Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task [edited post to change title and URL]

Does this also explain what happens with middle and upper management? As people have moved up the ranks during the course of their careers, I swear they get dumber.

That was my first reaction. Using LLMs is a lot like being a manager. You have to describe goals/tasks and delegate them, while usually not doing any of the tasks yourself.
- Fuck, this is why I'm feeling dumber myself after getting promoted to more senior positions and had only had to work in architectural level and on stuff that the more junior staffs can't work on.
  With LLMs basically my job is still the same.
- After being out of being a direct practitioner, I will say all my direct reports are “faster” in programs we use at work than I am, but I’m still waaaaaaaaaay more efficient than all of them (their inefficiencies drive me crazy actually), but I’ve also taken up a lot of development to keep my mind sharp. If I only had my team to manage and not my own personal projects, I could really see regressing a lot.
That's the Peter Principle.
My dad around 1993 designed a cipher better than RC4 (I know it's not a high mark now, but it kinda was then) at the time, which passed audit by a relevant service.
My dad around 2003 still was intelligent enough, he'd explain me and my sister some interesting mathematical problems and notice similarities to them and interesting things in real life.
My dad around 2005 was promoted to a management position and was already becoming kinda dumber.
My dad around 2010 was a fucking idiot, you'd think he's mentally impaired.
My dad around 2015 apparently went to a fortuneteller to "heal me from autism".
So yeah. I think it's a bit similar to what happens to elderly people when they retire. Everything should be trained, and also real tasks give you feeling of life, giving orders and going to endless could-be-an-email meetings makes you both dumb and depressed.
I’d expect similar at least. When one doesn’t keep up to date on new information and lets their brain coast it atrophies like any other muscle would from disuse.

The obvious AI-generated image and the generic name of the journal made me think that there was something off about this website/article and sure enough the writer of this article is on X claiming that covid 19 vaccines are not fit for humans and that there's a clear link between vaccines and autism.

Neat.

Thanks for the warning. Here's the link to the original study, so we don't have to drive traffic to that guys website.
https://arxiv.org/abs/2506.08872
I haven't got time to read it and now I wonder if it was represented accurately in the article.
- That’s a math article
Thanks for pointing this out. Looking closer I see that that "journal" was definitely not something I want to be sending traffic to, for a whole bunch of reasons - besides anti-vax they're also anti-trans, and they're gold bugs... and they're asking tough questions like "do viruses exist" 🤡
I edited the post to link to MIT instead, and added a note in the post body explaining why.

Microsoft reported the same findings earlier this year, spooky to see a more academic institution report the same results. https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf Abstract for those too lazy to click:

The rise of Generative AI (GenAI) in knowledge workflows raises questions about its impact on critical thinking skills and practices. We survey 319 knowledge workers to investigate 1) when and how they perceive the enaction of critical thinking when using GenAI, and 2) when and why GenAI affects their effort to do so. Participants shared 936 first-hand examples of using GenAI in work tasks. Quantitatively, when considering both task- and user-specific factors, a user’s task-specific self-confidence and confidence in GenAI are predictive of whether critical thinking is enacted and the effort of doing so in GenAI-assisted tasks. Specifically, higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking. Qualitatively, GenAI shifts the nature of critical thinking toward information verification, response integration, and task stewardship. Our insights reveal new design challenges and opportunities for developing GenAI tools for knowledge work.

~~Why is it referring to GenAI?~~ ~~It doesn't exist.~~
- GenAI is short for generative AI in this context
- I haven't read the paper but they might mean "Generative AI"

I wonder what social media does.

No wonder Republicans like it so much

So if someone else writes your essays for you, you don’t learn anything?

You write essay with AI your learning suffers.

One of these papers that are basically "water is wet, researches discover".

But does it cause this when when used exclusively for RP gooning sessions?

Somebody fund this scholar's research immediately
To date, after having gooned once (ongoing since September 2023), my core executive functions, my cognitive abilities and my behaviors have not suffered in the least. In fact, potato.

Isn't that the same guy that plays Michael Bolton in Office Space?

