Are we offloading too much of our thinking to AI?

I don't know if this is a good framing. "Too much" is subjective, and every heavy AI user will assert that they're just unlocking their potential, that calculators didn't make us dumber, etc.

But to latch onto the calculator argument: if you outsource adding numbers to a calculator, you're still you. On the flip side, if you use an LLM do most of your thinking, what's left? We have people here who use LLMs to raise their children, to manage relationships, to design products. So what's your unique contribution to this world - is it the prompt you once wrote? You're standing in front of a token-generating machine, pulling a lever, sometimes receiving gifts. Is that your edge, your unique experience, your purpose in life?

Many LLM maximalists say they use the tech to learn new things, but to what effect? Are you going to apply that knowledge of physics or computer science yourself, or will you just prompt the LLM again?

In my mind, it's pretty simple: I'm a human, LLMs are not. If a human writes a novel, it's inherently worth more because it's hard-earned and anchored to experiences we share. I want to support that. And I want to be a human who can write novels, the old-fashioned way. I'm not good at lifting weights or running, so my thinking is the only thing I have.

I know that the common refrain is “think of yourself as a manager now” but I’ve actually taken the opposite approach and have been telling anyone I train the same.

Diving deeper into technical understanding makes more sense to me at this point both as a way to make yourself more useful in the age of AI and also to use AI more effectively.

I regularly tell the kids to grab a text book on a subject that interests them and I do the same.

I’m willing to bet deep understanding is going to become a commodity soon.

It will be interesting to see when ads and propaganda are seamlessly (but purposefully) integrated into the output of LLMs. Who will decide what you should want or think? Will you even notice?

As for now, autocomplete is only as good as the training data. Once humanity collectively stop being autonomous beings and generating novel ideas, it all comes to a halt. LLM suggested ideas and preferences are nothing more than some mashup/average of what came before. The ability to actually think may become a rare treasure.

> What are we automating? Human work or human agency? Human tasks or human thinking?

I find it's so easy to convince oneself they're doing the former when it's increasingly the latter. The thinking part is so often provided by default by the models, or is a single prompt away. The thoughts are so syntactically (though not stylistically) perfect that it's difficult to ignore them and reason greenfield.

What's the solution? Given how keen models are to short-circuit the thinking process it could be the only solution is to silo off tasks/ideas. Choosing which mental tasks to silo off is itself incredibly difficult especially when there's a pressure to deliver rapidly (and in quantity) on those tasks.

When I use a calculator, I atleast try to get with in a few digits of what I think the anwser is in my head. Mostly since when I was younger I had a very passionate teacher about how much slower everyone is now because of calculators on simple math. I just apply the same thing with LLMs, just try and think of how and what I would have said and see how close I was. Only thing I change is I don't trust the anwsers and accept some nuance in the given context. It's a double edge sword because then I crash out over it more than if I don't. When it over and under explaining the wrong sections or when it gets to an objectively terrible solution that technically anwsers the question. It feels like a student trying to get brownie points and/or give fluffed anwsers for the sake of not leaving anything blank on a test.

The question presumes that most of us are "thinking" in the first place, when in actuality, most of us are just acting according to the patterns that have emerged from our encounters with the thoughts of others. We generally adopt them and/or try to hallucinate coherence when they conflict. Very few people actually "think". It's hard work and takes time. We neither have (take) the time nor are we particularly motivated to put in the work because the patterns we have learned from others are useful enough to achieve the low goals we set for ourselves.

IOW - modern AI is simply an extension of the lack of thinking that characterizes the modern life... It just does it faster and uses a hulluva lot more energy.

Clearly "mechanical" thinking has been automated, and you are better off, if not forced to outsource to AI. Humans have the biological needs, consciousness, taste and imagination, that's what we are left with thus far.

And just like people going to the gym to exercise their otherwise economically useless bodies, same thing will happen with the mind.

Folks, get over it.

I thought it was a myth until the junior developer on my own team responded with "I don't know" to a question about why he made a certain computation during a design review. Because the (wrong) computation was fully AI generated and he couldn't even tell the difference.

Most people don't use AI to learn new stuff. They use it to do "the job" for them and they don't even understand the result. What is the point of a person if they don't bring any value to the table other than being a "resource" to generate prompts?

Maybe it's a way of perspective. I adore to use AI to actually learn, so I don't feel like I offload thinking. I use AI to do all the research which earlier I did manual through google search. I still make my own decisions, I just let Claude spoon feed me all infos I want and need. Feels great man

I am increasingly finding my consulting work to be orientated around clearing up after people who outsourced their thinking to AI.

I'm seeing some incredibly dumb stuff: researchers spending months on Claude trying to do insane deduplication, unrelated to their research question, using regex; whole research methodologies YOLO'd out of ChatGPT.

