and skeptics of AI lies in how they enact their convictions. Rationalists fall to their knees in awe; skeptics run to the barricades.

In Language Machines, Leif Weatherby uses the term remainder humanism to damn the skeptics for this shared sin. It runs them in circles, hobbled as they are by “a distorting focus” on “a moving yet allegedly bright line between human and machine.” Their blurred picture leaves them unable to think clearly about machines that produce language through computational processes. The term aims to provoke literary theorists, linguists, and scientists who study cognition and data into updating their theories.

One alternative to remainder humanism, at least for those who do not wish to join what Dave Karpf calls “not exactly a cult, but not not a cult,” is to understand transformer-based language models as cultural technology. To see these models, as Weatherby does, as “culture machines, and more narrowly, language machines.” To agree with Alison Gopnik that they are analogous to cultural tech we take for granted after centuries of use: the printing press, the graphite pencil, the open-stack library—instruments that were shaped by humans, and now shape human thinking.

It is disappointing how little attention this idea has received over the past year, despite the work of Gopnik, Henry Farrell, and their collaborators. I lump Weatherby in with those writers and the AI-as-normal-technology duo of Arvind Narayanan and Sayash Kapoor, who are attempting to escape the semantic wrangling over cognition that the term artificial intelligence invites.

The medium is the massage

Kevin Munger, writing at , offers a hopeful take on the prospect of moving beyond wrangling. His McLuhanesque pronouncements—“the idea of ‘antimeme’ was once an antimeme but has now, in certain spaces, become a meme” and “The antimeme is the message of the medium, defined negatively”— suggest that if we can “identify what the medium is telling us by observing what it is unable to tell us,” we might bring the actual processes of large language models into focus. Not just the algorithmic processes that produce the words, but also the social processes that make those words meaningful. He says, “Processes are the antimeme haunting Western philosophy.” A revival of cybernetics is Munger’s main interest, but he usefully widens the frame to include a genealogy of process philosophy.1

Until Dan Davies gets interviewed by Ezra Klein and Tyler Cowen, and cybernetics becomes the new hotness, I fear we are stuck with warmed-over sci-fi tropes as the framework for debating what we should do about large language models. This frame lets the participants imagine themselves in a high-stakes battle, understood by one side as heroic defenders of humanity vs. diabolical agents of the machine, and by the other as heroic makers of the future vs. diabolical Luddites standing in the doorway to paradise. We would do better with more comedy of the sort offered by Davies in The Unaccountability Machine and Stafford Beer with his back-of-napkin illustrations.

Thinking about LLMs means attending to their social and computational processes, not telling romantic tales or tragic stories. All the arguing over benchmarks and bubbles, and the prospect of the always-almost arrival of AGI, has led to a notable lack of curiosity about what generative, pre-trained transformers actually do with words. Engineers talk compute and parameters. The social critics talk next-token prediction machines and text-extruding stochastic parrots. But there is more going on than just “mathy maths.” There is a there there.2

On beyond the dangers of stochastic parrots

Around the same time “On the Dangers of Stochastic Parrots” was published, poet Allison Parrish offered another way to think about the growing capabilities of language models, arguing that “the only thing computers can write is poetry.” Instead of insisting that words generated without human intention are without context or that the human brain is the sole source of language’s meaning, Parrish distinguishes between output as form and output as art: “The poem unites poetry with an intention. So yes, a language model can indeed (and can only) write poetry, but only a person can write a poem.”3

Parrish’s notion provides Weatherby with a start toward a better theory for what he calls “language in the age of its algorithmic reproducibility,” a theory that understands LLMs as “large literary machines.” This conception grapples with what he calls “a general poetics of computational-cultural forms.” Such forms exist not only in literature, but also music and the visual arts. This argues for better theory, and for greater consideration of musicians, visual artists, and poets who work with computational machinery. After all, humans have been using stochastic machines to create culture far longer than there have been technologies called artificial intelligence.4

This history is not directly relevant to Weatherby’s project, which is committed to the history of ideas rather than the history of cultural production itself. However, both theory and history provide frameworks for thinking beyond the dualism of human vs. machine, and for thinking about computational machinery that produces culture.

Language Machines is a marvelous exploration of social theory as it relates to LLMs, at least for those comfortable with the specialized philosophical language used to talk across disciplines. For this audience, Weatherby covers a wide range of debates and ideas, arguing that the structuralism of Ferdinand de Saussure and Roman Jakobson offers a promising theoretical lens for examining the complex interactions between language and computation within transformer-based language models, and visible in their outputs.

The essential Peirce

The book’s great value, even for those not so interested in structuralism, is its wide-ranging discussions—perhaps too wide for readers who prefer less winding paths—of relevant developments in social theory. The big thrill for me was discovering how many theorists are using Charles Peirce’s semeiotic these days. I frequently see references to his neologisms, like token and abduction, and assumed they were merely confirmations of my long-standing bias that Peirce, who died in 1914, is essential to understanding communication and media in the twenty-first century. Language Machines convinces me that something may be happening with this notoriously difficult thinker.5

Like his buddy William James, Peirce wrote against the dualisms that make up metaphysics as practiced by Europeans since Plato. Where James tends to mediate the two sides of any divide in search of new formulations, like his radical empiricism, Peirce names a third concept to disrupt the easy divisions of word/type, deduction/induction, or sign/signifier. This last dualism is at the center of Peirce’s semeiotic, or general theory of signs. Peirce’s third term, in this case, is interpretant, which names the process of making meaning from a sign.

Weatherby explains the process, starting with the importance of transmission in Peirce’s account of how a sign functions. “For Peirce, an untransmitted sign is no sign at all…the sign gives rise to another sign, and that second sign is called the ‘interpretant.’ This second sign is not a ‘user’ or maker of the second sign, not a mind or a person.”

Paul Kockelman’s Last Words: Language Models and the AI Apocalypse does a magnificently concise job of applying Peirce’s ideas to the problems and predicaments of LLMs. He writes, “In one sense, interpretation is simply the act or process of producing an interpretant.” The utility of this concept when applied to language produced by machines is that interpretation is no longer hidden by the blurry line separating humans and machines. Meaning happens to a sign—computationally in the case of LLMs; cognitively in the case of humans—though we need not get hung up on the distinctions.6

The breakthrough in 2022 was that we can no longer distinguish between the outputs of LLMs and those of humans, which renders the Turing Test useless. Now, we must answer questions about how machines manipulate language to create meaning, not await the arrival of thinking machines. Peirce gives us a handle. Both human and machine processes produce an interpretant. Again, Kockelman:

In other words, while machinic agents and human agents take very different paths, as it were, they can now arrive at similar destinations. At least insofar as the parameters of the former are aligned with the values of the latter, if only as channeled and distorted by the power plays of corporate agents.

Language as a service

Including Silicon Valley’s overwhelming influence in this account is critical. AI discourse is quite taken by ChatGPT and its competitors. The language of the largest AI models is channeled and distorted by their owners and their corporate agents, so when skeptics speak of AI, they speak the same language of apocalypse and optimization. But the largest models are not the only models, and chat is not the only way to use these tools.

Making other choices, in fact, just seeing that there are other choices, undermines the influence of corporate agents and tech oligarchs. Use small, open models. Boycott models made by OpenAI. Use language models to explore what Eryk Salvaggio calls “the gaps between datasets and the world they claim to represent.” Use culture machines to make art that illuminates how, as Celine Nguyen describes it, “the flaws of the medium become its signature.”

Of course, we should also pay attention to what the largest tech companies are doing as they desperately try to profit from these new toys and tools. Following Nick Srnicek, author of Platform Capitalism, Weatherby describes LLMs as offering “language as a service,” which both critiques the “as-a-service-ification" of the economy and suggests an alternative to intelligence and cognition for what the giant technology corporations are getting up to with LLMs.

“Language as a service” is not just theory. In September, Microsoft began offering Azure AI Language, “a cloud-based service that provides Natural Language Processing” to application developers. Services include detecting specific kinds of language, such as personal and health information, or the use of a non-English language or dialect. The features are basic, ranging from sentiment analysis and summarization to redacting health information and Social Security numbers. But in shifting from the brand Azure Cognitive Services to newly named services like AI Search, AI Language, and AI Foundry, Microsoft is providing a map to its offramp from the superintelligence highway to the mundane world of AI as normal cultural technology, a future in which language machines automate cultural labor just as machines powered by steam and electricity automated factory labor.

In this context, Weatherby's narrative aligns with the broader neo-Luddite and remainder-humanist critique, explaining how, through the creation of new jobs from the destruction of old ones, labor becomes a matter of attending to the machines. For example, human computers were replaced by machine computers, and new occupations, like programmer and tech support, were created. Whether you call this deskilling or reskilling, the creative destruction of employment has continued into the digital age, as bookkeeping has been subsumed into the work of business analysts. As language as a service grows, the work of coding, front-line customer service, and marketing becomes ensuring the quality of machine outputs and intervening when necessary. Task automation continues the story of the digital transformation. Managers armed with spreadsheets implement new enterprise platforms and software-as-a-service solutions, making their job of counting, optimizing, and forecasting easier. This increases profits while immiserating workers and customers.

Silicon Valley desperately needs these new language machines to be a genuine wave of optimization and automation, this time aimed at language workers at the bottom of the org chart. Perhaps it will work out as they hope, but this off-modern technology may not meet the high expectations of corporate managers. Large language models, like digital computers and printing presses, serve many social functions; some of them may work against the aspirations of the owners and managers of corporations. History offers plenty of evidence that language technologies are as double-edged as any sword.

New houses, new shoes, new language

The inventors of the transformer and the engineers building with it emphasize computation when they talk about LLMs, but the functions of natural language deserve as much attention. Artificial intelligence is a problematic name, as it creates mystifications that serve Silicon Valley’s pernicious influence over how social technology is made and made available. It masks the truth that all we call artificial is, in fact, a product of human thought and natural materials. As Emerson says, “Nature who made the mason, made the house.” This applies to technologies as old and familiar as shelter, and to those as new and complex as the transformer. It most certainly applies to language, that early and most human of all cultural technologies.

Critics who insist the language of language machine is nonhuman are like architects who start from a belief that they build “artificial caves” or podiatrists whose theory for treating foot pain is that their patients wear “artificial feet.” Houses and shoes are no more alien to human bodies than are the outputs of language machines to human minds. Poetry makes use of words as both subjective stream and external object, but academic social theory often seems overmatched by the complexity of language in process. For Emerson, Peirce, and James, and the modernist poets and cultural critics they influenced, the study of literature and the study of nature are inseparable.

The disciplinary bright lines that separate these activities explain something about the absent presence of cybernetics in the discourse about LLMs. Language Machines offers antimemetic insights in its discussion of Warren McCulloch and Walter Pitts, and its argument that Jakobson engaged directly with cybernetics. The connections between cybernetics and structuralism are all interesting, though I see no reason to limit the field to the structuralists. There were a lot of weird and interesting action happening under the banner of cybernetics during in the middle of the last century. For example, see Andrew Pickering’s The Cybernetic Brain and Ben Breen’s Tripping on Utopia.