Its so disturbing. Especially the bit about your brain activity not returning to normal afterwards. They are teaching the kids to use it in elementary schools.

I think they meant it doesn't return to non-AI-user levels when you do the same task on your own immediately afterwards. But if you keep doing the task on your own for some time, I'd expect it to return to those levels rather fast.
- That's probably true, but it sure can be hard to motivate yourself to do things yourself when that AI dice roll is right there to give you an immediate dopamine hit. I'm starting to see things like vibecoding being as addictive as gambling.
  Personally I don't use AI because I see all the subtle ways it's wrong when programming, and the more I pay attention to things like AI search results, it seems like there's almost always something misrepresented or subtly incorrect in the output, and for any topics I'm not already fluent in, I likely won't notice these things until it's already causing issues

cognitive decline.

Another reason for refusing those so-called tools... it could turn one into another tool.

More like it would cause you to need the tool in order to be the tool that you are already mandated to be.
It’s a clickbait title. Using AI doesn’t actually cause cognitive decline. They’re saying using AI isn’t as engaging for your brain as the manual work, and then broadly linking that to the widely understood concept that you need to engage your brain to stay sharp. Not exactly groundbreaking.
- Sir this is Lemmy & I'm afraid I have to downvote you for defending AI which is always bad. /s

Heyyy, now I get to enjoy some copium for being such a dinosaur and resisting to use it as often as I can

You're not a dinosaur. Making people feel old and out of the trend is exactly one of the strategies used by big techs to shove their stuff into people.

This isn't just being said, it's being shown with data.
- Not only that but broad gestures at society and the state of the world post-Internet
The people the paper talks about are the masses who think LLMs are "intelligent", then outsource their frontal lobe to Silicon Valley datacenters because it's seemingly easier. People who see LLMs as tools are much less (if at all) affected by this, if anything it's a trap for people who already have lower critical thinking skills in the first place and want GPUs to think for them.
You don't think it's odd that you use AI and here you are defending it?
- Realistically speaking, why would anyone think it's odd to defend something they use and/or enjoy? That doesn't really point to anything abnormal.
- It is not. As with all things, LLMs have their use. Unfortunately, they are slightly overhyped and the tech is very resource hungry, contributing to environmental and societal problems in at least the USA, probably everywhere to at least some extent.
  The hope is, ofcourse, that the same tech will help alleviate those problems in turn. Time will tell who's right.
- ...you are in a technology community? They're barely defending anything either, just a reasonable take about people saying the same thing about earlier technologies.
- Clearly their mind has been taken over by the machine /s
  Baffling comment tbh

I don't refute the findings but I would like to mention: without AI, I wasn't going to be writing anything at all. I'd have let it go and dealt with the consequences. This way at least I'm doing something rather than nothing.

I'm not advocating for academic dishonesty of course, I'm only saying it doesn't look like they bothered to look at the issue from the angle of:

"What if the subject was planning on doing nothing at all and the AI enabled the them to expend the bare minimum of effort they otherwise would have avoided?"

