The LA Unified School District's chatbot is a disaster and AllHere is close to bankrupt. What's next?
The Gartner Hype Cycle is famous enough to have its own Wikipedia entry, which helpfully explains that its purpose is βto represent the maturity, adoption, and social application of specific technologies.β As you might imagine, those who follow ed-tech like to keep an eye on where generative AI is on the chart. In August, Gartner pegged generative AI at the top, what they call the Peak of Inflated Expectations. They updated its position two weeks ago, showing that generative AI starting to slide into the βTrough of Disillusionment.β The colorful terminology is the best reason to use Gartnerβs visualization of this historical process.
A few days ago, generative AI slipped further down the trough, at least on the version of the chart inside my head. AllHere, the βaward-winning developer of AI solutions, trusted by over 9,100 schools nationwide,β announced its CEO is out, and the majority of its staff are furloughed until further notice. I wrote about AllHere without naming it in this post from March when I questioned the mostly uncritical news coverage of Ed, the Los Angeles Unified School Districtβs new chatbot. I suggested that based on what we could see of Ed, it felt like "the beginning of a Clippy-level disaster," referencing the Microsoft chatbot from the 1990s that famously irritated Microsoft Office users.
Turns out I was wrong. Ed is a much bigger disaster. Clippy simply caused Microsoft some embarrassment. They even brought Clippy back a few years ago as a 90s nostalgia play. Edβs not coming back. His probable future is to be quietly scrubbed from the LAUSD website and for the giant sun costume to be deflated and placed in a dumpster. Alberto M. Carvalho, the man standing not-at-all awkwardly next to Ed in the picture below, will issue a statement explaining that it wasnβt his fault. How could he, or anyone, have known this wouldnβt go well? Maybe some poor IT or procurement staff will lose their job, but I bet Carvalho will keep his annual salary of $440,000 and see his contract as superintendent renewed in 2026. He really knows how to put on a show.
Here is what I said about Carvalho and the other decision-makers who signed a reported $6 million contract with AllHere last year:
It is not hard to imagine how LAUSD got to last weekβs announcement. District executives, pressured by their board or excited about ChatGPT, told their staff to get moving on this AI thing. The result is a chatbot guardrailed up to its non-existing eyebrows and trained on data already publicly accessible on the LAUSD website. Ed is not a personalized tutorβ¦.its personalization consists of showing a student their report card and attendance record (Iβm sure thatβs fine, right?) so its advice mostly consists of referrals to existing services and websites.
I have no idea what role the LAUSD contract played in AllHereβs demise. I donβt know how many of the β9,100 schoolsβ that AllHere claims trust its AI solutions are in LA and, therefore, how much of its revenue is tied to the LAUSD contract. Maybe the District stopped paying AllHere because they were not delivering on the promised personalization or language translation features? Maybe AllHere expanded staff unwisely, anticipating that the announcement in March would yield additional big contracts? We donβt know.
It is worth pausing on that fact. Why donβt we know? In fact, why are you reading about this on some blog instead of in a newspaper or a website that covers educational news?
In addition to providing context to better understand news stories about educational technology, ππ ππ¨π reviews books and publishes essays about the social and historical contexts of generative AI.
Since The74 broke the story of AllHereβs impending demise on June 26, there has been no other news coverage (Updated 1 July 2024: Dana Goldstein has a good piece up this morning in the NYT). I hope reporters are busy digging into financial records and talking to sources among the furloughed employees and LAUSD employees. But maybe not. Education Week, which covered the rollout of Ed, is busy interviewing Google's head of education about how AI can save teachers time. The LA Times recently laid off a bunch of reporters and editors, including technology critic Brian Merchant. You should check out his book and Substack called Blood in the Machine. His latest post is a guided tour of the coming βtrough of disillusionmentβ with generative AI.
AllHere received glowing press coverage from when it started in 2016 to a recent video by freelance tech writer Luke Edwards celebrating its achievements. The CEO was a former teacher! She went to Harvard! It was first funded through the Harvard Innovation Labs! The story of a scrappy start-up and a district helping students through a magical new technology checked all the right boxes for enthusiastic, if uncritical, coverage. I guess the story of a school district flushing money down the drain on overblown tech and a venture capital start-up going bankrupt would not generate advertising clicks or write itself from a press release.
The news about Edβs demise will garner far less attention than its rollout. The reason is, of course, that the business structures of the news media are dissolving during the same period that the generative AI hype roller coaster has been climbing the Peak of Inflated Expectations. There is a growing conglomeration of quasi-journalistic and quasi-professional organizations that employ writers content creators to hype venture capital-powered start-ups with fawning coverage of young, charismatic entrepreneurs and game-changing technology. But donβt expect them to revisit this story now that it has turned dark or for any of the organizations that rewarded AllHere for its origin story and empty promises to try to explain its demise.