Weatherby concludes the book by reading a passage from Roland Barthes’s 1964 essay “The Old Rhetoric: An Aide-mémoire,” that describes “a cybernetic, digital program” rooted in the dialectic as practiced by the Sophists. I confess, I don’t quite understand what Weatherby means when, following Barthes, he urges “the opening of this dialectic-cybernetic paradigm to probabilistic interpretation.” My confusion may be simple resistance to what I see as structuralism’s over-investment in dualisms, or it may be the first step in understanding what’s next for cybernetic cultural theory. Perhaps, it’s both.

AI Log, LLC. ©2025 All rights reserved.

Share

Note: For a review of Language Machines by a leading theorist of the cybernetic revival, see Henry Farrell’s Cultural theory was right about the death of the author. It was just a few decades early. For a concise argument that AI is human technology, see Alison Gopnik’s Stone Soup AI. For an illuminating interview with Leif Weatherby, see this post from the Journal of the History of Ideas blog.

Note to readers: This blog is about AI and higher education, including the social and political contexts in which we teachers, students, and administrators make decisions about technology. Without the freedom to ask questions and argue our evidence, our institutions will no longer be capable of creating and disseminating knowledge about AI or anything else.

Subscribe now

Higher education is but a sideshow to the Trump regime’s actions this week, which included a command performance meant to drive professional soldiers out of the military, an ICE raid in Chicago that violently disregarded the safety and rights of non-citizens and citizens alike, and another air strike against boats off the shores of another nation. With those stark acts contradicting the laws and norms governing the U.S., worries about academic freedom can seem almost beside the point.

Given the scope of what’s happening, and the stakes, how much could it matter if a few professors get fired or if a few universities find some way to keep the money flowing for critical scientific research? It feels like time to hunker down, just get through the next few years.

Yet the defunding of science and destruction of freedom on campuses are all of a piece, an effort to preempt college campuses as sites where resistance and coordination will happen. The election next year and in 2028 will soon take center stage, and the regime needs to establish some means of control. Destroying resistance and demanding obedience from university leaders is both a strategy and a chaotic diversion. It plays out with campus police ordered to arrest “outside agitators” who include our students and teachers, with university presidents being ordered by the White House to protect “our right” to re-elect the President. When it gets worse, we won’t have to ask “how did this happen?” Columbia University has shown the way.

Timothy Burke in explains the terms of the deal offered to the nine institutions and why it is “very precisely” an offer for money “in return for surrendering any vestige of institutional autonomy to Trump’s inner circle.”

Chad Orzel in responds by calling Burke’s argument for refusing these carrots “a lovely sentiment,” illuminating what I fear is a common hope for the prospects for compromise. Orzel believes those who say that institutions should refuse the deal don’t “understand the centrality of money to the modern practice of STEM.” Orzel writes “Refusing federal funds isn’t just “hard” or even “hard”, it’s existential.”

Indeed. I’ve used Thorstein Veblen’s term “trained incapacity” to describe this way of thinking. When Orzel argues we should “thread the needle between currying favor and bold defiance and find a way to sustain as much funding” as possible, he makes perfect sense—if it were 2024. Arguing it today misunderstands our circumstances. There is no trade-off here, no actual compromise possible. The threat is existential, but it is not to the funding of science; it is that science itself will be no longer possible.

Since the 1940s, the U.S. has had a system in which the federal government gives universities massive amounts of funding and the world receives rewards in terms of useful knowledge. That system is gone, and in its place a system of patronage is emerging that will be used to reward the regime’s allies and punish its enemies. Knowledge is now what people like Robert F. Kennedy Jr., J.D. Vance, and their newly empowered political functionaries say it is. Anyone looking to preserve science through compromise with autocrats and kleptocrats is deluding themselves.

Orzel became one of my favorite bloggers about higher education because of his clear-eyed view of topics that befuddle most academics. That he cannot make out the writing on the walls is worrisome. If the presidents and board members who are examining the deal listen to scientists like Orzel… If they decide, based on their outdated model of fiduciary responsibility1, to make some sort of deal… If Harvard is not playing the long game, but holding out for more carrots or false guarantees, then we are lost.

Orzel invokes C. P. Snow’s two cultures to explain “a pretty comprehensive failure on the other side of campus to understand the centrality of money to the modern practice of STEM.” That would be a fair observation in 2024. But this is no longer a misunderstanding among scholars, merely a matter of social scientists and humanists fighting for their professional practices and autonomy versus scientists doing critical work of advancing science. Historians like Burke and Jon Zimmerman are not warning about how difficult it will be to teach or conduct research in their culture; they are warning that the entire enterprise, the work we do in both cultures, is no longer possible.

The freedom to say gender is a complex social and biological phenomenon, that generating electricity using wind and solar is superior to burning fossil fuels, and that the starvation and bombing of civilians is a war crime are disappearing. More unfreedom follows, especially if the nine institutions accept the deal, or if Harvard cuts a deal.

I’ll stop here and invite you to read A Rotten Compact by Brendan Cantwell. It is an excellent and brief analysis of the deal.

I urge you to support the AAUP’s Center for the Defense of Academic Freedom.

Most important, I urge to rally support among your colleagues. Speak clearly to your president, provost, and other leaders about why they must hold firm against the temptation to negotiate with autocrats and kleptocrats.

AI Log, LLC. ©2025 All rights reserved.

Share

The images here are screenshots from Prezi, which is my preferred presentation tool. I provide links to references and image credits here so interested audience members can find them.

I pulled together recent thoughts on chatbots and use of large AI models in higher education to talk about generative AI as analogous to technologies that once felt dangerous and exciting, but today feel ordinary, normal, banal, even boring. This talk was originally called The Boring Revolution but just about everything, it sounds cooler in French.

One weird thing about transformer-based AI models is that most of us access them through an interface that is over fifty years old. This new-wine-in-old-bottles aspect to ChatGPT makes it difficult to understand the potential of the technology and use it for new purposes. We anthropomorphize it as a companion, a co-intelligence, a helpful intern, when in fact, it is powerful new cultural technology, best understood as language machines. We are only beginning to understand what they can do.

To call AI banal or normal is to call attention to historical processes that are shaping how it is being developed and diffused, countering the hype created by those who speculate wildly about its potential, usually to sell something. Like all inventions, including the dynamo, the internal combustion engine, and the open-stack library, AI technology will take a while to be incorporated meaningfully into our work and lives. Autonomous vehicles are an example of an AI technology that seems about to pass a threshold that will soon make it feel as ordinary as your public library or the street light outside your house.

Will that happen with generative AI? Of course it will, though we cannot know when or how things will play out. But doesn’t it seem like ChatGPT is already beginning to feel like la révolution banale?

I begin by asking the audience to apply the duck/rabbit image made famous (for philosophers) by Ludwig Wittgenstein to how we first saw generative AI. In those early days, ChatGPT was seen as either the Duck of Doom or the Rabbit of Glad Tidings. Today, most see it as double-edged, both a duck and a rabbit, as having the potential to be beneficial and harmful.

Then we talk about the fact that the people who released ChatGPT in late 2022 did so as an experiment, a way to gather user data. They were as surprised as everyone else at the result: the creation of what is often described as the fastest-growing computer application in history.

Then we compare the interface of Eliza, one of the earliest chatbots…

to ChatGPT’s interface. Pretty much the same.

The key terms for the talk are:

1. Natural language capabilities of generative AI have a much greater range of applications than simulating conversation.

2. Generative AI may be a general purpose technology but it can be used for specific purposes and narrowly defined tasks, including search within platforms and analysis of unstructured datasets.

3. Orchestrating various AI models built for different purposes and re-engineering academic and business processes (workflows) to use AI will be harder and more important than figuring out how chatty companions can help individual knowledge workers.

I use this article from The Markup to explain how organizations are using natural language search to manage information. AI. Neutron Enterprise, a large language model developed by Atomic Canyon, is not deciding anything. It is not even summarizing the millions of pages of regulations and compliance guidance employees are reviewing in order to extend Diablo Canyon Nuclear Power Plant through 2029. All the AI model is doing is finding the relevant regulations and guidance for each decision the workers have to make, so the workers themselves can interpret the rules.

My favorite example from higher education, is a natural language search feature developed by Courseleaf, a platform that manages academic processes like course rostering, registration, and academic planning. One of their modules is Curriculum Manager. This is a system faculty members use to propose new courses, and administrators use to manage the curriculum review and approval process. Natural language search allows faculty proposing new courses and programs, and deans reviewing those proposals, a better understanding of similar courses and programs being offered.

The event is sponsored by Explorance, so the audience will see demonstrations of MLY, their AI model, in action. It is a purpose-built model for analyzing student comment data gathered through course evaluations. For more on MLY, see this section of my essay On Beyond Chatbots.

In all my talks, I plug my favorite purpose-built technology for understanding complex ideas: books.

Specifically, I recommend the book AI Snake Oil. The authors, Arvind Narayanan and Sayash Kapoor, write one of the best AI newsletters. Their new book project, AI as normal technology, and their relaunched newsletter under the same name, offer an alternative to the hype about imminent superintelligence and predictions of an immediate transformation of work and education. Treating generative AI as normal technology focuses our attention on the social processes of adopting this new social technology rather than speculation about the unknowable future it may bring. I use their framework as a way to think beyond the chatbot and about the history of technology going back much further than the last three years.

I talk about how we focus on the invention of technologies and downplay the historical processes of innovation and diffusion. There are speed limits in each part of the cycle. Pay particular attention to diffusion, the peach-colored area.

The speed of progress happens at the speed of humans and institutions. Electricity, the internal combustion engine, and the electronic computer all took decades to understand, to innovate, and to commercialize. Those technologies didn’t feel normal when they were invented; they became normal through social processes of adaptation and adoption. Generative AI will be no different.

This leads me to my favorite “law,” which is about AI but applies to so much more in software development and in life.

Time for historical analogies!

The talk in Edinburgh was at The National Robotarium at Heriot Watt University. When I was in school, I learned James Watt invented the steam engine.

Now that’s not true, and just like the transformer—the “T” in GPT that is the key invention for recent large language models—and the processes used to pre-train the models, there is no single inventor of a general purpose technology.

I really hope that schoolchildren a hundred years from now are not told Sam Altman invented ChatGPT.

What James Watt did in the 1760s was innovate on a technology that was developed about fifty years earlier. Watt’s innovations allowed steam engines to be used for many new purposes. In the decades that followed, other innovators improved and adapted steam engines, resulting in railroad locomotives and steam ships, and powerful new manufacturing machines.

Then, just as steam was starting to feel a bit boring, an exciting new technology appeared. In the following decades, generating electricity became a lot less exciting. Think about all the boring tools we take for granted that are enabled by electrification. Indoor lighting, toaster ovens, household machines that clean our dishes and our clothes....

All that is based on the wild and crazy idea that we could generate that scary naturally occurring phenomenon known as lightning, and direct it into your house using wires.

Think about how dangerous and revolutionary that sounded to any normal person in 1880.