I would argue that if you used AI, you still haven’t done any writing.
I don’t think you can definitely say that you wouldn’t have done it anyway. That’s speculative based on a theoretical situation.
It’s possible you might have been moved to write if AI never existed, maybe not. But whatever you do write without AI is actually something you made, good or bad. LLM output isn’t.
Could you expand with an example because what you said is too vague to really extract any point from. I'd argue that if it gives you wrong information, doing something wrong is worse than doing nothing.
- doing something wrong is worse than doing nothing.
  Is this a general statement right? Try to forget about context then and read that again 😅
  I actually think the moments when AI goes wrong are the moments that stimulate you and make you realize what you're doing and what you want to achieve better. And when you do subsequent prompts to fix the issue, you essentially do problem solving on figuring out what to ask to make it do the exact thing you want. And it's never going to be always right, simply because most of cases of it being wrong is you not providing enough details about what you actually want. So step-by-step AI usage with clarifications and fixes is always going to be brain-stimulating problem solving process.
sad that people knee jerk downvote you, but i agree. i think there is definitely a productive use case for AI if it helps you get started learning new things.
It helped me a ton this summer learn gardening basics and pick out local plants which are now feeding local pollinators. That is something i never had the motivation to tackle from scratch even though i knew i should.
- It helped me a ton this summer learn gardening basics and pick out local plants which are now feeding local pollinators. That is something i never had the motivation to tackle from scratch even though i knew i should.
  Given the track record of some models, I'd question the accuracy of the information it gave you. I would have recommended consulting traditional sources.
- Saw you down-voted and wanted to advise that I am glad you went on to learn some things you had been meaning to, that alone makes the experiment worthwhile as discipline is a rare enough beast. To be clear I myself have a Claude subscription that is about to lapse, and find the article unfortunately spot on. I feel fortunate to have moved away from LLMs naturally.
You haven't done anything, though. If you're getting to the point where you are doing actual work instead of letting the AI do it for you, then congratulations, you've learned some writing skills. It would probably be more effective to use some non-ai methods to learn as well though.
If you're doing this solely to produce output, then sure, go ahead. But if you want good output, or output that actually reflects your perspective, or the skills to do it yourself, you've gotta do it the hard way.
I'm in the same boat with many things I'm using AI for. I would never write natpmpc port-forwarding demons, I would never create my own DIY VPN, etc, if I had to do this all by myself. Not because I can't, but because I don't enjoy spending my time diving into tons of manuals for various utilities, protocols, OS level stuff, networking, etc. I would simply give up and use some premade solutions. But with AI, I was able to get it all done while also quickly getting to know some surface-level things about all of this stuff myself.
- A diy VPN is exactly the disaster scenario of vibe coding.
- a custom VPN without security minded planning and knowledge? that sounds like a disaster.
  surely you could do other things that have more impact for yourself, still with computers. use wireguard and spend the time with setting up your services and network security.
  and, port forwarding.. I don't know where are you running that, but linux iptables can do that too, in the kernel, with better performance.

relying on AI makes people stupid?

Who knew?

And using a calculator isn’t as engaging for your brain as manually working the problem. What’s your point?

Seems like you've made the point succinctly.
Don't lean on a calculator if you want to develop your math skills. Don't lean on an AI if you want to develop general cognition.
- I don't think this is a fair comparison because arithmetic is a very small and almost inconsequential skill to develop within the framework of mathematics. Any human that doesn't have severe learning disabilities will be able to develop a sufficient baseline of arithmetic skills.
  The really useful aspects of math are things like how to think quantitatively. How to formulate a problem mathematically. How to manipulate mathematical expressions in order to reach a solution. For the most part these are not things that calculators do for you. In some cases reaching for a calculator may actually be a distraction from making real progress on the problem. In other cases calculators can be a useful tool for learning and building your intuition - graphing calculators are especially useful for this.
  The difference with LLMs is that we are being led to believe that LLMs are sufficient to solve your problems for you, from start to finish. In the past students who develop a reflex to reach for a calculator when they don't know how to solve a problem were thwarted by the fact that the calculator won't actually solve it for them. Nowadays students develop that reflex and reach for an LLM instead, and now they can walk away with the belief that the LLM is really solving their problems, which creates both a dependency and a misunderstanding of what LLMs are really suited to do for them.
  I'd be a lot less bothered if LLMs were made to provide guidance to students, a la the Socratic method: posing leading questions to the students and helping them to think along the right tracks. That might also help mitigate the fact that LLMs don't reliably know the answers: if the user is presented with a leading question instead of an answer then they're still left with the responsibility of investigating and validating.
  But that doesn't leave users with a sense of immediate gratification which makes it less marketable and therefore less opportunity to profit...
- Don't lean on an AI if you want to develop general ~~cognition~~ essay writing skills.
  Sorry the study only examined the ability to respond to SAT writing prompts, not general cognitive abilities. Further, they showed that the ones who used an AI just went back to "normal" levels of ability when they had to write it on their own.
Yeah, I went over there with ideas that it was grandiose and not peer-reviewed. Turns out it's just a cherry-picked title.
If you use an AI assistant to write a paper, you don't learn any more from the process than you do from reading someone else's paper. You don't think about it deeply and come up with your own points and principles. It's pretty straightforward.
But just like calculators, once you understand the underlying math, unless math is your thing, you don't generally go back and do it all by hand because it's a waste of time.
At some point, we'll need to stop using long-form papers to gauge someone's acumen in a particular subject. I suspect you'll be given questions in real time and need to respond to them on video with your best guesses to prove you're not just reading it from a prompt.
It’s important to know these things as fact instead of vibes and hunches.
- Sure, and it’s important to know how to perform math functions without a calculator. But once you learn it, and move on to something more advanced or day-to-day work, you use the calculator.
You better not read audiobooks or learn form videos either. That's pure brianrot. Too easy.
- Look at this lazy fucker learning trig from someone else, instead of creating it from scratch!