The results invariably chaotic, resulting in huge amounts of stress and wasted time.

Non-technical people are treating LLMs like an oracle, making big assumptions and decisions with little regard for their implications, because their clanker told them to.

It's scary out there. The lack of critical thinking I'm seeing in some of these projects is horrific. Not unique to the post-AI era, certainly, but on a whole new level. Bad things are undoubtedly happening everywhere, right now, because someone's just like "let's ask Claude".

This is actually a very hard question to answer. If you’re truly AGI-pilled and believe in this progress continuing, then you basically have to contend with a scenario where AI is better than you at doing anything by an order of magnitude. And doing things suboptimally just for the sake of some notion of human independence doesn’t feel right to me.

Perhaps the only way forward will be if we figure out how to merge with the AIs so we can keep up. Otherwise, a soma-filled world likely awaits. And unlike Brave New World, I think it might actually be a lot more pleasant, but still one with a different set of tradeoffs.

I'm going to say: yes.

I think partially it is because of the amount of data you can get out of an LLM and because it looks pretty good, a lot of people treat it as authoritative.

This means they send it around and then other people have to go through the work of actually validating it before being able to act on it.

So what this really means is that the person you go the information from offloaded their thinking to the AI, while cannibalizing on yours.

I've been doing computer stuff all my life. I am learning the construction trades now. LLMs help you learn the rules for doing weird and specific stuff in the trades. One of my favorite LLM prompts is for plumbing tasks. For example: I have 1" pex b and I need to convert to a typical 1/4" RO connection line, what fittings on supplyhouse.com should I buy? So much easier than standing in Home Depot in the plumbing aisle for 25 minutes staring at the wall of fittings.

I think in the software trade you will definitely use your brain less. But in other trades, it removes the time sucks and gets you back to work.

The right question to ask is if it is enabling us to do more interesting things or take on harder or bigger problems that seemed too daunting before. That's what all delegation of tasks (to other humans or machines) have enabled humans to do – scale.

Whatever creativity/thinking/effort bandwidth that's available will now get shifted to a different place in the problem-solving effort bottleneck.

That's the hallmark of any delegation being effective. Do we see that happening with AI tools? Personally, I do see that working for me. Is it as good as the hype makes it to be or I wish it to be? maybe not, yet, for me. But that's the case with most things in life.

One major issue that the author ignores is that while we’re all having AI analyze our conversations or when we use it instead of search, there’s a chance it will provide an “answer” that is not correct, and literally drummed up out of thin air, and not even in the source material it’s “synthesizing”.

The article takes a position that assumes hallucinations do not occur, and then posits from that stance the question as to whether we rely on AI too much. We should be taking a step back before even asking that question and focusing on the part where AI does invent answers whole-cloth.

I have no idea what people are doing. I am thinking more about harder problems than ever before.

It goes like: "Here is this thing I wonder about", and the LLM is like "Yeah sure, consider these things that are super related to what you are doing, that you probably know nothing about yet (but you know... if you are interested...)".

And that goes in any direction, for any depth. Anything that is made trivial now, is just replaced by something more consequential a level or two higher. You can just get much better at things that matter more.

I'm not personally, since I don't use GenAI at all.

Especially given the comments I see here and on other tech and programming forums, I hate the direction things are going.

I still have some hope this will all fade, but the damage done will be worse the longer it goes on, I think.

Information is seldom presented in a way that makes sense to me. That’s not quite right. Let me attempt to explain.

I want certain answers that the docs and the code are not giving me yet. Nothing is more irritating than working through a tutorial on a new framework and then throwing all that work out because that’s not really how one should use the framework. Nothing is more frustrating than having to get through a treatise on why this framework is The Solution before I can actually see code that uses the solution. And it’s beyond annoying when this End All Be All framework has a glaring omission that’s not obvious until you’ve built large amounts of your project on top of it.

Hand the docs and the example code to the LLM, and now I can get answers. “How can I do X?” Example code. “Then I need Z” Modified code. “How is this going to handle Q?” Explanation. “That doesn’t seem quite right. Give me a reference to the doc or code showing this.” Links.

Great, in 15 min, I have learned what I need to know, I can see that this solves a problem that I have, and I have discovered that I need an implementation of S to complement this solution.

That is usefulness. And it requires experience.

The post is illustrated by a picture of handwritten notes, like that was supposed to shock the reader or something. I find this aesthetic tiring, and it usually comes from AI-maxxers. To me its saying: look at this quaint relic of the past, bereft of day-to-day utility, replaced by superior technology. Its life is now only as a symbol of a time where people actually used their brains.

Humans have been offloading thinking to something/some entity forever. From gods to influencers.