You can tell the story of electrification as a fight between big personalities with George Westinghouse and Thomas Edison as the Sam Altman and Elon Musk of the 1890s. You can also tell it as a boring story of patent lawsuits and corporate mergers.

Here is my version, and the protagonists are not inventors, entrepreneurs, or corporate lawyers.

My heroes are the innovators and diffusers. The people who worked for those early electric companies, the people who tinkered with light bulbs and lighting systems in their workshop or backyard. The local planners and residents who worked to light their cities and towns. The regulators and government officials who made it work on a national scale. The engineers who decided where to build power plants and the clerks who figured out how to bill customers. The people who solved all the little challenges of stringing electrical wires across the continent and connecting them to homes and businesses.

They slowed things down by asking questions, proposing alternatives, and listening to each other. To be sure, there were mistakes and problems, fires and accidental deaths. It took years of innovation to make sure we could safely light city streets and send power directly into homes.

An analogous process is playing out today with autonomous vehicles. And the heroes are not CEOs who make false predictions in order to generate attention.

That said, it looks like the experience of riding in an AV will be available next year in dozens of cities, including London.

Now people were saying something like that in 2009 when breakthroughs in mapping and sensors led to a lot of excitement and investment. How long before riding in a robotaxi is as banal as a tram ride or turning on a light switch?

There is an analogous revolution happening with information. Yet again. As Ada Palmer says, “We are an information revolution species.”

Since I started working at a university, we have gone from this:

To this:

To this:

We now talk with our data using AI models built out of nearly all the machine—readable information in the world, some linear algebra, and a great deal of computation powered by electricity. Just in the last two hundred years (my list starts with the telegraph), that’s information revolution four? five? six?

The challenge with understanding this revolution is less about making effective chatbots and more about reengineering and reimagining organizational processes. This is what Paul Bascobert, president of Reuters, calls orchestration in this Decoder interview.

In Edinburgh, I used an analogy drawn from the history of steam and electricity to illustrate orchestration with AI. Tim Harford in this essay draws on the work of economic historian Paul A. David to make the case that just as electricity required rethinking the physical arrangements and workflow of the factory in order to realize its value.

I believe orchestrating large language models into the flow of knowledge work will be how we realize the value of large language models. This analogy does not help predict exactly how this will happen (and may be misleading!), but it does suggest the people who work in institutions of higher education will play a crucial role in how these tools are to be used. After all, universities are factories where knowledge is created and disseminated.

That puts a lot of responsibility on people who work and study in institutions of higher learning. Technology does not determine how it is used, humans do. The process takes place over decades, not months, not years, but it happens as we make decisions together. These decisions should be made thoughtfully and with the needs of humans at the front of our minds, not the needs of large corporations like universities and technology companies.

Humans are tool creators. That’s what we do. We look around our environment, and we ask: what can we create that will make our lives better? And we invent things, and then improve them and share them. And these tools, these technologies are amazing! And often, we use them too much and we use them unwisely.

In closing, I talk about the contradictions in how I am teaching a course on AI this term. In order to learn about technology, my students and I put most technology aside. I didn’t ban or prohibit digital devices, but together we decided to leave our phones on a table by the door each day, and to keep our laptops closed unless we are doing something specific for the class.

And we have done it all semester. Gladly.

I teach about twenty-first century technology in a class that uses nineteenth century educational technology, and that most ancient of all edtech, the spoken word.

I believe we should inhabit this contradiction, and follow impulses that lead us to strengthen human social connections...put our screens away, and talk about what new technology means for our selves and our society. Face-to-face conversation was the first information revolution. It defines us as a species.

Note: These are paid speaking engagements where Explorance gives me a platform to share how I believe AI is changing higher education and I help their clients and other conference attendees think about how they use generative AI in their work.

I have known the CEO and leadership team at Explorance since 2008 and trust them when it comes to the responsible development of what we called machine learning until recently.

I do not receive money or other compensation for what I publish on AI Log. When I speak or write about AI as educational technology, the views I express are my own. That includes anything I say about a commercial product or service. See my disclosures statement on my website.

AI Log, LLC. ©2025 All rights reserved.

Share

I enjoy talking with people about how AI is changing higher education and what we should do about it. Click the button below for more information.

𝐀𝐈 𝐋𝐨𝐠 speaks

I wrote a piece back in May explaining why Trump’s executive order about AI in education prompted a chicken run of executives racing to line up for a chance at federal funding. Despite clear signs that the federal government was no longer acting in the interests of educators, NGO executives and higher education leaders responded eagerly to the prospect of new funding by saying nice things about AI and the Trump administration.

Universities may be non-profit enterprises, but they exist in a system that requires them to take in more money than they give out. Making budget decisions is the first job of a university president, and the board members who hire them are mostly successful corporate managers. Money is easier to measure than the creation and dissemination of knowledge, so presidents are evaluated in terms of their success in raising and maintaining funds. Nothing else really gets counted, so nothing else really counts.

Thorstein Veblen defined trained incapacity as “an habitual, and conventionally righteous disregard of other than pecuniary considerations.” Kenneth Burke drew a memorable analogy between the concept and the operant conditioning of farm animals.

If we had conditioned chickens to interpret the sound of a bell as a food-signal, and if we now ring the bell to assemble them for punishment, their training would work against them. With their past education to guide them, they would respond in a way that would defeat their own interests.

This week the bell rang, not to announce new funding, but to signal a new reality, one that was already apparent in examples being made of Columbia, Harvard, and UCLA. And so, to avoid punishment, university leaders fired staff and faculty for their speech. In Texas, the governor is insisting that Texas Tech go after students.

The response from campus leaders as they complied was incoherent. After Clemson University fired a staff member and removed two faculty members, they stated, “We stand firmly on the principles of the U.S. Constitution,” but then claimed “that right does not extend to speech that incites harm or undermines the dignity of others.” Other institutions dropped similar squawk bombs as they scrambled to comply.

Compliance involves more than censoring speech about a high-profile murder. After a professor was criticized in a video (posted on social media by a conservative politician) for violating “Trump’s laws” about gender ideology, the president of Texas A & M, Mark Welsh, fired the professor and removed her chair and her dean from their administrative roles. After deciding not to defend his faculty’s academic freedom, Welsh issued a statement claiming, “This isn’t about academic freedom; it’s about academic responsibility.” He resigned on Thursday as his critics complained his actions were too little, too late. His initial response to the student complaint had been to defend the professor.

Free inquiry and free speech are fundamental to the work of higher education. Campus leaders know this, but their training focuses their attention on the risk of budget cuts and layoffs. Those are terrible outcomes, and they hurt people, more people than the handful who posted ill-timed and distasteful words on the internet. I imagine that sacrificing a few individuals along with principles that protect speech felt like a difficult but necessary measure to see the campus through a challenging political situation. What could be worse than having money disappear?

Burke again:

If one rings the bell next time, not to feed the chickens, but to assemble them for chopping off their heads, they come faithfully running, on the strength of the character which ringing a bell possesses for them. Chickens not so well educated would have acted more wisely.

Fortunately, we are not all chickens. We have the capacity to see what is at stake and to defend the principles by which our institutions and our nation operate.

A wide, visible, and coordinated resistance is the best response to this abuse of power. Tell the institutions where you work and where you graduated that free speech matters. Donate to institutions that fight. Speak in support of leaders who try to protect academic freedom, even and especially when those attempts fail.

The bell is ringing. It is time for those who believe in free speech and academic freedom to speak and organize.

Share

Subscribe now

AI Log, LLC. ©2025 All rights reserved.

My work as a consultant brings me into contact with a lot of technology companies, and sometimes they give me access to their products at no cost, and occasionally they pay me to talk with them or give a keynote at a conference. That raises questions about influence and influencing, and so please indulge this effort at transparency.

The opinions I share here on AI Log and in talks like this one are my own. I do not accept payment to mention a product or company or to say anything specific about the products or the companies I discuss. I have a disclosures page on my website so that my readers can determine for themselves to what extent I am “a shill for AI,” an accusation once leveled at my colleague, Lance Eaton.

Subscribe now

Chatbots are an old technology

Three years ago, if you asked someone at OpenAI about putting a chat interface on their latest generative pre-trained transformer, they would have shrugged. At the time, OpenAI was super excited about their image generator, DALL-E 2, and the big AI story of the summer was about Google firing an engineer who believed the language model he worked on had become sentient.

There was no reason to believe chatbots were anything more than a novelty or an annoyance. After all, AI chatbots are more than fifty years old. Remember Clippy? Remember the early versions of Siri and Alexa? Heck, remember the last time you used Siri or Alexa!

Most AI researchers and product developers in 2022 believed that the chatbot was AI's past. AI's future was going to be something... futuristic.

The product developers who released ChatGPT in late 2022 did so as an experiment, a way to gather data. They were as surprised as everyone else at the result: the creation of what is often described as the fastest-growing computer application in history.

We don’t have time to go deep into why ChatGPT took off the way it did, but one big reason had to do with the collective sense that it was Shel Silverstein’s Homework Machine made real. The resulting homework apocalypse panicked educators and inspired pundits, resulting in a constant stream of over-excited speculation about what AI means for education.

You don’t learn much during a moral panic, but I believe that ChatGPT’s value, so far anyway, is in how it forces us to face important educational problems that long predate generative AI. Schooling has increasingly become a matter of measuring meaningless outcomes and not fostering student learning. As John Warner put it a few weeks after the release, ChatGPT can’t kill anything worth preserving.

Chatbots are yet another screen-based, attention-grabbing social media. Talking to “Chat” about your problems is just one more thing students do that isn’t spending time with others. The Eliza Effect, the tendency to treat chatbots like they are people, is now a mass phenomenon. Again, ChatGPT did not cause these social problems, and neither is it the answer.

Nearly three years into the GPT revolution, I think that the AI chatbot phase of this new technology is winding down. Today, we will look at a handful of tools that illustrate what I mean, but first let me offer a frame that avoids some of the distracting hype and speculation about AI.

Generative AI is normal technology

AI Snake Oil is my favorite book on AI from 2024. The authors, Arvind Narayanan and Sayash Kapoor, write one of the best AI newsletters. Their new book project, AI as normal technology, and their relaunched newsletter under the same name, offer an alternative to the hype about imminent superintelligence and predictions of an immediate transformation of work and education. Treating generative AI as normal technology focuses our attention on the social processes of adopting this new social technology rather than speculation about the unknowable future it may bring.

Excitement about how many people have AI companions or used ChatGPT to do homework obscures the wider potential impacts of the Generative Pre-trained Transformer. The hype around this invention is normal, even though the scale of capital being deployed to develop and promote generative AI is unprecedented. But all that money and hype won’t do much to speed up the pace of organizational change. This frustrates the big AI companies and their investors, as well as enthusiastic early adopters looking for evidence to back up their predictions.

The speed of progress happens at the speed of humans and institutions. Electricity, the internal combustion engine, and the electronic computer all took decades to understand, to innovate, and to commercialize. Those technologies didn’t feel normal when they were invented; they became normal through social processes of adaptation and adoption. Generative AI will be no different. This leads me to my favorite “law,” which is about AI but applies to so much more.

