MIT Experiment Finds ChatGPT-Assisted Writing Weakens Student Brain Connectivity and Memory 36
ChatGPT-assisted writing dampened brain activity and recall in a controlled MIT study [PDF] of 54 college volunteers divided into AI-only, search-engine, and no-tool groups. Electroencephalography recorded during three essay-writing sessions found the AI group consistently showed the weakest neural connectivity across all measured frequency bands; the tool-free group showed the strongest, with search users in between.
In the first session 83% of ChatGPT users could not quote any line they had just written and none produced a correct quote. Only nine of the 18 claimed full authorship of their work, compared with 16 of 18 in the brain-only cohort. Neural coupling in the AI group declined further over repeated use. When these participants were later asked to write without assistance, frontal-parietal networks remained subdued and 78% again failed to recall a single sentence accurately.
The pattern reversed for students who first wrote unaided: introducing ChatGPT in a crossover session produced the highest connectivity sums in alpha, theta, beta and delta bands, indicating intense integration of AI suggestions with prior knowledge. The MIT authors warn that habitual reliance on large language models "accumulates cognitive debt," trading immediate fluency for weaker memory, reduced self-monitoring, and narrowed neural engagement.
In the first session 83% of ChatGPT users could not quote any line they had just written and none produced a correct quote. Only nine of the 18 claimed full authorship of their work, compared with 16 of 18 in the brain-only cohort. Neural coupling in the AI group declined further over repeated use. When these participants were later asked to write without assistance, frontal-parietal networks remained subdued and 78% again failed to recall a single sentence accurately.
The pattern reversed for students who first wrote unaided: introducing ChatGPT in a crossover session produced the highest connectivity sums in alpha, theta, beta and delta bands, indicating intense integration of AI suggestions with prior knowledge. The MIT authors warn that habitual reliance on large language models "accumulates cognitive debt," trading immediate fluency for weaker memory, reduced self-monitoring, and narrowed neural engagement.
It almost writes itself. (Score:3)
Re:It almost writes itself. (Score:5, Insightful)
yes. It would seem that when it comes down to it, actually doing the work is what causes one to learn the material and basically shortcuts don't result in learning the material.
For what it's worth I didn't even like taking notes on a computer, I liked taking them by hand. And don't get me wrong, I don't like handwriting and my penmanship leaves a lot to be desired. I just found that if I wrote out my notes on paper that I learned the material, I did not even have to refer back to said notes most of the time. The act of writing them down helped me learn far more than typing the notes did.
Re: (Score:2)
About the most fundamental statement in neuroscience is 'cells the fire together wire together'.
Of course the more time you spend thinking about and doing something the better you remember it and the more quickly you will be able recall the details.
I would think it would be obviously to anyone that using tools that allow you to skip over a lot of the composition effort, organizing facts you want to present into a narrative means you get less recall of those facts later.
Re: (Score:2)
yes. It would seem that when it comes down to it, actually doing the work is what causes one to learn the material and basically shortcuts don't result in learning the material.
For what it's worth I didn't even like taking notes on a computer, I liked taking them by hand. And don't get me wrong, I don't like handwriting and my penmanship leaves a lot to be desired. I just found that if I wrote out my notes on paper that I learned the material, I did not even have to refer back to said notes most of the time. The act of writing them down helped me learn far more than typing the notes did.
Exactly. I posted in one of these threads that anything I have had written for me with any AI just doesn't get committed to memory.
Note taking - I take copies note most of the time. I seldom reference them, because the act of writing them down commits them to memory.
So I couldn't agree more - AI might find the answers, but we'll need to keep asking the same questions over and over again.
Re: It almost writes itself. (Score:2)
Re: (Score:2)
Either way it will at least be a problem; since the current reliability of bots basically requires knowledgeable and experienced people to supervise them and know when to just give their output a look and pass it along, when to prod them on errors to try to get them fixed, and when to j
Re: (Score:2)
I'm not sure. I myself made a similar argument about GUI management tools when the CLI tools were still required for troubleshooting, advanced configuration, and other similar things. I was concerned that the use of the 'dumb GUI tool' would lead to generations of IT engineers that couldn't fall back on basic principles.
