The Impact of Generative AI on Critical Thinking [pdf]

I think this study does not say what most people are taking it to say.<p>&gt; our research questions - when and how knowledge workers perceive the enaction of critical thinking when using GenAI (RQ1), and when and why do knowledge workers perceive increased&#x2F;decreased effort for critical thinking due to GenAI (RQ2)<p>This is about the application of critical thinking to AI outputs and in AI workflows. Are people cognitively lazy when some other entity hands them plausible sounding answers?<p>The answer of course is yes. If some entity gives you a good enough result, probably you aren’t going to spend much time improving it unless there is a good reason to do so. Likewise you probably aren’t going to spend a lot of time researching something that AI tells you if it sounds plausible. This is certainly a weakness, but it’s a general weakness in human cognition, and has little to do with AI in and of itself.<p>In my reading, what this study does not say, and does not set out to answer, is whether or not the use of AI makes people <i>generally less able or likely to engage in critical thinking</i> as a result of use of AI.

Study (PDF): <a href="https:&#x2F;&#x2F;www.microsoft.com&#x2F;en-us&#x2F;research&#x2F;uploads&#x2F;prod&#x2F;2025&#x2F;01&#x2F;lee_2025_ai_critical_thinking_survey.pdf" rel="nofollow">https:&#x2F;&#x2F;www.microsoft.com&#x2F;en-us&#x2F;research&#x2F;uploads&#x2F;prod&#x2F;2025&#x2F;0...</a><p>I did not link to it directly because the PDFs title - &quot;The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers&quot; - was way too long for the HN title length limit.<p>Abstract (from the PDF):<p>&gt; The rise of Generative AI (GenAI) in knowledge workflows raises questions about its impact on critical thinking skills and practices.<p>&gt; 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.<p>&gt; 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.<p>&gt; 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.

If a study comes out which demonstrates to high confidence that some extremely horrible outcome is certain with this technology, would any changes be made? Would we tell OpenAI to close shop? I don&#x27;t think, with the way our society and politics is set up, we would be able to deal with this. There&#x27;s too much money, ego, and general inertia at play to change course.

That&#x27;s a brilliant strategy to reach ASI, lowering the bar on humans.

Interesting, in my experience LLMs hallucinate so much on stuffs I know about that I instinctively challenge most of their assumptions and outputs, and found out that this kind of dialectic exchange bring the most out of the &quot;relationship&quot; so to speak, co creating something greater than the isolation of us two.<p>Relevant 2018 essay by Nicky Case «How To Become A Centaur»: <a href="https:&#x2F;&#x2F;jods.mitpress.mit.edu&#x2F;pub&#x2F;issue3-case&#x2F;release&#x2F;6" rel="nofollow">https:&#x2F;&#x2F;jods.mitpress.mit.edu&#x2F;pub&#x2F;issue3-case&#x2F;release&#x2F;6</a>

As I expected. When I got a cell phone I forgot everyone&#x27;s phone numbers, and I&#x27;m afraid if I use AI I&#x27;ll forget how to think.

A warning on this topic, in short story form: <a href="http:&#x2F;&#x2F;web.archive.org&#x2F;web&#x2F;20121008025245&#x2F;http:&#x2F;&#x2F;squid314.livejournal.com&#x2F;332946.html" rel="nofollow">http:&#x2F;&#x2F;web.archive.org&#x2F;web&#x2F;20121008025245&#x2F;http:&#x2F;&#x2F;squid314.li...</a><p>(By Scott Alexander)

Don&#x27;t exercise your body, become fat.<p>Don&#x27;t exercise your brain, become stupid.<p>LLMs give us another excuse not to exercise our brains.

I&#x27;ve thought about this myself and try to cultivate a mindset of distrust towards the LLM. It&#x27;s hard because the way it is set up is just so convincing that you regularly need to remind yourself that it might be bullshitting.<p>Really the user interface should be like &quot;here&#x27;s a system where you can enter some text and it outputs statistically generated text that might be useful to you&quot;. Of course that won&#x27;t happen because the charade (the conversational interface) is part of its success.

