Taking a slight detour this week to dive into the recent MIT article with the very catchy name Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.

Given this is a 200+ page academic article, I’ve enlisted some AI help, to find some quicker ways to absorb it (I promise, I did take a look at the original though!).

I asked Perplexity for an analytical review of the coverage – so the articles about the article, and then loaded the URLs of the pages Perplexity had referenced, along with the PDF of the MIT report into Notebook LM. And, yes, I did create a podcast.

The other thing I did is ask it which of the six articles Perplexity pulled up gave the best (as in most balanced, and thorough) review of the original paper. The answer is ChatGPT might be draining your brain from THE DECODER.

The author is a tech guy with a background in philosophy, which rocks for this kind of discussion.

So, in short, the MIT research team ran a bunch of tests – multiple sessions over time on the same people, where they had three groups write essays in a 20 minute time-frame. One group used ChatGPT, one used ‘traditional’ search (Google but without the AI summaries), and the third just used their own brains.

What they found is that using ChatGPT to help write essays in this way led to what they’re calling ‘cognitive debt’ in the group that relied on that tool. Basically, relying on the LLM weakened users’ learning and critical thinking (yikes!).

The researchers did things like get the essays marked by both AI and human assessors, and also asked the students who wrote the essays to recall something they put in there, pretty soon after they’d finished writing.

Apparently the markers tended to find the AI-assisted essays generic and ‘soulless’, while over 80% of the LLM group couldn’t recall anything they’d put in their essay, as compared with 11% for the brain-only team.

They repeated this exercise three times to see what happened over time (four months) and one of the things I found interesting is that the brain-only group’s neural activity increased over the sessions.

In an interesting twist on the tale, a final (fourth) session was conducted with a smaller group of the test students, where they swapped. So the ChatGPT group had to write an essay with no tools, and the brain-only group were asked to use ChatGPT. The LLM to brain-only group showed weaker neural connectivity BUT their neural activity was still stronger than in the first writing session for the brain-only group, so it’s not as dire as some headlines are making out, but it’s definitely a thing to be aware of.

Limitations

Keep in mind, this study was on a small group of subject (54 students overall and 18 in that final session), on a very specific task (writing an essay in 20 minutes), and the only AI tool used was ChatGPT.

It doesn’t look at other tools, or the various ways you can use the tools that might impact brain engagement differently, and of course it’s had be be done in a relatively short space of time.

Regardless, it’s a great starting point for discussion, and further investigation.

What this means

First up, I think it’s confirming what lot of people have been worrying about. If you use generative AI as a shortcut, you don’t engage with the work, or learn. It’s a much more passive activity than either searching for, and pulling information together, or hauling work 100% out of your own brain.

The question is, how do we set things up to encourage people (especially students) to spend some time in their own brains first? And what are ways we can use these tools to enhance our engagement and neural activity rather than shortcut around it?

Okay, yes that’s two questions, and the reason I’m calling out students in particular is that this is where they’re going to create the habits they bring into adulthood. We need to give them the best opportunities possible to succeed, with and without AI tools.

This is a simplified summary, and I’d recommend a look at the Decoder article if you want a better overview (I turned the article page into a pdf and it’s about 5 1/2 pages, so a decent read, but not academia length).

I’m pretty impressed by this study, the conversations it’s sparking, and where we might go next.

What are your thoughts? What are ways to fire up the mind (especially those of bored, disengaged teenagers in an out-of-date education system)?