In my thought experiment I was proven wrong, but only out of those GUI tools failing to provide a complete enough solution to begin with that the IT engineers could forego the CLI. They s
Re: It almost writes itself. (Score:2)
True.
And for a number who are "visually oriented", making hand sketches and graphs is as strong a generator of neural connections.
Re: (Score:3)
I don't think there's anything wrong with those sorts of general observations (I mean, who remembers dozens of phone numbers anymore now that we all have smartphones?), but that said this non-peer-reviewed study has an awful lot of problems. I mean, we can focus on the silly, embarassing mistakes (like how their methodology to suppress AI answers on Google was to append "-ai" into the search string, or how the author insisted to the press that AI summaries mentioning the model used were a hallucination, w
Re: It almost writes itself. (Score:2)
Funny. Because any mention of "class" in TFS or TFA was not in reference to student's socio-economic status. It was most probably a reference to their activity involving sitting at desks, listening to some instructor drone on about some mind-numbing subject.
It's possible that continued exposure to news about LLMs adversely affects some posters ability to develop meaningfull semantic relationships.
Re: (Score:2)
So ... (Score:2)
Re: (Score:2)
From the paper... (Score:2)
Reducing the capacity for independent thinking sounds like a wet dream for repressive governments, oligarchs, and marketing people.
Re: (Score:1)
Maybe the lesson here should be that students learn to give the teacher what they want (an essay or test). Why retain that information beyond what's necessary for completion of the objective? Math and some sciences we can use
Re: From the paper... (Score:3)
Because understanding history goes a long way toward understanding current events and politics. The leftists and AGW proponents would rather that you had never heard of Stalin or Lysenko.
Re: (Score:2)
And the fascists would rather you never knew about slavery in America or the fact that LGBT people exist.
Education is inimical to both extremes of the political spectrum.
Re: (Score:2)
AGW proponents
As in, people who accept reality, rather than being utterly blinded by politics.
AGW is happening. The greenhouse effect has been known about for 201 years now and well established for over a hundred.
Re: From the paper... (Score:2)
Re: From the paper... (Score:2)
Re: (Score:2)
A lot of that is more about imposing homework discipline and patterns of thinking, rather than actually expecting you to remember every detail.
Yeah, I think ultimately school is less about teaching you subjects and more about just teaching you to learn. Most people will never remember the history they learned in grade school, but maybe they will learn when things are happening in the world, to go back to history books and see if any previous time repeated this same situation. Also any job without step by step instructions require the employee to keep learning new things and how to apply them, hopefully they learned to do this in school, rather th
Re: (Score:2)
Sure, most of what I learned in school I never used either.
But it's almost impossible to predict what some kids will end up using. Just because you were uninterested in (say) history, it doesn't mean learning it was a waste for every single kid.
Also, I'm of the strong belief that more knowledge is better, even if you think you'll never use it.
Re: (Score:2)
like a calculator (Score:5, Insightful)
Just like a calculator does not help you remember how to do long division. Tools are great but I think learning involves learning the basics first then learning how to things with tools. If you don't understand the basics you cannot check for correctness, and with generative AI you need to do that.
ok, but (Score:2)
Ok, I'm willing to believe the study. But chances are it can't be replicated.
Re: ok, but (Score:2)
It makes sense to me. If the tool does the work for you, you do less. Sooner or later you lose skill and become dependent on the tool.
Data collected so far does not support conclusion. (Score:2)
At least that's what I heard skeeted on Bluesky: https://bsky.app/profile/danbi... [bsky.app]
And of course this is no surpriseâ"an inconclusive study that can be used to justify a controversial headline is gold in the infotainment biz.
Re: (Score:2)
BTW, the ascii glorp above is a unicode em-dash, proving that I am an LLM and hence not a disinterested party to the discussion. :)
Our schools don't exist to teach people (Score:1)
Mind you teachers will try to use the school system for unintended purposes like improving the lives of children. But understand it's not the actual purpose.
Our school systems are a machine that turns kids into corporate profits. Anything else that comes out of it is purely accidental or coincidental.
In practice this means that if a kid can't just le
Just wait till you see their code, right? (Score:2)
I'm sure using ChatGPT will have no negative impact on their code quality
On the bright side (Score:2)
They clicked through ads more often. So there's that.