Any chance the link can be changed to the original paper (<a href="https:&#x2F;&#x2F;www.microsoft.com&#x2F;en-us&#x2F;research&#x2F;uploads&#x2F;prod&#x2F;2025&#x2F;01&#x2F;lee_2025_ai_critical_thinking_survey.pdf" rel="nofollow">https:&#x2F;&#x2F;www.microsoft.com&#x2F;en-us&#x2F;research&#x2F;uploads&#x2F;prod&#x2F;2025&#x2F;0...</a>) and the title changed to &quot;The Impact of Generative AI on Critical Thinking&quot;?

I have this horrible feeling that in the future people are going to walk around with AR glasses and some AI is just doing to tell people what to do all the time. They will become meat robots to do the bidding of some hallucinating model which has been finetuned with an agenda.

I want them to do the same study with managers. And maybe executives.<p>Does relying on employees kill the same critical thinking skills? Or is this a skill that has to be cultivated like every other skill?

personal related example: my 12 year old son is a terrible speller; whenever I&#x27;ve tried to encourage him to improve in it, he&#x27;s like &quot;why do I need to know how to spell, I just use grammarly&quot;.<p>And so we raise a next generation who grow up with AI and see no more value in learning a host of knowledge skills.

I agree with the study participants. Using LLMs makes me feel less capable at some cognitive tasks, but it&#x27;s hard to say whether I&#x27;ve actually declined or if use of AI has dispelled some of the Dunning-Kruger effect for me around my own skills.<p>For instance, LLMs are excellent at rewriting text in a specific style&#x2F;tone. I used to believe I was quite good at that, but LLMs always do better. Now I no longer believe I am quite good. Did I become worse at text synthesis, or did I simply become more aware of my skills and limitations?<p>This is emblematic of a broader issue with self-reported data. It&#x27;d be good to measure critical thinking skills against earlier benchmarks for a clearer picture.<p>It&#x27;s also important to not only focus on one skill, because while AI might make certain areas of our mind atrophy, others might now be more engaged. Just like when pocket calculators became popular and we all got worse at mental arithmetic but much better at applied mathematics overall. High level programming languages have made software engineers worse at comp-sci, but much better at applying it; commodification of cars made the average driver much less capable of understanding car mechanics, but much better at driving; and so on. My thesis is that human brainpower is not generally reduced by technological innovation, but changed.<p>One thing&#x27;s for sure - cognitive changes in society (and especially how they relate to technology) is an area I&#x27;d like to see more research in. I think there are a few discoveries to be made there.

Review code or waste time thinking and coding? I guess that&#x27;s the question.

Does that mean that being a manager also kills critical thinking?

I want a study on the impact of generative AI on knowledge. I fear that AI hallucinations have the potential to destroy all knowledge. For now, when an AI hallucinates, I can still check an established source to see if the fact is correct. But what happens when most established sources are themselves AI generated? There will almost be no ability to verify any fact. The very foundation of society collapses and any shared knowledge. Oh wait, that already happened with flat earthers, Covid and conspiracy theories. But AI may be the final nail in the coffin for Western civilization.

i’m surprised they published this with how much they are invested in ai…

The Greeks found that relying on writing kills memory.