Let me emphasize a point Naryanan and Kapoor make: calling generative AI normal technology is not to discount its impact. These tools are powerful, and their capabilities have improved rapidly over the past four years. Transformer-based AI models are changing how we understand culture, education, and the relationship between language and mathematics.

Alison Gopnik and her colleagues help make sense of that change when they argue that transformer-based AI models are cultural technology, analogous to the printing press, the open-stack library, or the graphite pencil, each one normal in the sense that it took decades, and even centuries, for humans to understand and make use of them.

What do large AI models actually do?

What do large AI models like Claude and GPT do besides simulate a conversation and do homework? The range of answers is vast.

Here are three things that AI models do that I will show you today:

• They provide natural language interfaces to unstructured and semi-structured data.

• They are used to develop purpose-built tools that solve specific problems, including academic problems.

• They can be added academic and business processes, making them more complex, and therefore, more powerful and adaptable. This is often called orchestration.

The best way to think of how large AI models provide interfaces to data is to briefly review the history of information storage and retrieval systems. When I started my career, we used this technology, developed at the end of the nineteenth century.

Slowly, over the past thirty years, we adopted this technology.

Today, thanks to powerful large language models, we can now talk with our data by typing into a box or speaking into a microphone.

Chat interfaces are but one example of how AI models can help understand data. Let’s talk about some others.

Natural language search

My favorite example of natural language interface is this story from The Markup, from back in April. The headline is scary, right? AI managing a nuclear power plant sounds like the plot of a B-movie.

But as the article explains, AI is not managing anything. The product is a natural language search engine that uses “open-source sentence embeddings” to search through nuclear data to help humans find what they need—very useful if you work in a nuclear power plant, or just about any other enterprise that seeks to manage information and knowledge.

Here is how the article describes what’s going on at Diablo Canyon.

The artificial intelligence tool named Neutron Enterprise is just meant to help workers at the plant navigate extensive technical reports and regulations — millions of pages of intricate documents from the Nuclear Regulatory Commission that go back decades — while they operate and maintain the facility.​

The state of California recently extended the plant’s approval to operate through 2029. That means workers there are facing hundreds of decisions to replace or extend their safety equipment, looking at depreciation schedules, and updating procedure manuals.

Millions of pages of documents going back decades!

To be clear, the algorithm decides nothing. All the AI model does is find the relevant regulations and guidance for each decision the workers have to make, so the workers themselves can interpret the rules and decide what to do.

Given the overwhelming amount of digital information we contend with, having a tool that helps sort through unstructured data to help find what’s relevant is useful. There is nothing that requires jumping to have it automate decisions.

Natural language search of academic data

Some of you are on campuses that use Courseleaf, a platform that manages academic processes like course rostering, registration, and academic planning. One of their modules is Curriculum Manager. This is a system faculty members use to propose new courses, and administrators use to manage the curriculum review and approval process.

As anyone who has proposed a new course or is involved in reviewing curriculum proposals knows, one of the big questions in curriculum development is what else is out there—either already existing courses or thoe being proposed—that is similar to what you are proposing? You don’t want to poach students from another program with your new course…or, maybe you do, but you want to know that before it goes to the dean for review. A keyword search might get you a list, but we all know that jargon and specialized language in each academic discipline make identifying subject matter overlap difficult.

Courseleaf is developing a feature that, just like Neutron Enterprise, uses natural language search to find similar courses at your institution, both existing courses and those that are currently under review.

Of course, you could ask ChatGPT or Copilot to do that sort of task, but think about the points of failure with a chatbot. You would have to get it to ingest a huge amount of institutional data. It might confabulate or hallucinate some or all of its answers. Talking with it might be fun, but really, you just want a list of courses, not a little buddy telling you how insightful your questions are.

Just like AI note-takers and customer service bots have been appearing everywhere, this kind of advanced natural language search is coming to platforms and apps. Instead of Clippy-like companions and co-pilots, we will see more built-in natural language functions to existing products that retrieve the information you want and help you take action on it.

A chat+ interface to the Web

Adding natural language functions to your internet browser reimagines a tool we take for granted. Instead of a Google search box or a chatbot, you may soon find yourself using natural language functions that add layers to chat so that you can interact with the World Wide Web.

All the big AI companies are working on some version of this, but The Browser Company’s Dia seems to be a little ahead of the curve. This concept is a little hard to explain, so here is a video featuring students explaining how they use Dia, and here is blogger and professor of economics, Brad DeLong, describing his experience using it, and designer Jurgen Gravestein writing about the trend of AI powered browsers.

While there are reasons to be skeptical about English becoming the hottest new programming language, it does seem clear that using natural language to understand the data in your browser tabs is useful. Features like Courseleaf’s “similar courses” and redesigned browsers like Dia point toward a future where natural language functions within existing technology, rather than as an external co-intelligence you talk to.

Purpose-built to solve academic problems

This next example flips from data interface to data analysis tool. As I said earlier, AI is changing how we understand the relationship between language and mathematics. I’d love to talk about how the digital humanities is rising to these questions, but I promised to stay focused on the practical tools. So, let’s talk about a tool that harnesses a large language model, which, after all, is the application of mathematics to language, to understand and solve bureaucratic problems.

I managed Penn’s course evaluation system for many years. It always made me sad to think about all the information we gathered from student comments that never went beyond the instructor. Some department chairs might read through them, or maybe cherry-pick a few good comments for a promotion packet. But mostly all that information just sits there in a database or warehouse.

Think back to my example of a purpose-built language model searching through nuclear-related data. Explorance MLY is an AI model built for analyzing text-based information, a cousin to Neutron Enterprise, that can function in a similar way for course evaluations or any other database of natural languge feedback. It includes a classification mechanism that provides sentiment analysis and sorts comments into categories.

Here is Explorance’s CEO, Samer Saab, talking about MYL in the context of responsibility:

MYL offers practical value in its ability to identify information in the unstructured course evaluation data. Say you want to know what students had to say about a specific classroom or compare how students describe small seminars versus large lecture courses. Or, to bring up a pet project of mine, let’s say you wanted to read all comments specifically about active teaching methods, even those for courses that do not use such methods. To get even more practical, let’s say you wanted to screen potentially biased or harmful student comments before they are sent to brand-new teachers. MYL helps you do that.

Teaching as orchestration

We covered natural language beyond chat and purpose-built models. Now let’s talk about the last, and in many ways, the trickiest of the ways I see us moving beyond chatbots. What happens when we add one or more generative AI tools to academic or business processes, making those processes more complex? This involves orchestrating the work of the models themselves along with the work of the humans involved. Here is a simple example drawn from my teaching.

Last fall, I worked with the folks at the Penn Center for Learning Analytics to use a large language model tool called JeepyTA in my history class. I wrote about the experience here.

Given the lay offs and hiring freezes along with the fundamental inability of large language models to replace human teachers, I am always careful to describe JeepyTA as an instructional tool, not a teaching assistant. The frame of AI-powered digital twins or All Day TAs cedes too much to the tech oligarchs and cost-cutting bureaucrats who dream of replacing teachers with chatbots.

I use JeepyTA narrowly, to add a step to the process of peer review. Here is how I described it in a previous essay:

Rather than have JeepyTA emulate how I give writing feedback in order to substitute for feedback I would give, I conceived its role as providing an additional layer of feedback, a preliminary step each student takes before showing a draft to a peer or a member of the teaching team. We configured JeepyTA to give this feedback by post-training it on the assignment and examples of previous first-draft feedback.

At our first peer-review workshop, each student had a peer essay + JeepyTA’s feedback to review. This had the effect of helping students get past their hesitancy to criticize a classmate’s writing because they started with machine-generated feedback. My students and I believe the simulated feedback helped jump-start peer review. I will use the same technique for the course I am teaching this semester.

That course is called How AI is Changing Higher Education (and what we should do about it), and, given the subject, I wanted to give students some practical experience building AI tools. For that, they are using Boodle Box.

Orchestrating class assignments

The assignment involves what I call getting our hands dirty by working with AI. We are all familiar with the dilemma of spending hours crafting a syllabus and assignment guides only to have students not read them. This week, I asked my students to create what Henry Farrell calls a course syllabus that speaks as a way to have students engage with my course materials and experiment with generative AI tools.

Using BoodleBox as an AI toolkit, each student will develop a custom bot designed to answer questions about how to succeed in the seminar. They will experiment with using different AI models (the ability to switch out models is a feature of BoodleBox). Then, we'll conduct peer review and testing of the bots and compare the experience of talking to an AI bot with the old-fashioned method of reading the syllabus and course guides.

Note what I am doing here with JeepyTA and BoodleBox. Instead of an AI chatbot helping me design lesson plans or answer questions when I’m unavailable, I am managing various AI tools, orchestrating their use for specific purposes, while also directing the activities of my students. All this is to create opportunities to learn about the application of AI tools in higher education.

If you want more complex examples of orchestration in teaching, the place to go is large introductory STEM courses. As a participant in the talk pointed out, one of the best-known examples is Eric Mazur’s work developing tools like Perusall and using them for peer instruction in Physics. Increasingly, tools built from large AI models are being used for a variety of narrow instructional purposes.

I don’t have examples in the administration of higher education of complex orchestration. (If you know of any, please get in touch!). I’m grateful to the participant in my talk who asked me to explain what I mean by orchestration.

The best example I can come up with is Reuters, which has been adopting and adapting new information technologies since the telegraph. This Decoder interview with Paul Bascobert, describes how operations of a newsroom spread across the globe are being transformed by the incorporation of large language models and other types of AI. As you'll hear, this is far more than chatbots and copyediting.

If you want to get a sense of thinking about AI models working together, try out BoodleBox and explore the collaboration tools they have just added to their platform. You might do what my students are doing this week: build a custom AI tool to help accomplish a task, and then think about how to incorporate it into the processes of your course or your office.

Higher education should shape the market in AI tools

At the end of the the session, one of the participants talked about the value of using smaller, open source models instead of the latest and largest models. This is a view I share. So much attention is focused on the giant technology companies and their products that alternative models or companies rarely enter the conversation on campuses.

These giant companies have made giant bets on AI chatbots, and it is not clear whether those bets will pay off. Competition is fierce and growing users numbers is all important. How much will people pay for a chatbot subscription? What happens when the costs of producing large AI models decreases and competition increases? These are important questions…for AI companies.

Higher education has enough problems without borrowing those of Google and OpenAI. Educators should keep this in mind as the pressure to speed up progress on AI adoption continues. There are economic and ethical reasons to use small, open source models, and to buy products from companies who act responsibly, not just talk about “responsible AI.”

Educational institutions should refuse to work with companies that “move fast and break people.” To the extent that faculty and teachers have a voice in institutional decision-making, we should call out companies that quietly erode data privacy while giving their product away for free to students, and demand procurement processes take into account ways AI companies are commercializing their products.

Given their track record of dissembling and illegal behavior, why would any college or university do business with OpenAI? We should insist on better from companies looking to be our “AI partners.” There are hundreds of AI companies competing for our business. Let’s work with those that take care not to harm our students.

