A provocative new study out of MIT has ignited headlines claiming that ChatGPT may be harming your mind. However as is usually the case with viral AI tales, the reality is much extra nuanced.
On Episode 155 of The Artificial Intelligence Show, I broke down the examine with Advertising AI Institute founder and CEO Paul Roetzer to find out what’s price listening to.
The Research at a Look
Titled “Your Mind on ChatGPT,” the examine analyzed how totally different instruments affect cognitive engagement throughout essay writing. Members have been break up into three teams:
- One wrote essays utilizing solely their reminiscence (“brain-only”)
- One used a search engine
- One used ChatGPT (GPT-4o)
Utilizing EEG scans and linguistic evaluation, researchers discovered that contributors who relied on ChatGPT confirmed weaker neural connectivity and fewer engagement in reminiscence and decision-making areas of the mind. Their essays have been extra uniform and fewer unique. And so they had extra hassle remembering or quoting from their very own writing—even simply minutes after finishing it.
However there’s a catch.
Why the Panic Is Untimely
The paper, as AI expert Ethan Mollick points out, is being badly misinterpreted in viral posts and sensational headlines. He writes on LinkedIn:
“This new working paper out of the MIT Media Lab is being massively misinterpreted as “AI hurts your mind.”
It’s a examine of faculty college students that finds that those that have been instructed to write down an essay with LLM assist have been, unsurprisingly, much less engaged with the essay they wrote, and thus have been much less engaged once they have been requested to do comparable work months later. It says one thing vital about dishonest with AI (when you let it do your work you will not be taught) nevertheless it does not inform us something about LLM use making us dumber total.
This misinterpretation is not helped by the truth that this line from the summary could be very deceptive: “Over 4 months, LLM customers persistently under-performed at neural, linguistic, and behavioral ranges.” However the examine doesn’t check “LLM customers” over 4 months, it exams (9 or so!) individuals who had an LLM assist write an essay in an experiment writing an analogous essay 4 months later.
To be clear this is not a protection of blindly utilizing AI in schooling, they’ve for use correctly to be efficient. We all know from this well-powered randomized managed research that simply having the AI offer you solutions lowers check scores.
However that does not imply that LLMs rot your mind.”
Roetzer agrees, declaring you could in a short time get a way of Mollick’s level when you truly learn additional concerning the examine, moderately than simply consuming headlines.
“It is like saying we gave calculators to a management group who did not know tips on how to do math, and we discovered that individuals who relied on the calculator to do math did not truly be taught math,” he says.
The actual takeaway? In case you use AI to bypass essential pondering, you’ll assume much less. That’s not a revelation—it’s simply frequent sense.
From Analysis to the Actual World
The examine appears methodologically stable and factors to an actual impact. But it surely truly strengthens the case for accountable AI use, says Roetzer. In schooling and enterprise, the true crucial is to show folks tips on how to use AI instruments to speed up—not substitute—studying and pondering.
Meaning:
- Beginning with human comprehension
- Utilizing AI to check, refine, or develop concepts—to not generate total outputs blindly
- Creating environments that reward cognitive engagement, not simply completed deliverables
Roetzer additionally launched a useful framework to grasp the place there are cognitive gaps in AI outputs that human brains might have to pay attention to and/or verify:
- The Verification Hole: People have to fact-check AI outputs.
- The Pondering Hole: People have a restricted capability to critically consider AI-generated content material.
- The Confidence Hole: With out really participating with the underlying materials, discomfort arises while you current materials created by AI however don’t absolutely perceive it.
Collectively, these gaps clarify a rising dynamic within the office:
As we produce extra with AI, if we do not perceive and tackle these gaps, we threat retaining and understanding much less.
In Roetzer’s personal expertise, this dynamic performs out even in day-to-day conferences.
“I nonetheless kind out all the pieces in each assembly I am going to,” he says, moderately than depend on an AI notetaker. “If I simply have the notetaker, there’s much less cognitive load. However that cognitive load is definitely what embed it in my reminiscence.”
The Backside Line: Essential Pondering Nonetheless Issues
The MIT examine, when learn fastidiously, doesn’t say AI is inherently harmful to our brains. It says lazy use of AI is. And that’s an vital distinction.
So what’s the good takeaway? Use ChatGPT. Use it typically. However by no means outsource your mind. As a result of the talent that issues most within the age of AI is similar one which mattered earlier than it: figuring out tips on how to assume.