Anyone who doubts this should ask their parents how many phone numbers they used to remember.

In a few years there'll be people who've forgotten how to have a conversation.

I don't see how that's any indicator of cognitive decline.
Also people had notebooks for ages. The reason they remembered phone numbers wasn't necessity, but that you had to manually dial them every time.
- And now, since you are the father of writing, your affection for it has made you describe its effects as the opposite of what they really are. In fact, [writing] will introduce forgetfulness into the soul of those who learn it: they will not practice using their memory because they will put their trust in writing, which is external and depends on signs that belong to others, instead of trying to remember from the inside, completely on their own. You have not discovered a potion for remembering, but for reminding; you provide your students with the appearance of wisdom, not with its reality. Your invention will enable them to hear many things without being properly taught, and they will imagine that they have come to know much while for the most part they will know nothing. And they will be difficult to get along with, since they will merely appear to be wise instead of really being so.
  —a story told by Socrates, according to his student Plato
The other day I saw someone ask ChatGPT how long it would take to perform 1.5 million instances of a given task, if each instance took one minute. Mfs cannot even divide 1.5 million minutes by 60 to get get 25,000 hours, then by 24 to get 1,041 days. Pretty soon these people will be incapable of writing a full sentence without ChatGPT's input
Edit to add: divide by 365.25 to get 2.85 years. Anyone who can tell me how many months that is without asking an LLM gets a free cookie emoji
- I want a free cookie emoji!
  I didn't ask an LLM, no, I asked Wikipedia:
  The mean month-length in the Gregorian calendar is 30.436875 days.
  Edit: but since I already knew a year is 365.2425 I could, of course, have divided that by the 12 months of a year to get that number.
  So,
  1041 ÷ 30.436875 ≈ 34 months and...
  0.2019343313 × 30.436875 ≈ 6 days and...
  0.146249999987 × 24 ≈ 3 hours and...
  0.509999999688 × 60 ≈ 30 minutes and...
  0.59999998128 × 60 ≈ 35 seconds and...
  0.9999988768 × 1000 ≈ 999 milliseconds and
  0.9999988768 × 1000000 ≈ 999999 nanoseconds
  34 months + 6d 3h 30m 35s 999ms 999999 ns (or we could call it 36s...)
  Edit: 34 months is better known as 2 years and 10 months.
- I swear the companies hard code solutions for weird edge cases so their investors are fooled into believing that their LLMs are getting smarter.
- You forgot doing the years, which is a bit trickier if we take into account the leap years.
  According to the Gregorian calendar, every fourth year is a leap year unless it's divisible by 100 – except those divisible by 400 which are leap years anyway. Hence, the average length of one year (over 400 years) must be:
  365 + 1⁄4 − 1⁄100 + 1⁄400 = 365.2425 days
  So,
  1041 / 365.2425 ≈ 2.85 years
  Or 2 years and...
  0.850161194275 × 365.2425 ≈ 310 days and...
  0.514999999987 × 24 ≈ 12 hours and...
  0.359999999688 × 60 ≈ 21 minutes and...
  0.59999998128 × 60 ≈ 36 seconds
  1041 days is just about 2y 310d 12h 21m 36s
  Wtf, how did we go from 1041 whole days to fractions of a day? Damn leap years!
  Had we not been accounting for them, we would have had 2 years and...
  0.852054794521 × 365 = 311.000000000165 days
  Or simply 2y 311d if we just ignore that tiny rounding error or use fewer decimals.
- Rough estimate using 30 days as average month would be ~35 months (1050 = 35×30). The average month is a tad longer than 30 days, but I don't know exactly how much. Without a calculator, I'd guess the total result is closer to 34.5. Just using my own brain, this is as far as I get.
  Now, adding a calculator to my toolset, the average month is 365.2425 d / 12 m = 30.4377 d/m. The total result comes out to about 34.2, so I overestimated a little.
  Also, the total time is 1041.66... which would be more correctly rounded to 1042, but has negligible impact on the redult.
  Edit: I saw someone else went even harder on this, but for early morning performance, I'm satisfied with my work
People don't memorize phone numbers anymore? Why not? Dialing is so much quicker than searching your contacts for the right person.
- This is the furthest thing from my experience lol I can type 2 letters in my phone, see the right name and press call. I haven’t memorised a phone number since before the year 2000* (*hyperbole)
They’ll have forgotten how to remember anything.
I still remember all my family’s phone numbers from when I was a kid growing up In WV in the 70s
I currently have my wife’s number memorized and that’s it. Not my mom, my kids, friends, anybody. I just don’t have to. It’s all in my phone.
But I’m also of the opinion that NOT having this info in my head has freed it up for more important things. Like memes and cat videos 🤣
But seriously, I don’t think this tool, and AI is just a tool, is dumbing me down. Yes I think about certain things less, but it allows me to ask different or better questions, and just learn differently. I don’t necessarily trust everything it spits out, I double check all code it produces, etc. It’s very good at explaining things or providing other examples. Since I’m older, I’ve heard similar arguments about TV and/or the Internet. LLMs are a very interesting tool that have good and bad uses. They are not intelligent, at least not yet, and are not the solution to everything technical. They are very resource intensive and should be used much more judiciously then the currently are.
Ultimately it boils down to if you’re lazy, this allows you to be more lazy. If you don’t want to continue learning and just rely on it, you are gonna have a bad time. Be skeptical, questioning, employee critical thinking, take in information from lots of sources, and in the end you will be fine. That is unless it becomes sentient and wipes us all out.

