The Impact of Generative AI on Critical Thinking: Self-Reported (2025)
dl.acm.org
dl.acm.org
You get more done without realising you're learning less.
Unfortunately, I think many folks now just don't place any value on learning or knowing things because they can "just ask AI".
That's probably my biggest fear with AI is that people just don't see any value to knowing things anymore. And quite ironically, AI is amazing for learning new things.
People ABSOLUTELY place a ton of value on. Learning new things? You yourself JUST called this out but somehow “many folks” don’t place value on learning new things? What’s many? 20? 50? A million? Ten million?
Contemptible suppositions.
Now that it can be outsourced,we will slowly lose this ability that we barely had.
Early LLMs spit out easily identifiable nonsense. Now, it's much more subtle, but it's still absolutely everywhere, even with the best models.
You have to constantly be on your toes to detect it. Some people won't, but I suspect that's mostly the people who were never capable of critical thinking anyways. I'd say it's an upgrade if those people blindly trust an AI over some random person on the internet like they did previously.
They all do.. something[1]. They write visually great looking papers. Even good on a quick read.
But once I try porting their proposed Section ideas into the paper, I notice that for each sentence I either need to do a complete rewrite, or a deep dive into it (plus the literature), to confirm or delete that idea/sentence.
If it were my field, it would be much simpler to just write everything directly, rather than attack every sentence of LLMs. As it's not my field, it's a decent way to understand the nuances and learn, but I get increasingly disillusioned by AI writing skills.
[1]: Old and new Gemini models are too categorical, everything is a breakthrough and a marvel; Sol/Astra gets bogged down with too many details, and "it's X, not Y" garbage. For some reason, it's very defending against potential referee questions which don't add anything to the main idea; Opus 5.5 or Fable are basically the same. Maybe it's their way of anti-hallucination..
Authors analyzed ~300 workers doing ~1000 tasks. They collected self-reported measure of using critical thinking [yes/no], and estimate of cognitive effort of doing the task. Authors also measured workers' confidence in themselves doing the task or the AI doing the task.
They found: if someone is confident that AI can do the task, they report using critical thinking less, even though they think the task is easy. When someone is confident in doing the task and in evaluating AI responses, they use critical thinking more, even though they think the task is harder.
I think the paper is saying: low-domain expertise (read: confidence) + unrestricted AI tooling can eliminate the human from the loop. The fallout is better automation or unreliable systems. We need to resolve this ambiguity.
The question is: should low-domain expertise be coupled with more restricted AI tooling as a critical thinking forcing function? Are schools/workplaces right in restricting AI so students/new workers can reach the point of high domain expertise?
Amount of "critical thinking" is basically amount of oversight, no? What's interesting is when the amount of oversight is inappropriate, leading to overconfidence in bad results (or, if too much oversight, lack of efficiency).
This paper does not seem to discuss results and correctness at all, ONLY the amount of critical thinking.
Critical thinking is actually something that people exercise (or don't) all the time. It is a skill and practice that is inarguably valuable in all situations. This paper is not about the effects of critical thinking on "AI results", but the exact inverse of that.
Do you value the capacity for critical thinking in yourself? In others? If so, the effect reported by this paper is probably of interest to you.
If I want to know when Obama was born, and check wikipedia and believe what it says, by their measure I am practicing less critical thinking than if I found independent verification. Independent verification is the main way they measured critical thinking. They are NOT measuring that I thought about how much I could trust the result, and decided that yes I could.
And, well, yes, they're right. But for most purposes, wikipedia is pretty credible for something like that, and it's just a waste of time to follow up on it. What's important is that my calibration is correct.
I'm not even saying I disagree with the idea that there is likely a problem here, with people offloading thinking. But I'm thinking critically about this particular study, which seems totally useless.
The example's not relevant; the two things are not analogous. Moreover your point only works if one accepts your conclusion that:
Independent verification is the main way they measured critical thinking
But they also reported assessing things about the content, like style/personality, logical soundness, relevance to a topic, or suitability for a specific purpose.
Studies don't need to justify themselves with concrete conclusions; gathering the data for future use is a fine goal.
Regardless, confidence in oneself is hardly measured, self report is likely as good as it gets.
and here is a direct link to the pdf without all the rigamarole:
https://www.microsoft.com/en-us/research/wp-content/uploads/...
https://www.rxjourney.net/how-artificial-intelligence-ai-is-...
To try, go and type a weakly argued point into Claude and watch it take you apart. Try these:
People only post on hn for the pathetic dopamine hit of upvotes
Hn comments are getting weaker
James Acaster is the best commedian of the last 200 years.
Your challenge is to turn it into an argument strong enough to not get an instant rejection
I agree that AI is a useful tool, but it's arguably a very easy tool that people can offload their thought process to - which you should not be doing.
AI has an obvious issue with incorporating how happy users are in its responses. It LOVES to "take your argument seriously", push back lightly, and then tell you you are really smart and make amazing points. If you start with an argument out of left field, you will shift the Overton window very easily.
I’ve been burning spare weekly quota on fringe-adjacent physics. I often copy-paste the LaTeX format “paper” back and forth for “peer review of a friend’s work”.
Gemini: “Your friend’s insight hits the nail right on the head!”
Opus: “Tell your friend to stop wasting his time now, it’s more kind than to keep indulging him.”
Brutal.
If students in the public k-12 system were actually being taught critical thinking, the schools would be the targets of strategic bombing by the US military.
Everyone I know who is working closely with LLMs is doing some of the most acute critical thinking of their lives these days.
I'm very skeptical of these kinds of takes; I'm reminded that Socrates railed against books on the basis that they were going to rot the brains of the youth.
According to them.
Yes. In contrast to you, I really think many of my coworkers seem to get dumber by the day. Not all of them yet, fortunately, but many.
I think the problem with these sorts of studies is that the industry is moving so fast, and by the time your paper gets peer reviewed and published it is no longer relevant.
In this paper, the participants are using mostly ChatGPT - well, there's a HUGE difference in free chatGPT, gpt 6, gpt 4o, gpt 3.5?
What might have been the case for say, GPT 4, is probably no longer the case with GPT 6.
My experience with Google's gemini right now makes this obvious, as I run out of Gemini 4 Astra in a conversation, and then it switches me to Gemini flash lite.
You could argue that these are just examples of laziness, but I think these are both tasks that require critical thinking, and it’s easier to hand them off to an llm that at least purports to do this for you.
I’d argue that both setting requirements, and self code review are both things that need careful attention. (What _exactly_ do you need doing? Does your work correctly do the thing that needs doing?).
I have no fundamental issue with using ai, but there’s no doubt that applying critical thinking to both these tasks directly, and to the broader question of “should I use ai for this job” should result in a different outcome.
Personally I would also not agree with your take that working with llms has allowed me to do “acute critical thinking”. It’s sped up my development time, sure, but I would argue that my evaluation of, say a new sdk now using Claude is far less detailed than it would’ve been in the past when I did it by reading docs. As a side note, I’ve been rereading Douglas Hofstadter’s GEB recently in paperback, and I’ve found a lot of joy in wrestling with paragraphs and writing notes in the margins. This is way more focussed attention than I really give anything when working with llms.
Geometry existed for centuries before Euclid's formalization. Similar for algebra before Al'Khwarizmi. People were using logic before Boole.
"External Intelligence" is not new. But we are still in the informal stage of LLM use.
Or course "everyone you know" might be doing something different, but AI is offering the same delegation possibilities for critical thinking. Do you think there won't be people who'll do the equivalent of driving into the river when using AI?