Share

AI Log, LLC. ©2025 All rights reserved.

I enjoy talking with people about how AI is changing higher education and what we should do about it. I’d love to come to your campus and talk with you and your colleagues. Click the button for more information.

𝐀𝐈 𝐋𝐨𝐠 speaks

Superbloom is largely concerned with developments in communication technology over the last twenty years, and, as I argued in part one, is strengthened by its engagement with earlier ideas, particularly those of the often overlooked Charles Horton Cooley, who coined the term social media in 1897.

That makes it rather unfair for me to pick the nits off Carr’s discussion of the exchange of ideas between John Dewey and Walter Lippmann in the 1920s, which Carr treats with far more understanding than most writers. Yet, he frames their ideas with the oft-repeated but too-simple story of Lippmann as a clear-eyed democratic realist and Dewey as a dreamer of democratic ideals. In fact, these two thinkers argued that technocracy isn’t just anti-democratic; it is impossible to implement. I’ve indulged in footnotes and links more than usual for readers who want to learn more.

The notion that intelligence is a personal endowment or personal attainment is the great conceit of the intellectual class, as that of the commercial class is that wealth is something which they personally have wrought and possess.

—John Dewey, The Public and Its Problems, 1927.

Dewey, Lippmann, and their Publics

In Part One of this essay, I said that among Superbloom’s many contributions to the current discourse about technology is its highlighting the social ideas of Charles Horton Cooley. Carr credits Cooley for anticipating the insights of later media studies scholars like Harold Innis and Marshall McLuhan, and points out that Cooley’s optimism about the power of technology to bring people together blinded him to the ways the social media of his time frustrated democratic political and social movements. Carr turns to the Lippmann-Dewey Debate to discuss two writers who did think carefully and critically about the problems social media creates for democracy.

Carr’s presentation of this set piece of American intellectual history rightly understands John Dewey as extending Walter Lippmann’s insights from Public Opinion (1922) and The Phantom Public (1925) about the challenges mass media present for democratic politics. The Public and Its Problems (1927) agrees with Lippmann's insights about the mass media’s social effects, and shows Dewey was just as troubled by what those effects mean for democracy. Yet, for the past three decades, philosophers and pundits have followed media scholar James Carey's framework of a debate, focusing on the one place where Dewey departs from Lippmann. In so doing, they squeeze Dewey into the role of optimistic advocate for democratic ideals, and portray Lippmann as a realist advocate for the exercise of technocratic power. Carr follows this unfortunate tendency, ending his chapter “The Democratic Fallacy” with this characterization: “Dewey told us what we want to hear. Lippmann told us what we need to hear.” 1

This gets it exactly backward.

Lippmann argues that because the democratic masses are not capable of the necessary wisdom and intelligence for self-government, elites should exercise political power, sometimes using anti-democratic means. Dewey agrees with Lippmann’s premise about the masses, but not his solution. In the spirit of Churchill’s quip about democracy being the worst form of government “except for all those others that have been tried,” Dewey believes that our only hope, inadequate as it is to the current moment, is to develop the wisdom and intelligence of the masses through democratic education. This can only happen if scientists and policymakers align their interests with the unwise and politically incapable masses, and engage with them to understand and develop the public interest.

The problem with Dewey’s answer is not that he tells us what we want to hear. Dewey is as skeptical as Lippmann about the prospects of democracy in an age of social media. The problem with Dewey is that he tells us readers of books—intellectuals, social scientists, and policy makers—what we don’t want to hear: that we are not to be trusted, that “rule by those intellectually qualified, by expert intellectuals” is a seductive impossibility.

Dewey’s critique of what he calls “this revival of the Platonic notion that philosophers should be kings” is far more than a moral objection or a cynical attack on intellectual self-regard. Cynics, Dewey writes, might call this technocratic rule

a revery entertained by the intellectual class in compensation for an impotence consequent upon the divorce of theory and practice, upon the remoteness of specialized science from the affairs of life: the gulf bridged not by the intellectuals but by inventors and engineers hired by the captains of industry.

Indeed, cynics say exactly that, then and now, pointing out that in their eagerness to earn status among their peers by doing narrow, statistically complex research about small questions, academic intellectuals have left the masses to the offerings of inventors and engineers who make pretty toys, and oligarchic captains who build great wealth commercializing these products. Today, the offerings of giant technology companies and their owners’ prognostications distract our attention from serious things and, worse, as Carr makes clear, they build public communication platforms that place conflict at the center of social life.2

But Dewey was no cynic, and this criticism of experts wielding power does not address the actual problem with intellectuals acting as shepherds to control what Lippmann calls “the trampling and the roar of a bewildered herd.” The real problem, as Dewey argues, is that technocratic solutions do not accomplish anything and may strengthen the position of oligarchs.

If the masses are as intellectually irredeemable as its premise implies, they at all events have both too many desires and too much power to permit rule by experts to obtain. The very ignorance, bias, frivolity, jealousy, instability, which are alleged to incapacitate them from share in political affairs, unfit them still more for passive submission to rule by intellectuals. Rule by an economic class may be disguised from the masses; rule by experts could not be covered up. It could be made to work only if the intellectuals became the willing tools of big economic interests. Otherwise they would have to ally themselves with the masses, and that implies, once more, a share in government by the latter.

Dewey’s Answer

This is Dewey’s answer: social science and policy making aligned with the interests of the public is democracy’s only possible future. Intellectuals who speak for or about the people or groups of people as abstractions or write “studies show” analyses that merely support their policy preferences are flailing. They risk irrelevance or, worse, unintentionally becoming the tools of the anti-democratic economic interests they think they are opposing.

“Rule by an economic class may be disguised from the masses; rule by experts could not be covered up” is Dewey’s one-sentence summary of where he differs from Lippmann. In the 1930s, Lippmann came to agree with him.3

Focusing on their books misses the fact that Dewey and Lippmann shared a critical view of technocratic social scientists. Lippmann was so dismayed by the arrogance and misleading claims of intelligence experts that he wrote a series of essays in 1922, published in The New Republic, attacking IQ tests just as they were being introduced to the public as a tool to cull the “mentally deficient” from the herd. Dewey joined Lippmann in this attack, and the two of them presented a united front against academic experts promoting policy frameworks based on Galtonian social science.

Lewis Terman, who put the “Stanford” in the Stanford-Binet Intelligence Scales, answered with an essay of his own, but came off so arrogant that colleagues urged him not to continue the exchange. The efforts of Lippmann, Dewey, and others did not stop the use of IQ testing by policymakers and courts, but they gave voice to opposing views and helped raise doubts that eventually blunted the reach of the eugenics movement. Lippmann’s exchange with Terman was an actual debate, one that, as Tom Arnold-Forster says, “reveals a broader range of ideas about the politics of expert authority in a modern democratic society.”4

The differences between Dewey and Lippmann, and there are many, are of emphasis. For Lippmann, liberalism is paramount. When democracy threatens individual freedom, he argues for taking anti-democratic measures. For Dewey, democracy is the priority, and, for him, the fact that the masses act unwisely changes nothing. He believes democracy is a way of life to be realized, not simply a means for making political decisions. Progress toward that way of life would create a society that respects individual liberty and reforms political institutions from the bottom up. Developing a democratic culture is a long-term project, and is by no means certain to succeed.

Creating a democratic culture was not a vague aspiration for Dewey. He does not say much about schooling in The Public and Its Problems because when it was published, Dewey was already famous for arguing that school is where democratic social change must begin. Those unsympathetic to or unaware of Dewey’s ideas lump him in with the worst aspects of progressive reform, missing the extent to which he was a critic of the technocratic social science and the romantic child-centered idealism that underpinned so much of the educational reform efforts during his long life.

Dewey never abandoned what the women teachers of the Chicago Laboratory School, his wife Alice among them, taught him in the 1890s. He turned these lessons into books about schooling in a democratic society that brought him to the public’s attention. As he moved from Chicago to New York to become one of the best-known public intellectuals in the world, he remained committed to the idea that the school, which he calls “an embryonic community,” is where democratic culture begins.

Dewey’s essay “Politics as Education,” a short piece published in The New Republic in 1924, was written as he began grappling with Lippmann’s analysis of problems of democratic government in an age of mass media.5 It is pessimistic about the state of politics and schooling, but Dewey concludes on a hopeful note, one that makes clear that his loyalties and faith are with teachers and not expert reformers. His view is that if our systems of schooling reorient around developing critical intelligence and a broad, participatory understanding of the scientific method, and if “teachers become sufficiently courageous and emancipated to insist that education means the creation of a discriminating mind,” then schooling could become the democratic means to an improved end.

When this happens schools will be the dangerous outposts of a humane civilization. But they will also begin to be supremely interesting
places. For it will then have come about that education and
politics are one and the same thing because politics will have to be in
fact what it now pretends to be, the intelligent management of
social affairs.

Don’t misread “intelligent management of social affairs” as technocratic systems of schooling. Dewey means embryonic communities embedded in local communities that cultivate the social intelligence of the public. His hope that schooling would soon create conditions for a growing democratic culture was fading by the time he read Lippmann’s Public Opinion, but he did not abandon public schools as the long-term means to achieving democratic ends.

Dewey is pessimistic because he saw more clearly than Cooley that the liquefaction of the social medium is a liquefaction of the social bonds that connect people to their local community. His diagnosis in the 1920s speaks to the ways social media has “invaded and largely destroyed” the “self-centered locality.” Unlike many social scientists of his day and ours, Dewey is uncertain about cause and effect when it comes to how this process changes our social experience.

No one knows how much of the frothy excitement of life, of mania for motion, of fretful discontent, of need for artificial stimulation, is the expression of frantic search for something to fill the void caused by the loosening of the bonds which hold persons together in immediate community of experience.

Does social media draw our attention because we have lost connections to others? Or does social media cause the dissolution? Social science is good at measuring the correlations we notice, but assumptions about cause turns experts into augurs. The incentives of the attention economy encourage prognostication and confident prediction. Headlines and “studies show” best-sellers confirm the worst, while the complexities go unexamined. When each side of every debate has the science on its side, when everyone’s pet theory to improve schooling is evidence-based, then it all becomes something of a joke.

No silver linings

The case for eliminating connected devices at school is strong enough based on common sense alone. More charts will not change the calculus for policymakers or make bans any easier to enforce. And even if schools are free of personal devices, school systems spend a huge portion of their shrinking budgets on equipping every classroom with the latest digital technology and new software, whether teachers want them or not.