The abstract seems to suggest that in the long run you'll out perform those prompt engineers.
- How does it suggest that?
Thing is, that "trivial question asking" is part of what causes this phenomenon
what should we do then?
i also personally wouldn’t use AI at all if I didn’t have to compete with all these prompt engineers and their brainless speedy deployments
Gotta argue that your more methodical and rigorous deployment strategy is more cost efficient than guys cranking out big ridden releases.
If your boss refuses to see it, you either go with the flow or look for a new job (or unionize).
- I'm not really worried about competing with the vibe coders. At least on my team, those guys tend to ship more bugs, which causes the fire alarm to go off later.
  I'd rather build a reputation of being a little slower, but more stable and higher quality. I want people to think, "Ah, nice. Paequ2 just merged his code. We're saved." instead of, "Shit. Paequ2 just merged. Please nothing break..."
  Also, those guys don't really seem to be closing tickets faster than me. Typing words is just one small part of being a programmer.

I mean, that’s not surprising.

16 hours after posting it I am editing this post (as well as the two other cross-posts I made of it) to link to MIT’s page about the study instead.

Better late than never. Good catch.

Thank you for providing a better Source and editing the post!

Ok, if the ai knows

Been vibe coding hard for a new project this past week. It's been working really well but I feel like I watched a bunch of TV. Like it's passive enough like I'm flipping through channel, paying a little attention and then going to the next.

Where as coding it myself would engage my brain and it might feel like reading.

It's bizarre because I've never had this experience before.

Are history teachers wasting their time?

Don’t worry scro

Good article.