An AI Thought Experiment

To help folks think through what we should be considering regarding the impact on education of generative AI tools like large language models, I want to try a thought experiment.

Imagine if, in November 2022, OpenAI introduced ChatGPT to the world by letting the monster out of the lab for a six-week stroll, long enough to demonstrate its capacities—plausible automated text generation on any subject you can think of—and its shortfalls—making stuff up—and then coaxing the monster back inside before the villagers came after it with their pitchforks.

Periodically, as new models were developed that showed sufficient shifts in capabilities, the AI companies (OpenAI having been joined by others), would release public demonstrations, audited and certified by independent expert observers who would release reports testifying to the current state of generative AI technology.

What would be different? What could be different?

First, to extend the fantasy part of the thought experiment, we have to assume we would actually do stuff to prepare for the eventual full release of the technology, rather than assuming we could stick our heads in the sand until the actual day of its arrival.

So, imagine you were told, “In three years there will be a device that can create a product/output that will pass muster when graded against your assignment criteria.” What would you do?

A first impulse might be to “proof” the assignment, to make it so the homework machine could not actually complete it. You would discover fairly quickly that while there are certainly adjustments that can be made to make the work less vulnerable to the machine, given the nature of the student artifacts that we believe are a good way to assess learning—aka writing—it is very difficult to make an invulnerable assignment.

Or maybe you engaged in a strategic retreat, working out how students can do work in the absence of the machine, perhaps by making everything in class, or adopting some tool (or tools) that track the students’ work.

Maybe you were convinced these tools are the future and your job was to figure out how they can be productively integrated into every aspect of your and your students’ work.

Or maybe, being of a certain age and station in life, you saw the writing on the wall and decided it was time to exit stage left.

Given this time to prepare, let’s now imagine that the generative AI kraken is finally unleashed not in November 2022, but November 2024, meaning at this moment it’s been present for a little under six months, not two and a half years.

What would be different, as compared to today?

In my view, if you took any of the above routes, and these seem to be the most common choice, the answer is: not much.

The reason not much would be different is because each of those approaches—including the decision to skedaddle—accepts that the pre–generative AI status quo was something we should be trying to preserve. Either we’re here to guard against the encroachment of the technology on the status quo, or, in the case of the full embrace, to employ this technology as a tool in maintaining the status quo.

My hope is that today, given our two and a half years of experience, we recognize that because of the presence of this technology it is, in fact, impossible to preserve the pre–generative AI status quo. At the same time, we have more than info information to question whether or not there is significant utility for this technology when it comes to student learning.

This recognition was easier to come by for folks like me who were troubled by the status quo already. I’ve been ready to make some radical changes for years (see Why They Can’t Write: Killing the Five-Paragraph Essay and Other Necessities), but I very much understood the caution of those who found continuing value in a status quo that seemed to be mostly stable.

I don’t think anyone can believe that the status quo is still stable, but this doesn’t mean we should be hopeless. The experiences of the last two and a half years make it clear that some measure of rethinking and reconceiving is necessary. I go back to Marc Watkins’s formulation: “AI is unavoidable, not inevitable.”

But its unavoidability does not mean we should run wholeheartedly into its embrace. The technology is entirely unproven, and the implications of what is important about the experiences of learning are still being mapped out. The status quo being shaken does not mean that all aspects upon which that status quo was built have been rendered null.

One thing that is clear to me, something that is central to the message of More Than Words: How to Think About Writing in the Age of AI: Our energies must be focused on creating experiences of learning in order to give students work worth doing.

This requires us to step back and ask ourselves what we actually value when it comes to learning in our disciplines. There are two key questions which can help us:

What do I want students to know?

What do I want students to be able to do?

For me, for writing, these things are covered by the writer’s practice (the skills, knowledge, attitudes and habits of mind of writers). The root of a writer’s practice is not particularly affected by large language models. A good practice must work in the absence of the tool. Millions of people have developed sound, flexible writing practices in the absence of this technology. We should understand what those practices are before we abandon them to the nonthinking, nonfeeling, unable-to-communicate-with-intention automated syntax generator.

When the tool is added, it must be purposeful and mindful. When the goal of the experience is to develop one’s practice—where the experience and process matter more than the outcome—my belief is that large language models have very limited, if any, utility.

We may have occasion to need an automatic syntax generator, but probably not when the goal is learning to write.

We have another summer in front of us to think through and get at the root of this challenge. You might find it useful to join with a community of other practitioners as part of the Perusall Engage Book Event, featuring More Than Words, now open for registration.

I’ll be part of the community exploring those questions about what students should know and be able to do.

Join us!