Attempting to make schools the dangerous and interesting outposts of a democratic culture feels inadequate to the size and scale of today’s pressing public problems. Yet, if we take democratic education to mean more than efficiently managed mass schooling, Dewey’s answer is the best we have. That answer requires redirecting “the forces which have effected uniform standardization, mobility, and remote invisible relationships” so that they “flow back into local life, keeping it flexible, preventing the stagnancy which has attended stability in the past, and furnishing it with the elements of a variegated and many-hued experience.” Then, Dewey says, “organization may cease to be taken as an end in itself.”6

The recent pandemic hit big school systems hard. Families and students are choosing alternatives to big systems, and voters are supporting Empowerment Scholarship Accounts (ESAs) and other unregulated versions of school voucher programs. Although we will no longer have reliable national numbers about school enrollment, it seems clear that this shift in public education funding and declining public school attendance will be at the center of politics in the coming decade. Working to make microschools the incremental, small-scale means for democratic social change runs smack into an ascendant political movement directing public funding away from big systems of schooling. That movement, shaped by a coordinated network of billionaires and right-wing operatives, has turned the idea of microschools like the one Alice Dewey managed in Chicago into a hot potato on the left as some defend the ideals of a common system of schooling and others seek to build small sanctuaries of learning for students who are too often its victims.7

The reading and math wars, so beloved of experts and pundits, seem almost quaint today as education researchers are laid off and national data programs are defunded. The destruction of long-standing programs to gather and analyze information about schooling will hobble attempts to improve our systems of education at the national level. But perhaps that is not the tragedy it seems. The legacies of No Child Left Behind and the Common Core provide little support for another wave of national reform. Perhaps, the result of the current chaos will be to invigorate the laboratories of democracy.

There is no silver lining in the tornado ripping through our federal agencies and universities, but whatever emerges afterward should aim for something better when it comes to supporting public schools. Let’s reduce the incentives for social scientists to play the game of applying complex statistical methods to small datasets and stop paying consultants and technology companies to optimize what historian David Tyack called “the one best system.” Instead, experts and oligarchs should turn their attention toward furnishing teachers “with the elements of a variegated and many-hued experience” and help strengthen our resolve against over-capitalized technology companies desperately trying to make Anna Julia Cooper’s dark joke come true:

I wonder that a robot has not been invented to make the assignments, give the objective tests, mark the scores and—chloroform all teachers who dared bring original thought to the specific problems and needs of their pupils.

Public school teachers bring original thought to their classrooms every day, and yet they spend their time and energy on being a counter-friction to machinery imposed upon their work by well-meaning but narrow-minded experts and managers with the best of intentions. The promise that creating new systems out of large AI models will free up teachers’ time, letting machines perform meaningless bureaucratic work while teachers teach, is the same mirage offered by those selling previous versions of personalized teaching machines. More elaborate machinery will not solve the problems of public education. Given what’s been happening lately, the prospects for resisting are grim.

One thing we can count on is that the future will be stranger than we can imagine. Here is a short science-fiction story about a utopia built by kindergarten teachers after society collapses.

Share

The rock and the margins

Thinking about the long history of social media helps correct the tendency to imagine that digital technology is the sole cause of the public problems facing us today. The liquefying effects of the railroad, telegraph, telephone, and cheap mechanical printing were as devastating to those alive in 1900 as the transformations wrought by digital networked devices that so occupy our attention today. And now, as we enter a new phase of the digital age with transformer-based large AI models, we feel as though our social selves are drowning in the flood, that we live at the moment of maximum liquefaction.

Superbloom ends by finding hope in a rock, specifically, Samuel Johnson’s famous kick of a rock refuting George Berkeley’s idealism. Carr calls that rock “an emblem of stability, solidity, resistance”; it reminds us of the material world as experienced by our human bodies. He urges us to think about “what separates the human animal from the AI machine.” Yet, when we distinguish the machine from the human, we draw not a line of separation, but a two-way arrow. Every communication technology is an expression of human intelligence. Dewey put it this way: “The level of action fixed by embodied intelligence is always the important thing.” Dewey means embodied both in the material self and in society as a whole.

The moral perfectionism offered by Thoreau and Carr and the reconstruction of the social self offered by Cooley and Dewey offer a basis for restoring solid ground. In A Common Faith (1934), Dewey argued that democracy is not just material and social; it offers all the “elements for a religious faith that shall not be confined to sect, class, or race,” and that our task is to “make it explicit and militant.” Carr hints at this potential when he tentatively offers the word salvation in his conclusion.

I take his use of salvation in the spirit of William James, who understood it as social and material, as well as spiritual: “You may interpret the word 'salvation' in any way you like, and make it as diffuse and distributive, or as climacteric and integral a phenomenon as you please.” Climacteric means a major turning point or critical stage, which describes where we are today when it comes to social media—especially the latest social media, large, transformer-based, AI models—and their relation to schooling.

The feeling that we are living in climacteric times has existed as long as social media have existed, which is to say, for all of written history, and very likely before. Carr and Dewey point us in a similar direction to meet the challenges of our moving present. In Carr’s rendering, salvation lies in individuals making “personal, willful acts of excommunication” that create movements “not outside of society but at society’s margin, not beyond the reach of the informational flow but beyond the reach of its liquefying force.” The school could be just such a place, where small groups work together at the margins to grow the embryos of a democratic future.

Recall Cooley’s metaphor about the “thick, inelastic liquid” transmitting large waves, which bear “countless wavelets and ripples.” Surfing those wavelets and ripples on social media makes them seem larger and more powerful than they are. We can always stand on solid ground, joining with what Cooley called our primary group, and dive together into the slow, molasses waves. Stand again, look into the eyes of our companions, listen to what they say, and answer in our own voices.

Dewey concludes The Public and Its Problems with a meditation on Emerson’s image from Self-Reliance: “We lie in the lap of immense intelligence, which makes us receivers of its truth and organ of its activity.” Dewey argues that each of us has access to this persistent and transformative power.

There is no limit to the liberal expansion and confirmation of limited personal intellectual endowment which may proceed from the flow of social intelligence when that circulates by word of mouth from one to another in the communications of the local community. That and that only gives reality to public opinion.

As delightful and addictive as small screens and AI chatbots appear to be at the moment, we can put them aside and use the oldest communication tool at our disposal, the original technology of human connection, the spoken word. School is the obvious place to organize our immense social intelligence. Many schools and teachers already start each day by severing connections to the distracting digital networks so that students can develop their intelligence in communion with others. This work is slow, difficult, and it loses value as it gets scaled and standardized.

I believe the moral perfectionism Thoreau and Carr offer is essential to democratic culture, but that Dewey shows us how to bring that culture into being. He urges us to look beyond individual self-perfection and join our work with those immediately around us. Our salvation, understood as a direction of travel toward a better culture, is to be found in the explicit and militant work of creating democracy through teaching and learning together. That work would be easier if experts and oligarchs joined teachers and students in the attempt to make schools supremely interesting places.

AI Log, LLC. ©2025 All rights reserved.

The railway, telegraph, telephone and cheap printing press have changed all that. Rapid transportation and communication have compelled men to live as members of an extensive and mainly unseen society. The self-centered locality has been invaded and largely destroyed.

—John Dewey, “Education as Politics” in The New Republic (1922)

Henry David Thoreau kept a journal—he might have called it a log—from the time he graduated college in 1837 until a few months before his death in 1862. We usually think of Thoreau writing about the natural beauty he experienced at Walden Pond or while traveling on the Merrimack River. Or, picture him sitting in jail protesting slavery and an unjust war, writing sentences like “Let your life be a counter-friction to stop the machine.” Yet Thoreau was a keen observer of the technologies changing his world. The new modes of manufacturing, the railroad, and the telegraph were as essential to his writing project as his observations about nature and social justice.

On January 2, 1851, inspired by the new year, he began writing in a new notebook. Most of his entry on the first, fresh pages describes in detail what he saw while visiting the Gingham mills, one of the many new manufacturing facilities appearing around Boston. Thoreau could be enthusiastic about what he saw in new technology. Later that year, he wrote in his journal:

As I went under the new telegraph wire, I heard it vibrating like a harp high overhead. It was as the sound of a far-off glorious life, a supernal life, which came down to us, and vibrated the lattice-work of this life of ours.

He returned to this theme a few days later, writing that listening to the vibrations “reminded me, I say, with a certain pathetic moderation, of what finer and deeper stirrings I was susceptible, which grandly set all argument and dispute aside, a triumphant though transient exhibition of the truth.”

The next week, he abandoned any sense of moderation, to describe putting his ear to the wooden posts on which the wires were strung, and hearing music, “as if every fibre was effected and being seasoned or timed, rearranged according to a new and more harmonious law. Every swell and change or inflection of tone pervaded and seemed to proceed from the wood, the divine tree or wood, as if its very substance was transmuted.”

He concludes this passage by proposing a modern Muse to honor the invention, and proclaims the telegraph, “is this magic medium of communication for mankind!”

Superbloom does not mention Thoreau, though these reflections on the telegraph illustrate the main theme of Nicolas Carr’s new book: our longstanding cultural habit of greeting each new communication technology with enthusiasm, and then slowly realizing the problems it creates. As he puts it, we don’t see the ways that “technologies of connection tear us apart” until it is too late to do anything about them.

Superbloom explains how this cycle has played out over the past 180 years. From Thoreau’s “magic medium” to Marconi's boast that radio makes “war impossible” to technology writer Orrin Dunlap’s prediction in 1932 that television would “usher in a new era of friendly intercourse between the nations of the earth,” Carr describes how each new communication technology appears as the light at the end of the tunnel of human misunderstanding only to discover Harold Innis’s disappointing truth: ”Enormous improvements in communication have made understanding more difficult.” Carr’s point is that it is not just that our new tools fail to live up to their promises; they roll right over us, destroying social structures built with previous technology and eroding modes of communal living.

Thoreau was quick to see the dark side of the technologies of connection of his day. By the time he wrote Walden, two years after his reveries about the telegraph, he had turned skeptical. He calls the railroad and the telegraph “pretty toys, which distract our attention from serious things. They are but improved means to an unimproved end.” Among Thoreau’s most memorable passages about technology is one about the human cost of building networks of iron rails held together by wooden ties, also known as sleepers.

We do not ride on the railroad; it rides upon us. Did you ever think what those sleepers are that underlie the railroad? Each one is a man, an Irishman, or a Yankee man. The rails are laid on them, and they are covered with sand, and the cars run smoothly over them. They are sound sleepers, I assure you.

For the past decade, we have been asking similar questions about the digital networks that promised so much when we first encountered them. We have come to understand that we do not communicate through the internet; the internet communicates through us. As Thoreau says, we have become the “tools of our tools,” and the internet-connected platforms that seemed to promise freedom and democracy in the early days of Google and Facebook now make us feel oppressed and trapped. Carr describes how these same companies, now joined by OpenAI and Anthropic, are building a technology that runs smoothly over our minds: “Live in a simulation long enough, and you begin to think and talk like a chatbot. Your thoughts and words become the outputs of a prediction algorithm.”

As we awaken to the social and psychological problems associated with using these newest pretty toys, the question is, what can we do to change things for the better? Can we shape the tools that now shape us? Carr believes we lost the chance: “Now, it’s too late to rethink the system. It has burrowed its way too deeply into society and the social mind.” His hope, such as it is, is not to change the system. Too late for that, he says, “But maybe it’s not too late to change ourselves.” As Carr sees it, living our lives as a counter-friction to these machines is the only remaining option.

