Universities should arm students with AI ‘eval’ powers
The writer is professor of political economy at the Stanford Graduate School of Business
Today, young people are watching enormous power concentrate in a handful of AI companies whose systems will shape their careers, their information and their opportunities — and some of them feel, understandably, that they have little say in any of it.
At the same time, many of the universities educating these students are responding with understandable suspicion, restricting or even banning AI in the classroom. The institutions meant to prepare the next generation for a world remade by AI are walling the technology off from the people who will most need to understand how to wield it.
There is a better response.
In a recent essay, Satya Nadella, chief executive at Microsoft, described what he calls the new “loop” — a process by which an organisation combines its own expertise with AI and then regularly evaluates how well AI systems are serving its particular goals. The organisation tests the models against what it actually needs, repeatedly optimises it and ends up in a position where it — and not the companies that train the underlying models — understands where the value lies and how to capture it.
As companies wake up to the threat that AI tool providers can capture their data, copy their competitive edge, and render them obsolete, this loop offers them a new kind of sovereignty. The test should be that if they can replace one model with another in their workflows — swap out OpenAI’s ChatGPT for Anthropic’s Claude, say — without losing any value, then they have successfully made themselves sovereign.
Nadella was writing about companies, but the loop is really a theory of how anyone stays in charge of their own destiny in an AI world.
More from the AI in Education reportAI labs muscle in on $6tn education market; Universities face difficult choices over how to integrate AI; Is AI killing critical thinking in the classroom?
I have been experimenting with teaching it i…