Except for a handful of blogs and small non-profit journalism outfits like The74, EdSurge, and The Markup (use the comments to point to others!), there just is not much critical coverage of ed-tech. There is little in place to counter the relentless hype of generative AI, even as we should be preparing people for the coming plunge.
Right now, across the US and worldwide, ed-tech companies selling products βpowered with AI!β are making sales calls to the offices of superintendents, chancellors, and provosts with a pitch that starts with questions shaped by the narrative that generative AI has already changed the world.
Do you understand that students must use AI tools to succeed in our new, AI-powered world? Your competitors are moving fast and signing contracts with us. Shouldnβt you get your head out of the sand and start implementing generative AI at scale today?
The sales teams at these companies know about the hype cycle, or at least they feel the current power of the media blitz featuring Sal Khan, his book, and the staged demos of a magic computer talking to Khan and his son. These companies need to book sales while the roller coaster is still up high. They donβt have much time before the disillusionment sets in as people actually use the technology and find it doesnβt live up to the hype.
My advice is not to try to explain why the assumptions embedded in the sales pitch are wrong. The world of AI-powered sales is defined by the phrases βat scaleβ and βscaling at speed.β The notion of moving slowly and taking care sounds like weak-minded nonsense. They cannot understand the perspective of institutions responsible for the welfare and learning of young people because, as Upton Sinclair explained long ago, βIt is difficult to get a man to understand something when his salary depends upon his not understanding it.β So, when they pitch you, it is better to turn the tables and ask them some questions:
Why would any school system or college invest money in this technology right now? There are no clear educational uses for generative AI, and we barely understand what it can do or how it works. All we have seen are flashy demos. Since youβre so sure this is transformative, why donβt you let us use your product for free this academic year so we can try it out with some small pilot programs?
Well-positioned and well-funded companies will take you up on that offer. If the LAUSD had taken that approach with AllHere, they might have avoided losing however much of the $6 million dollar contract paid so far. And maybe AllHere would have taken a more conservative approach to growing its payroll. There are potential use cases for generative AI in education, even if it is not yet clear that the technology can reliably do the job.
Venture capital-funded technology is not all bad.1 Youβre reading this blog for free on a nice interface without ads, thanks, in part, to the largess of Andreessen Horowitz. But the move fast, break things mentality necessary to produce a few big winners out of thousands of bets runs on a hype machine and on a media ecosystem aimed at creating intense feelings of missing out. Stories like Ed and AllHere disrupt that feeling. Teachers and those who purchase technology on their behalf need to know those stories so as to not get carried away by the demos and the enthusiasm of early adopters.
The story of Ed the chatbot should remind us that a return on investment and scaling at speed are Silicon Valleyβs problems to solve. Our problem is how best to educate students in a rapidly changing world. Solving our problem means putting generative AI tools in the hands of teachers and students for experimental pilots. Then, listen to what they have to say and consider what investments make sense at a scale and timeline that supports effective adoption.
Big projects are always a risky proposition, more so now that we are approaching an inflection point. The story of LAUSD and AllHere is an important reminder that booms are followed by busts. Letβs make sure people hear the story.
ππ ππ¨π offers links, analysis, and reflection on developments in AI and education. Learn more here.
π¨π° π³ππ Β© 2024 by Rob Nelson is licensed under CC BY-SA 4.0.
Iβm skeptical of Silicon Valley but not as cynical as most skeptics. It is a deeply weird place with a lot of deeply weird ideas. Think about what it must be like to go to parties where you argue about how likely it is that AI is going destroy humanity and then get up the next morning to work on building that technology as quickly as possible. Iβm too interested in the old, weird cultural history of 19th-century American business and confidence artists not to be fascinated by its 21st-century manifestations.
I get the discomfort with handing out loads of capital to young people (mostly men) in a rigged game where the tiny percentage of big winners attracts thousands of people who will end up in their thirties with no job and little savings. Before I pick up a nice throwing rock, I have to admit that Iβve spent my career getting paid by institutions that run a similar game for young male athletes. At least young entrepreneurs at startups get paid before they bust out. Plus, they end up with job skills and good stories. The majority of young football players mostly just end up with minor brain damage and student debt.
Great essay. The NYT reported on this debacle this morning. They missed the opportunity to use that amazing pic of Cavalho and "ED," however. https://www.nytimes.com/2024/07/01/us/ai-chatbot-los-angeles-schools.html
I briefly forgot this was your blog, @Rob, and was going to post to my LinkedIn and tag you before I realized, haha. Great points all around - it's a little (or a lot? ) fools gold right now.