In looking to the individual self as the means for social change, Carr is writing in the tradition of “moral perfectionism,” which philosopher Stanley Cavell understands as originating with Emerson and Thoreau, especially in Emerson’s democratic faith that each of us can shape the world through the attempt to perfect “our unattained but attainable self.” Walt Whitman is the great poet of such striving, and James Baldwin, perhaps, the great example in the twentieth century.1

Not trusting social or political movements, these writers look to the self and self-expression to reform an immoral society. They hold democracy sacred and believe that in each individual, in Whitman’s phrase, “Nestles the seed Perfection.” After all, what stops me—what stops any and all of us— from attaining a self that closes my social media accounts, shops local, votes in every local election, and uses only open-source computing products and software?2

Carr’s neglect of the writers who inaugurated the tradition in which he writes is easily forgiven because, by reminding us of Charles Horton Cooley, he does something more valuable than revisiting Thoreau, who, after all, still regularly appears in our cultural discourse as the leading techno-skeptic of American letters.3

Charles Horton Cooley, the original academic techno-optimist

Superbloom rescues Charles Horton Cooley from obscurity. A faculty member at the University of Michigan starting in 1894 until his death in 1929, Cooley was among the generation of writers and researchers who created social science as we think of it today, organized into disciplines and housed in universities. His major works range across those disciplines, covering communication, organizational dynamics, transportation economics, and social psychology. His most influential concepts were the looking-glass self and primary groups, which described processes of identity formation just as the boundaries between psychology and sociology were being marked.

Like most of the group associated with the Chicago School (think symbolic interactions in the 1890s, not neoliberal economics in the 1950s), he kept his distance from the pseudo-scientific racism that permeated academic biological and social science in those days. One of his first academic papers was a critique of Francis Galton’s Hereditary Genius (1892). Yet, again like many of his colleagues, he never involved himself in politics and ignored the path-breaking sociological work of W.E.B. Du Bois in The Philadelphia Negro.4

As Carr demonstrates, Cooley was among the first academic theorists of communication technologies. He coined the term social media in 1897 in “The Process of Social Change,” an essay published in Political Science Quarterly. In it, Cooley sketches ideas that later academics like Harold Innis and Marshall McLuhan made famous, applying the now ubiquitous term quite broadly, and understanding social media’s origins in the ancient combination of writing and maritime navigation that spread the words of the earliest social media influencers, including those we still know today: Buddha, Confucius, Jesus, and Socrates.5

Understanding media is, for Cooley, as for Innis and McLuhan, a matter of thinking about the processes of social and technological change across long periods of history. Just as McLuhan brought broad historical context to understand the new media of the mid-twentieth century, Cooley addresses those of the nineteenth: “the new means of communication, — fast mails, telegraphs, telephones, photography and the marvels of the daily newspaper, —all tending to hasten and diversify the flow of thought and feeling and to multiply the possibilities of social relation.” He explains the increased speed of the means of communication and the attendant processes of social change with an extended analogy.

A thick, inelastic liquid, like tar or molasses, will transmit only comparatively large waves; but in water the large waves bear upon their surface countless wavelets and ripples of all sizes and directions.

The new means of communication produce an effect that “may be described as a more perfect liquefaction of the social medium.” From the perspective of users of today’s technology, information in the past moved slowly across space and time, making change visible in the archeological record and then in written historical records. What appears today as the slow movement of information via sailing ships and the printing press accelerates rapidly, starting in the decades Thoreau wrote his journal, as new networks were built from telegraphs and steam engines. The combination of steam power and the telegraph sped information around the globe, spreading cultural, political, and scientific change.

If Cooley worried about the effects of these changes, say in the truthiness apparent in the journalism being invented by writers and editors working for Joseph Pulitzer and William Randolph Hearst, he does not mention it. In his blindness to what Carr calls “the more ominous implications of his argument,” Cooley is not so different from those who wrote about how digital social media would accelerate the spread of democracy or those who see the latest AI models ushering in an age where humans will have their material wants provided by machines of loving grace.

In Cooley’s time, as now, social media operates through systems managed by webs of wealth and influence that work to preserve the power of elites. Despite the persistence of anti-democratic politics in every democratic political system, there is something about the arrival of each novel means of communication that opens our minds to the hope that, as Thoreau put it, we will “grandly set all argument and dispute aside” to experience “a triumphant though transient exhibition of the truth.” Carr positions Cooley and Mark Zuckerberg as bookends of the twentieth century’s optimism about each new mechanism for communication. If the First World War and the rise of totalitarian regimes ended the hope that Cooley’s new media were the harbingers of peace, then the story of Facebook’s fall from a company commercializing a social technology that promises human connection to one that delivers a product that tears us apart brings us to the end of another cycle.

Realizations about what the commercial internet had become were just sinking in when the pandemic severely limited our means of socializing. In Carr’s telling, this was the moment when “the social and the real have parted ways. No longer tied to particular locations or times of day, social situations and social groups now exist everywhere all at once.” And, it’s true that the liquefying force of digital communication technology was intensely felt as we were confined and isolated. But, when exactly did the social and real become estranged? When Zuckerberg built a tool “to make the world more open and connected”? When radio and film gave us two new magic mediums of communication? When Thoreau heard the magical telegraph “vibrating like a harp high overhead”? When Socrates told Phaedrus that writing “will introduce forgetfulness into the soul of those who learn it”? Does it all go back to those damn kids painting in the caves?6

Social media is bad for you, always has been

In Thoreau’s Axe: Distraction and Discipline in American Culture (2023), Caleb Smith offers “a genealogy of distraction and the disciplines of attention, going back to Thoreau’s era,” putting together an impressive range of documents that explore the nineteenth-century sources of our predicament. He points to the failure of individual moral perfectionism to address the problems.

Even after we understand that distraction’s real causes are in the large-scale economic systems and technologies that shape our world, we keep trying to solve the problem with personal, moral remedies.

Like Smith, I find meaning in nineteenth-century writing as “relics of a troubled past and as crafted objects of sustaining beauty.” But the moral perfectionism of Thoreau and Carr offers little evidence that being the change you want to see in the world is a recipe for solving problems created or exacerbated by social media. Smith’s tracing of anxieties past provides context for the social scientific worries of the present, especially the writing of Jonathan Haidt and Jean Twenge.

Carr brings an even hand to their attention-grabbing work, pointing out narrowly focused social science methods “obfuscate as well as illuminate,” and that “sometimes you have to lift your eyes and take a wider view.” That view offers plenty of evidence that adolescents feel more depressed these days. But the reasons for this are unlikely to be singular. Why are adolescents increasingly likely to stay home and on their screens, instead of taking part in traditional activities like gathering in church on Sundays, singing together, and celebrating the harvest each fall listening to popular music together, consuming alcohol and other recreational drugs, and having sex with their friends? Pinning the problems on a particular technology ignores a complex set of correlated and confounded factors at play, and the long history of anxiety about the cultural habits of the young.

As a fan of free-range parenting who worries about constant supervision of young people, I’m partial to studies that suggest we may be seeing ill effects from keeping kids indoors, permanently cancelling recess, and eliminating old patterns of free play in the woods, in the streets, and at the malls. Social media, especially the nightly TV news and newspaper headlines, digital news feeds and Amber alerts, play a part in these changes, producing a sense that stranger danger is an ever-increasing threat and creating parenting habits that would seem bizarre to earlier generations.

Keeping kids at home and always in sight, to say nothing of turning schools into contests centered on high-stakes, standardized tests, is a recent shift in behavioral norms that helps explain why escaping into screen time is so alluring. It also frames data suggesting an increase in exercise and a reduction in alcohol use, especially among young people, in an interesting light. Maybe the kids are finding their own path back to alright?

I am convinced by Carr and others that young people are choosing to live in what he calls a “world without world.” I am not convinced that these simulated worlds are all that different from the worlds created by adolescents using earlier social media. Is crafting a self on a YouTube channel so different from writing a journal? Are the teens disappearing into their screens to play video games and watch make-up routines so different from a young person in the 1850s writing in a diary about reading a novel by a Brontë sister or a teenager in the 1980s listening to Prince on a Walkman and starting a zine?

What makes Cooley such an interesting writer for this moment is that he and his colleagues constructed a theory for understanding self-formation through social groups and shared social identities. Self-fashioning as a social experience was described by writers ancient and modern, but it was first articulated in terms of the new, post-Darwinian social science in the late nineteenth century, most notably by William James in Principles of Psychology (1890).7

The social self

Carr writes about this idea, placing Cooley’s looking-glass self alongside similar ideas of George Herbert Mead and Erving Goffman as extending James’s description of the social sources of the “I” and “me.” Instead of exploring this neglected concept as a basis for thinking about social change in the digital age, Carr ends the discussion by pointing out that Goffman’s conception of the self “was defined by in-person interactions in actual places—schools, offices, factories, homes, bars, churches, shops.” Carr then turns to Joel Meyerson’s No Sense of Place (1986) and its pessimism about the social effects of electronic mass media and online social spaces.

For Carr, the looking-glass self and other notions of the fundamentally social nature of human cognition have been abandoned for “the mirrorball self” of digital culture, “a whirl of fragmented reflections from a myriad of overlapping sources.” Social media have disappeared the ground beneath our collective feet, by “removing social interactions from the constraints of space and time and placing them into a frictionless setting of instaneity and simultaneity,” leaving us in an unreal social world that “vibrates chaotically to the otherworldly rhythms of algorithmic calculation.” This is insightful and impressive writing, but it ignores the ways that the medium’s message has always provoked a sense of disorientation among older people, and that social identity is always formed out of the materials and tools at hand.

It is true that digital social media pull us toward a disembodied and dislocated sense of self, but then so do many religious and cultural practices. In introducing his discussion of the self in Principles, James describes the way people are often “ready to disown their very bodies and to regard them as mere vestures, or even as prisons of clay from which they should some day be glad to escape.” He introduces the social self, along with the material self and the spiritual self, as a conceptual framework for making sense of the formation of individual identities. The social self gets all the attention, and is of obvious value in thinking about communication technologies, but our bodies and our relation to what James called “the higher universe” are merely forgotten in the moment, not lost completely, when our attention is subsumed by media.

I find Carr’s case for concern more convincing when he turns his attention to Silicon Valley’s “idealists with means,” who are looking to “give us a utopia of their own device, one that furthers their ethical and political as well as their commercial interests.” Carr has Marc Andreessen and Zuckerberg stand for the wider and wilder range of theories about this utopia, where we all become “virtual epicureans.” But we should not grant today’s postmodern feudalists too much originality. This is not our first crop of oligarchs with weird social ideas getting bent toward the right by their workers’ protests. When Henry Ford first began to engineer the automobilification of North America, he was optimistic about labor relations, but turned violently anti-labor and publicly anti-Semitic when it was apparent that workers had their own ideas about fair wages and working conditions. Elon Musk’s purchase of Twitter echoes Ford’s purchase of his hometown newspaper, The Dearborn Independent, which became a vehicle for spreading anti-Jewish and anti-Labor conspiracies. At least so far, giant tech companies have not had their Pinkertons open fire or physically beat protesting workers.8

Despite the undeniable pull of screens and the immensity of the capital being deployed to capture our eyeballs, Carr too readily accepts the notion that we are lost to dopamine hits and digital toys. The recent habit of spending too much time on digital devices is like those charts beloved by tech investors, with lines that go up. Trends reverse. Habits change. Pretty screens and AI chatbots get boring. Attention may be captured, but our bodies and spirits persist, offering ground for a collective counter-friction to all the machinery Silicon Valley is desperate to sell us.

