How schools can survive (and maybe even thrive) with AI this fall

Last November, when ChatGPT was released, many schools felt as if they’d been hit by an asteroid.

In the middle of an academic year, with no warning, teachers were forced to confront the new, alien-seeming technology, which allowed students to write college-level essays, solve challenging problem sets and ace standardised tests.

Some schools responded — unwisely, I argued at the time — by banning ChatGPT and tools like it. But those bans didn’t work, in part because students could simply use the tools on their phones and home computers. And as the year went on, many of the schools that restricted the use of generative artificial intelligence — as the category that includes ChatGPT, Bing, Bard and other tools is called — quietly rolled back their bans.

Ahead of this school year, I talked with numerous K-12 teachers, school administrators and university faculty members about their thoughts on AI now. There is a lot of confusion and panic, but also a fair bit of curiosity and excitement. Mainly, educators want to know: How do we actually use this stuff to help students learn, rather than just try to catch them cheating?

I’m a tech columnist, not a teacher, and I don’t have all the answers, especially when it comes to the long-term effects of AI on education. But I can offer some basic, short-term advice for schools trying to figure out how to handle generative AI this fall.

First, I encourage educators — especially in high schools and colleges — to assume that 100% of their students are using ChatGPT and other generative AI tools on every assignment, in every subject, unless they’re being physically supervised inside a school building.

At most schools, this won’t be completely true. Some students won’t use AI because they have moral qualms about it, because it’s not helpful for their specific assignments, because they lack access to the tools, or because they’re afraid of getting caught.

But the assumption that everyone is using AI outside class might be closer to the truth than many educators realise. (“You have no idea how much we’re using ChatGPT,” read the title of a recent essay by a Columbia undergraduate in the Chronicle of Higher Education.) And it’s a helpful shortcut for teachers trying to figure out how to adapt their teaching methods. Why would you assign a take-home exam, or an essay on “Jane Eyre,” if everyone in class — except, perhaps, the most strait-laced rule-followers — will use AI to finish it? Why wouldn’t you switch to proctored exams, blue-book essays and in-class group work, if you knew that ChatGPT was as ubiquitous as Instagram and Snapchat among your students?

Second, schools should stop relying on AI detector programs to catch cheaters. There are dozens of these tools on the market now, all claiming to spot writing that was generated with AI, and none of them work reliably well. They generate lots of false positives, and can be easily fooled by techniques like paraphrasing. Don’t believe me? Ask OpenAI, the maker of ChatGPT, which discontinued its AI writing detector this year because of a “low rate of accuracy.”

It’s possible that in the future, AI companies may be able to label their models’ outputs to make them easier to spot — a practice known as “watermarking” — or that better AI detection tools may emerge. But for now, most AI text should be considered undetectable, and schools should spend their time (and technology budgets) elsewhere.

My third piece of advice — and the one that might get me the most angry emails from teachers — is that teachers should focus less on warning students about the shortcomings of generative AI than figuring out what the technology does well.

Last year, many schools tried to scare students away from using AI by telling them that tools like ChatGPT are unreliable, prone to spitting out nonsensical answers and generic-sounding prose. These criticisms, while true of early AI chatbots, are less true of today’s upgraded models, and clever students are figuring out how to get better results by giving the models more sophisticated prompts.

As a result, students at many schools are racing ahead of their instructors when it comes to understanding what generative AI can do, if used correctly. And the warnings about flawed AI systems issued last year may ring hollow this year, now that GPT-4 is capable of getting passing grades at Harvard.

Alex Kotran, the chief executive of the AI Education Project, a nonprofit that helps schools adopt AI, told me that teachers needed to spend time using generative AI themselves to appreciate how useful it could be — and how quickly it’s improving.

This article originally appeared in The New York Times.