People are worried AI is making us dumber. You hear it all the time. GPS wrecked our sense of direction. Spellcheck killed spelling. Now it’s AI’s turn to supposedly rot our brains.<p>It’s the same old story. New tool comes along, people freak out about what we’re “losing.” But they’re missing the point. It’s never about losing skills, it’s about shifting them. And usually, the shift is upwards.<p>Take GPS. Yeah, okay, maybe you can’t navigate with a paper map anymore. So what? Navigation isn’t about memorizing street names. It’s about getting from A to B. GPS makes that way easier, for way more people. Suddenly, everyone can explore, find their way around unfamiliar places without stress. Is that “dumber”? No, it’s just… better navigation. We optimized for the outcome, not the parlor trick of knowing all the streets by heart.<p>Same with the printing press. Before that, memory was king. Stories, knowledge – all in your head. Then books came along, and the hand-wringing started. “We’ll stop memorizing! Our minds will get soft!” Except, that’s not what happened. Books didn’t make us dumber. They democratized knowledge. Freed up our brains from rote memorization to actually think, analyze, create. We shifted from being walking libraries to… well, to being able to use libraries. Again, better.<p>Now it’s AI and coding. The worry is, AI code assistants will make us worse programmers. Maybe we won’t memorize syntax as well. Maybe we’ll lean on AI to fill in the boilerplate. Fine. So what if we do?<p>Programming isn’t about remembering every function name in some library. It’s about solving problems with code. And AI? Right now, it’s a tool to solve problems faster, more efficiently. To use it well in its current form, you need to be better at the important parts of programming:<p>- Problem Definition: You have to be crystal clear about what you want to build. Vague prompts, vague code. AI kind of forces you to think precisely.<p>- System Design: AI can write code snippets. As of right now, designing a whole system? That’s still on you. And that’s the hard part, the valuable part.<p>- Testing and Debugging: AI isn’t magic. At least, not yet. You still need to test, validate, and fix its output. Critical thinking, still essential.<p>So, yeah, maybe some brain scans will show changes. Brains are plastic. Use a muscle less, it changes. Use a new one more, it grows. Expected. But if someone’s scoring lower on some old-school coding test because they rely on AI, ask yourself: are they actually worse at building software? Or are they just working smarter? Faster? More effectively with the tools available today?<p>This isn’t about “dumbing down.” It’s about cognitive specialization. We’re offloading the stuff machines are good at – rote tasks, memorization, syntax drudgery – so we can focus on what humans are actually good at: abstraction, creativity, problem-solving at a higher level.<p>Don’t get caught up in nostalgia for obsolete skills. Focus on the outcome. Are we building better things? Are we solving harder problems? Are we moving faster in this current technological landscape? If the answer is yes, then maybe “dumber” isn’t the right word. Maybe it’s just... evolved. And who knows what’s next?<p><a href="https:&#x2F;&#x2F;tulio.org&#x2F;blog&#x2F;dumber-no-different&#x2F;" rel="nofollow">https:&#x2F;&#x2F;tulio.org&#x2F;blog&#x2F;dumber-no-different&#x2F;</a>

No shit Sherlock

Those with critical thinking skills already realize this.<p>Isn&#x27;t MS heavily involved in AIBS?

<i>&gt; Thus, knowledge workers with higher levels of trust in GenAI — generally or for specific tasks — perceive engaging in critical thinking activities to be less effortful. A possible explanation, supplemented with our qualitative analysis in RQ1 (see Section 4.3.2), is that trust and reliance on GenAI inhibit the enaction of critical thinking, i.e., users underinvest in critical thinking when using GenAI.</i><p>There is a reduction in certain kinds of effort when using any kind of tool.<p>Another possible explanation is that using LLMs, much like using a pry bar for leverage with physical energy, provides leverage for cognitive energies. Does the person using a pry bar become less physically fit than if they were removing floorboards by hand or do they just get more work done during the day?<p>I don&#x27;t think there is a problem in exploring the trade-offs for using any kind of tool or technology but the discussion needs to be balanced.<p>Personally, I use LLMs but I&#x27;d say I have a critical understanding of their limitations and when it is better to start using another tool, like a book, to learn the skills required to use the tools to achieve the tasks at hand. I will say that using a tool that provides leverage for cognitive energies allows for the cognition to be applied to other parts of the process, typically higher order.<p>For example, I spend much more time thinking about the overall architecture of the memory model in C applications.