In Democracy and Education (1915), John Dewey writes, “The moving present includes the past on condition that it uses the past to direct its own movements.” People in the early decades of the last century faced a global rise in authoritarianism and a thrilling sense of possibility in new technologies, both sped along by waves traveling through the liquefying force of new social media. The political response to the pandemic of 1918, the Great Depression, and global wars did not solve our social problems, but it did create social safety nets that improved people’s material lives and civil rights movements that spoke to their spiritual longings.

Superbloom reminds us of the past but discounts its potential to direct us in the moving present. Dewey's The Public and Its Problems plays a role in Carr’s account of the past, but like so many who write about John Dewey’s exchange of ideas with Walter Lippmann in the 1920s, Carr misses the anti-technocratic projects Lippmann and Dewey collaborated on during this period and their shared belief that liberal democracy is the least worst answer to the problems of political organization in a social world shaped by globalizing capital and new social media.

Here is part two of this essay.

Share

Support 𝐀𝐈 𝐋𝐨𝐠

The first edition of AI Log went live on August 25, 2023, with a LinkedIn post titled Five things you might have missed in AI news this summer. In those earlier days, I was trying to find a middle ground between the unrelenting hype of Silicon Valley and the unrelenting freakout among educators about ChatGPT. Two years later, those early posts read embarrassingly breezy. Eventually, I wrote my way to a more critical perspective on educational chatbots and found a more distinctive voice writing book reviews and essays about my experiments with LLMs in the classroom. I gave up hot takes and round-ups, though I still blow my top about the headlines on occasion.

My interest in writing long-form led to a change in platforms from LinkedIn to Substack, which, for now anyway, is ad-free. I also stopped using AI image generators. Behind those visible changes to AI Log was a shift in focus to longer-term social contexts for technological change, as I came to believe that the problems we associate with digital social media and AI models have roots going back hundreds and even thousands of years.

To celebrate this publication’s second anniversary, here are five things I discovered through writing AI Log.

Anthropomorphizing AI tools is a bad idea

Pretending chatbots are people has always felt problematic to me, which is why Ethan Mollick is one of my favorite writers about AI and education. He is the grand champion of anthropomorphizing AI models, the leading evangelist for the benefits of treating AI models as if they were human assistants and tutors. My opposition to thinking of AI as people was sharpened on the rock of Mollick’s vision of co-intelligence.

Recent stories about obsessive behavior have more people worried, but the Eliza Effect also interferes with understanding how transformer-based computational models actually work, and it blinds us to a wide range of ways we might put these computational algorithms that confabulate language to good purpose.

Large AI models are educational technology

If Alison Gopnik and Henry Farrell are right that Large AI models are a cultural technology, then it follows that they are an educational technology. The question then becomes, what educational value do these tools offer? Instead of answering that question, mostly we talk about their social harms (there are plenty!) or speculate about what they might become (science fiction is fun!).

Understanding how large AI models function culturally and educationally is a path to moving beyond endless cycles of skepticism and speculation to think about how we might use these technologies to change educational institutions for the better. That path includes regulations to prevent companies from moving fast and breaking people and to break up monopolies, but it also means experimenting with generative AI models to try to accomplish socially beneficial goals. Here is a list of writers exploring AI as a cultural and social technology.

Large AI models are normal technology

I use Arvind Narayanan and Sayash Kapoor’s framework in my public talks with educators and technologists because it focuses attention on where we are now with transformer-based AI models as the social process of innovation and diffusion plays out. Think about the history of major inventions like the railroad, telegraphy, electric generators, internal combustion engines, and the networked personal computer. It takes decades to learn what economic and social value a new technology offers—and understand the social problems each creates— because it takes a whole lot of people working together across many types of organizations to discover how new tools can be adapted and put to use. And then, it takes even more time to encourage people to adopt the tools.

Investors in technology companies desperately want this process to move faster than it does, which places pressure on the people who manage technology companies. But that’s their problem. Higher education has enough problems without borrowing those of Microsoft and OpenAI.

Spending millions on an enterprise contract with OpenAI or Anthropic feels like a bold move for a president or board to make. Less bold, and just as stupid, is paying Microsoft money for Copilot when there is no evidence that it adds value.

For all the seeming urgency of AI job training and AI literacy, it is still early days. We have a lot to learn about how AI will change the economy, organizations, and educational institutions, but treating AI as a normal technology will help us make better decisions.

Small is the next big thing

I try to avoid prognostication, but it seems pretty clear that the hype about the largest and latest AI models is fading. I wrote about what happens after the AI bubble earlier this summer. What will replace breathless coverage of how the latest releases perform on benchmarks or outrage over the latest disastrous attempt by some tech company to speed the commercialization of a product they don’t understand?

I hope it will be stories about smaller, open models solving medium-sized problems for organizations of all sizes. If generative AI is indeed revolutionary, I suspect it will come to feel boring, the way double-entry bookkeeping, the filing cabinet, and digital spreadsheets are boring.

Nineteenth-century ideas explain twenty-first-century technology

In 2024, I wrote several essays about how William James helps make sense of AI. I intended to follow up with similar pieces about two of his lesser-known contemporaries, Charles Sanders Peirce and Anna Julia Cooper, but I am a slow reader and even slower writer, especially when it comes to the history of ideas. Peirce makes a brief appearance in On Confabulation, which is my favorite essay on AI Log so far.

I quote Cooper in this review essay of Nicholas Carr’s Superbloom, which is largely about how the ideas of Charles Cooley and John Dewey make sense of today’s digital social media. I expect all five of these thinkers will figure in Phase 2.

Subscribe now

Phase 2 of AI Log

I wrote about Lizzie Douglas’s theory of cultural work in a market/attention economy in my review of John Warner’s More Than Words. You can hear the original formulation of her ideas here: I’m selling my pork chops, but I’m giving my gravy away.

AI Log is all gravy, but Phase 2 involves cooking up a pork chop with the working title of Solving for AI: Technology and the Future of Higher Education. I do not expect a book to get me anywhere near Phase 3, but there are other reasons to write, and ways to earn a living wage while writing.

I don’t turn on paid subscriptions to AI Log because I prefer to ride for free on this platform. That said, I could use your help in letting people know about that delicious AI Log gravy.

AI Log, LLC. ©2025 All rights reserved.

Share

Better beginnings

I have a tendency to go on too long when I talk to students about the ideas that inform my teaching. I was winding up to do exactly that on the first day of class with a mini-lecture on how Talks to Teachers on Psychology by William James is the foundation for my structured, active in-class approach to teaching, when Shakespeare rescued me. Not William himself, rather I heard Peter Gould, the director of Get Thee to the Funnery, introduce the work he and his colleagues do to produce a Shakespeare play—this year it was Macbeth—with a group of teenage actors for two weeks each summer.1

Peter describes the program’s educational framework as working our hearts, minds, bodies, and voices together to encourage individual growth in the service of a collaborative production. This captures the essence of James’s ideas about embodied cognition and human experience that I hope my students and I can use to think and feel our way toward a better understanding of how AI is changing higher education.

That framework will start us off by clearly distinguishing between human cognition as a complex stream of emotional, physical, intellectual, and performative experience and computational neural networks producing outputs from fancy math applied to large datasets of words and images. Heart, mind, body, and voice are an elegant sufficiency to start our work, defining simply that most remarkable natural phenomenon, human intelligence. From there, we will begin to understand these most recent versions of artificial intelligence and what they mean for institutions of higher learning.

On the first day, we will discuss my radical proposal to limit the use of digital technology during class time, while having no limits outside of class time. My hope is that students will accept this sharp division, and that we can use it to create critical distance to think together in class about our educational use of AI out of class. I was inspired to try this by my students last fall, who enthusiastically endorsed the idea of creating digital-free class time.

A Brief Tour of “How AI Is Changing Higher Education”

I love reading other teachers’ lesson plans and syllabi, but my course material pales in comparison to beautifully crafted intellectual structures like Henry Farrell’s A Political Economy of AI: a Syllabus, Jane Rosenzweig’s To What Problem is ChatGPT the Solution?, and Eryk Salvaggio’s Critical Topics: AI Images. Reading them is the intellectual equivalent of watching MTV’s Cribs. My class is more Flip or Flop.

By design, my courses are open-ended and risk disaster each time we meet. There is little script and no score. Class meetings are jam sessions. Think backyard hootenanny in Appalachia or a late-night party after the clubs close in New Orleans. Everything depends on people showing up ready to play. Ambitious failure is encouraged, the stakes are low, and when it works, the payoff is a feeling of communal joy in individual and group performance.

“First we read and then we write” is the theoretical structure for the first eight weeks of the course. We will read More Than Words by John Warner and AI Snake Oil by Arvind Narayanan and Sayash Kapoor. I am asking students to buy physical copies so that we can do good, old-fashioned reading of the text together in class, books open to the same page. The in-class activities and writing assignments will be grounded in collective engagement with the words on the page. We will do a lot of group writing, with sentences and paragraphs going up on the whiteboard for discussion.

The first major task is inspired by John Warner’s advice to find your guides. I’ll ask each student to identify a writer, one who has something interesting or useful to say about AI and higher education. They choose an essay by that writer to be read by the class, they write a short essay about what makes their chosen writer interesting and useful, and they deliver a brief, informal talk in class about the value of reading that writer. To help them get started, I give them a list of some of my guides, which includes more than a few subscribers to AI Log.

• Here is A Guide to Finding Your Guide.

The class is built on the experience of writing six short essays. We will experiment with using JeepyTA, an LLM tool, to provide simulated feedback in the drafting process. As I did with my fall course, we will include the tool’s feedback as an additional layer of their peer review and discuss what, if any, educational value it brings. This is an experimental exploration; it does not replace human effort. Writing feedback itself will come in the traditional form of in-class peer review and comments from me.

• Here are the short essay writing assignments as they will appear on the Canvas course site.

• Here is my guide to writing the essays.

The class culminates in a student-led workshop where I sit back and the students get to work educating themselves and their peers (and me!). They will spend the last half of the course working in groups to develop an educational program designed to help students understand how AI is changing higher education, and what we should do about it. If all goes well, the first Monday in December, we’ll take over the wonderful new Research Data and Digital Scholarship Exchange space in Van Pelt Library to run workshops for whoever shows up.

First impressions

Later this week, I will send everyone who registers for the class a welcome letter, a preliminary syllabus, and a link to an introductory video. As I say in the letter, there are some weird things about the course, and it may be better for students who feel anxious about experimental and social approaches to learning to find more traditional options.

Welcome letter & preliminary syllabus

Welcome video (link live on August 20)

Here is the answer to the inevitable question: How does grading work in this class?

AI Log, LLC. ©2025 All rights reserved.

Share

Feel free to share any of the teaching materials I publish here with friends and colleagues.