AI is the current popular way (at least in these circles) and if it's too much or too little it might not matter. What will matter is if this offloading is making people unhappy and having any negative impact in civilization. I have no idea

This is exactly the lens I use myself. I write AI software and I use it in my development process, but I try to use the AI to do things that don't remove my agency, but extend my capabilities: - Debug things. It knows way more than I do in many areas and it sees things I will miss. If I'm struggling to find the answer, maybe it will succeed. - Review things. It has a wealth of experience I couldn't possibly have. Ask it to critique my work and provide an alternative perspective that I can't provide myself - Implement a design. I have already gone through the thoughtful engineering to decide what to do and how to do it. The rest amounts to translating pseudocode to the programming language. Let it type what I would have typed anyway and save me the hassle of typos, looking up function and parameter names, and other such mechanical details. Let me use that time and mental effort to better consider my design, try more alternatives, or build more things, providing more value overall. - Suggest ideas. Even as a 20 year professional, there are things I don't know or haven't considered. Is there a newer, faster, or more maintainable way of doing this? Is what I wrote clear to anyone other than myself? Before AI, I would ask coworkers, search the web, or reference other sources. Now I can get an immediate suggestion from something with tremendous knowledge at almost no cost. It's full up to be to consider what it suggests, further explore the used and learn about them - I don't take the AI at its word and let it decide what is best for me. But I do use it to gain perspective and explore alternatives.

There are some common traits about the thighs I use AI for. They are this that I either couldn't possibly do myself (because I'm biased, or unfamiliar, or have no access to the expertise) or that I would spend a lot of time while having little agency (mechanical translation). I am not replacing learning, thinking, or deciding. I think this is the key difference.

This is something I see and feel everyday. And it’s very frustrating and annoying.

For example I send some doc asking for a feedback and someone without reading it generate a feedback with llm with so much ambiguity that I have to get back and wait couple of more days to get a reply.

One of the most silliest thing I see is a middle manager feeding Microsoft planner to Claude to generate a report and generate future steps and sometimes it makes no sense what he present couple of weeks ago because what he present today is contradicting to the one before.

At this point I feel it’s cheaper to replace them with AI. They are just physical vessels for AI.

It’s just not that maybe they were not good enough. But now they just fully depend on AI.

Our brains take time to ingest things and learn and using an AI tool to make decisions when taking in vast inputs of data requires then an overview of those decisions and understanding to some degree of care. If you don't take the time to go over these decisions and understand them and weigh them and course correct as needed, then you are offloading too much. You will become this approver buffer in between claude and nothing more if you don't have this methodical checking and understanding.

What is frightening is with something like neuralink that in a future hypothetical time would have very fast capability to keep informed and advised, you could be a zombie decision maker and nobody would really be able to tell. Even when you were pressed to why you made the decision, it's just another AI response, it's like a con artist or imposter dream scenario.

I noticed that atm, before these crazy hypotheticals potentially happen, the people that seem to take the time to understand things deeper are still way more valuable than those that just use tools more than not. Its obvious atm due to the lag in time and the way people respond in meetings, at least for now. :)

I think this is absolutely an issue.

The rise of knowledge work made many people far less physically active because moving one's body was no longer a given part of one's job. This led to a lot of people (who assumed sports was exercise on top of one's work, not the only source of exercise) moving very little. This meant we needed to rediscover the importance of exercise as a pillar of health.

I think something similar will happen with knowledge work, where we have to do a lot less cognitive exercise due to AI (as well as the decline of reading and rise of short-form video), which will likely lead to eventual issues and subsequently, a rise in activities designed to replicate the cognitive exercise work used to provide.

I saw a post the other day pulling a Frank Herbert quote from Dune about men offloading their thinking to machines and being controlled by the men who controlled the machines.

Somebody asked an AI how to interpret it.

To answer the title's question directly? Yes, I think so. Or at least that it is a legitimate hazard.

I think the analogy to hyper-palatable, calorific food works well:

Humans adapted for a world with too little food. Then once there was too much, obesity and overeating became a problem for the first time. Self discipline is the cost we have to pay for this kind of abundance.

We now have a general-purpose way to offload mental effort, and are discovering in real-time the negative consequences of that.

I use AI for coding, but I feel I've moved past the honey-moon period and am now learning how to use it in a way that is not a detriment. If I care about the work I'm doing I don't want the AI to do it for me, even if it could. Deciding what work I want/should be doing myself vs what can be delegated is a new skill I, and I believe we all need to learn.

Maybe, let me ask my coding agent what he thinks about this.

Just like calculators became crutches for school children memorising their times tables and our phones contacts list a crutch for memorising phone numbers, I see AI becoming a crutch for many, in all the areas we seem important and humanity has spent years refining (communication, thinking, writing, etc)

It's a well done and thought-provoking article.

The reality is that most humans do very little actual thinking of their own anyway, and, if you believe that what LLMs produce constitutes a form of intelligence, it does seem "more intelligent" than most humans.

