AI will revolutionize education the way a gambling addiction will revolutionize your bank account
I have a new essay in The Hedgehog Review on AI; I hope you'll read it
I have been thinking a lot about AI lately. Which is to say, I work (some) in higher education. The current state of discourse about AI in higher ed is completely unglued. People are saying the craziest things. In response, I decided to say something not-crazy about learning with and from other people, real people, in the limited time before we die.
The main “action item” for readers of this post is to read my much longer and better essay, “ChatGPT Is a Gimmick,” which is just out as a web feature for The Hedgehog Review, one of America’s best little magazines. I don’t know if I can offer a concise summary of the essay here (you’ll need AI for that, I suppose). I just hope you’ll read and enjoy it. There is no paywall.
Thanks to Matt Dinan, Atlantic Canada’s viral-postingest humanities professor, the essay is getting attention on social media. People are saying the essay is beautiful; they love the last paragraph. There’s your “social proof” that the essay is worth reading. Please do!
Matt also had a very good, short Substack post recently about intellectual skills, AI, and the future of education. I think he’s right about everything in that post, except for one thing, which is, of course, what I’m going to focus on. Matt says that paradoxically, AI will spur learning to “shift back in the direction of genuine education delivered person-to-person. Sadly, this will become the reserve of the rich and the privileged.”
The part I disagree with is that AI-delivered “education” will be a cheap substitute for in-person education. And the simple reason is, AI is not cheap! The real cost of AI is hidden even from users who pay for “premium” ChatGPT accounts. I’ve been diving into1 the financial position of OpenAI, the company that basically is the AI industry in the US. It is shocking how much of investors’ money the company is spending, and how little revenue it is getting in return. In short, OpenAI is not remotely close to breaking even, and it has no obvious path to profitability in the years ahead. Its costs are only increasing.
Remember MOOCs? The massive open online courses that were allegedly going to revolutionize education 10-15 years ago? I do. They offered a truly terrible education to any but the most motivated learners, but they in principle relied on economies of scale. The initial capital investment was huge, but you could run a MOOC many times for infinite students at minimal operating cost, because the course content was the same for everyone.
Not so with AI. The whole point of generative AI is that it responds to your every prompt in real time. And that requires tremendous computing power, which is expensive. It’s like responding to inefficiencies in the public transit system by encouraging more people to drive their own cars.
Now you might think that once OpenAI develops a really good product, it can sell or license it to universities and make a profit there. I doubt it. AI companies will have to charge universities a fortune for “enterprise” services (like an in-house large language model) that will still have the same operating costs and will in fact cost even more to train on a university’s bespoke content archive. The universities racing to implement AI are racing to burn through billions of dollars for absolutely no purpose. Smart universities will hang back.
I feel pretty confident that AI education will not be cheaper than a full-time prof teaching 50-100 students per semester. Never. It will always be more expensive. Let’s acknowledge this and get back to the difficult yet ultimately satisfying work of teaching.
There is more I want to say about AI, particularly AI and death, but it will have to wait. In the meantime, please read my Hedgehog essay! Thank you.
reading one article about