I spent some time this past week talking with my students about artificial intelligence (AI) and whether and when it is appropriate to use in school. Their responses were wide-ranging and nuanced. Some cited the environmental impacts and opposed any use of AI, others thought it could help them break down complicated topics, and a few (of course) joked about having AI do all of their homework.
We discussed it because I wanted to hammer home one central idea: AI is a tool. It’s not “good” or “bad” any more than a screwdriver is “good” or “bad.” A screwdriver might be really great if you are assembling a bookshelf, but it’s not so great for cutting down a tree.
(I’ll take a moment here to specify that my conversations with my students were focused on large language model chatbots like Gemini and ChatGPT. There are obviously more consequential uses of AI, and those uses warrant even more thorough discussion and scrutiny.)
I explained to students that large language models are a poor fit for much of what we do in school. For instance, letting students use AI when I am assessing their knowledge of a historical event would completely miss the point. I don’t need to know what the AI knows; I need to know what the students know.
Yes, AI can be a helpful research tool, but we discussed the significant risk that AI misinterprets something or confuses facts. Thus, anything a large language model says must be fact-checked with credible sources.
There are, however, times when AI can be helpful in school. Some students said it can help explain how to do a math problem step by step. I mentioned that some teachers have used chatbots to give students a chance to “chat” with historical figures like George Washington. Large language models can also be meaningful idea generators during the brainstorming phase of projects.
Yet, an unexpected thing happened as I discussed this topic with students: they were surprised that I was willing to admit that AI could be beneficial. Their response struck me, because if anything, I’ve always thought of myself as more of an AI skeptic. More often than not, I have found AI tools to be underwhelming, at least when compared to the pie-in-the-sky promises of Sam Altman, Elon Musk, and their ilk. Now this roomful of students was acting as if I were some kind of AI evangelist simply because I did not treat it like the forbidden fruit.
Breaking out of the binary
The students’ response, I suspect, goes back to the difficulty many people have breaking out of the binary good/bad paradigm when it comes to AI. In a school context AI=bad, they figured, and thus the idea that AI might sometimes be helpful did not compute.
It’s quite understandable, actually. When it comes to AI, most of the things we hear about it are black-and-white. It’s either a magical cure-all or an economic time bomb. Yet, most of our everyday experience of AI has been, well, gray. Or, to put it in teenage vernacular: Most AI is decidely mid.
I spent some time this week using Google Gemini to create a tool that will automate a menial task I have to do each schoolday. It took a couple of hours to build it, and it will save me five to ten minutes each school day from here on out.
I would not have been able to build the tool without Gemini’s help. Yet, Gemini was far from perfect. Over and over again, I would present a problem with the code Gemini generated, and it would respond by explaining why the error happened, and then proposing a fix. Virtually every time, the first fix failed. Often, the second fix did not work, either. In response, Gemini would explain exactly what had gone wrong, and then propose a new fix. A few times, it declared it had found a “bulletproof” solution. Usually—but not always—the bulletproof solution worked. The tool was worth the time investment, but if I had thought that Gemini would spit out a perfect solution in two minutes, I would have been sorely disappointed.
Indeed, if Altman had launched ChatGPT by holding a flashy press conference and then proceeding to use the next two-and-a-half hours going back and forth as he prompted, corrected, and troubleshot ChatGPT in order to design a simple tool, I doubt OpenAI would have the mystique it enjoys today.
However, I also doubt OpenAI would be as loathed as it is today.
If the company and its peers had chosen to market their products as helpful-but-not-perfect research and composition tools rather than as be-all-end-all, maybe-help-you, maybe-kill-you terror machines, the companies might not have the 12-figure valuations they boast today. Then again, they might also have much improved long-term viability.
Backlash hits schools
Within the past week, the nation’s two largest school districts have announced moratoria on the use of AI in school. In doing so, they took the opposite approach to Miami schools, which have tried to position themselves as early adopters of AI.
It’s possible that the moves by New York City and Los Angeles are attempts to get a better handle on when and how to use large language models in school. It’s also possible they are just a temporary reaction to the current backlash against AI. I hope those school districts use the next year not only to think about whether to use AI, but also about how to teach students about it. It matters that students have a realistic understanding of what AI is and is not.
If people truly want to put AI in its place, though, they can start by literally putting it in its place: a toolbox of computer applications, some of which work better than others.