AI tools in education

It is inevitable that students will be tempted to use generative AI tools to do schoolwork. No doubt some students already are. So it is totally reasonable for us to be disconcerted by certain possibilities, such as students using ChatGPT to write essays with almost no effort on their part. But it is also important to understand that regardless of how much future AI tools live up to the current level of hype, today’s AI tools have real capabilities today that aren’t going away and some of which already make them legitimately useful. They also have serious limitations, some of which make them, at least presently, inappropriate tools for many tasks.

In order to formulate thoughtful policies on the acceptable use of AI tools in school we need to consider possible uses along two dimensions. The obvious dimension, and one that as educators we should be well equipped to evaluate, is whether a particular use will hurt or help students’ learning.

In some cases this is easy: we can probably all agree that a student pasting an essay prompt into ChatGPT and turning in ChatGPT’s output without even reading it has probably not learned much of anything. On the other hand, a student struggling with a concept who asks ChatGPT for an explanation and then asks follow-up questions about the parts they still didn’t understand may learn quite a lot, at least assuming the topic is one for which ChatGPT produces accurate explanations.

Which brings us to the other important dimension that we need to consider: whether current AI tools are actually good for a given use. Asking an AI tool to explain concepts is obviously only a good use of the tool if its explanations are good. Unfortunately the limitations of current AI tools make it easy to be misled about what they are actually good at.

For example, ChatGPT is often quite good at summarizing text that you paste into the chat box. However if you ask it to summarize something from a well-known book without actually providing the text it will probably make something up that, while plausible sounding, is not based on the contents of the book at all. From the point of view of how ChatGPT actually works, these are very different problems but from the point of view of someone not immersed in how GPT works and who has been told GPT was trained on “the whole internet” they might seem like very similar problems. Its success on the first task may mislead us or our students into thinking it will also be good at the second task.

This is just one example but if we want to guide our students towards uses of AI tools that will serve desirable educational purposes, we need to build our own understanding of the different possible uses of AI tools and where they sit on the dimensions of both learning and utility.

Additionally, I believe AI tools present a new educational challenge, beyond how we allow or disallow their use in school, which is we now have to teach our students to make their own evaluations of what AI tools are good for. Our students will live in a world where these tools exist and as citizens, workers, and leaders of the future, they will need to be able to continue to assess whether different uses of AI are appropriate or not as AI capabilities continue to evolve.

tl;dr?

Just for grins, I asked ChatGPT to summarize the above text:

The text discusses the use of artificial intelligence (AI) tools, like ChatGPT, in schools and the potential benefits and drawbacks these tools present. It highlights that while some students may misuse these tools to complete schoolwork with minimal effort, there are also ways in which these tools can assist in learning, such as providing explanations for difficult concepts.

However, the effectiveness of AI tools depends on their capacity to perform a task well, which varies based on their current limitations. For example, ChatGPT may excel at summarizing given text, but it can't generate accurate summaries for content it hasn't been directly given, like books. This can lead to misunderstandings about the capabilities of AI tools.

The text suggests that to formulate effective policies on the use of AI tools in schools, educators must consider two dimensions: the potential impact on students' learning and the actual effectiveness of the tool for a given task. Moreover, it suggests that as AI tools continue to evolve, students should be taught how to critically evaluate their uses, as they will become citizens, workers, and leaders in a world where these tools are prevalent.