So: is outsourcing thinking a net improvement for a majority of users?

I use several models, daily, and they seem "reasonably conditioned" that they are only input to my thinking and not "my thinking". I correct them constantly; they are wrong (in reasoning/logic, in actual facts) frequently. They are demonstrably "not smarter" than I am. And yet I know many people who can "do more" with them as a "thinking" tool. I can say that "the problem" is they can't spot the errors, but they can't or won't do that in their ordinary lives, either, so, again, is it a net improvement for them?

Interesting times and all that.

Almost definitely Id think - I think we have to treat it something like power tools / chairs and a gym.

AI makes me massively more productive as a quant, and more creative in the sense that it can often find calculations I don't know how to start, BUT the flip side of that is that I can also feel my skills atrophy and as such am trying to make sure I do maths exercises and so on. I don't worry about programming skills because programming isn't about code.

I do feel like I'm offloading thinking to an AI, but I think that's a good thing. I envision a world where users and AI are aligned without corporate interference. AI lets me offload things that I don't need to know and frees up my brain to push farther than I could before. At least that's how it feels to me.

Related discussion: Outsourcing thinking (270 points, 5 months ago) https://news.ycombinator.com/item?id=46840865

I'm probably not the only person that types math problems into google search because I'm too lazy to open up a calculator (or do the math myself). So yes, given I do that, I probably am susceptible to offloading some of my thinking to AI.

Are CEOs outsourcing too much of their thinking to their employees?

I find I'm not thinking less per say, just thinking about different things. Maybe you could argue there are CEOs who get too far out of touch with the reality on the ground and should get more directly involved in the work. However, I don't know how well one could argue that the CEO should do all of the work.

I see at least the current iteration LLMs and harnesses as me managing and coordinating them and thinking differently, not less.

The value of thinking in language is going down, the value of acting on your dreams is going up

Some of us are, yes.

I've noticed it when interviewing interns. A surprising number seem unable to think on their feet or solve problems without immediately reaching for chatgpt. I don't necessarily expect you to be able to solve a problem entirely without tools, but you should be able to give me the outline of how to go about something and why you would go that way.

After all, if you are just going to spit out AI, I will just get AI to do your job...

It really depends on which angle you look at it. Is it purely to meet a business goal? Or is it also for personal growth? I think it's a mix of both, but for me it's always important that I engage mentally with the process, learn something, and solve puzzles, even if that involves letting the AI take care of the coding, which is an abstraction. You could still code and not think creatively.

Decisions are special things. One of the golden rules of life is that a person (or entity) making decisions is somehow impacted or otherwise getting feedback on the repercussions of those decisions.

When you cognitively surrender to AI, or to another person (be it a leader/manager, or a subordinate/report), you are asking for trouble.

This is the kind of headline I would expect from one of those news that appear as you progress in cookie clicker.

I don't think we are offloading thinking to AI. We just started to use it. AI is a tool useful to write text, for a fast searching, for the boring work. Personally I don't take the first answer, I like to challenge, to ask why, to tell it's wrong.

as a six-figure salary software engineer haver I find that a lot of colleagues take their cushy jobs for granted.

these are the people who are now first against the wall when the revolution comes.

why would I want to hire you to work the three hours you feel like working per day (per week?) when I can have AI with a deeper knowledge set available 24x7?

Yes, I think we are . And because of super competition, most of us are trying to make AI (agentic) do the work.

Personally, I use AI to learn more about Backend Engineering actually, so it's fine for me. Beside I can also use AI to suggest and it's me verifying the idea so that's a no for me

How much of the thinking is involved in asking the right question, versus coming to the correct answer? I don't have a real answer to that but it does seem to be worth considering.

imo no way

But this varies from person to person

Some of us overthink already and offloading to AI just enables us to overthink more in other directions than we would if we didn't have ai

Ironically I just caught myself offloading the thinking about this article to the comment section before I read it

Having a very dangerous AI standoff at work, where people are debating wether or not to use a particular connection pooling / threading strategy to fix a production issue, and everyone is unqualified to answer and is instead arguing what their agent said.

They are just straight up admitting they don't know anything, and advocate fiercely for their agent's recommendation.

No one cares, no one tries to stop this behavior. It's seen as good, apparently. I admitted that I don't know enough to have an opinion at the moment, I certainly don't know how to judge the contradictory opinions of multiple frontier AIs, and I fear that just made me look incompetent.

This is a load-bearing title.

The answer to this question is: Politicians, not you!

Perhaps the question to ask is: who is making all of the final decisions for the things that really matter to you in your life?

No direct democracy, just people deciding for you. You can choose once every four years. Are we surprised of how easily we delegate decisions? May be AI can do it better

You’d lose your ability to interview at a different workplace.