Ask HN: Which skills are least likely to be replaced by AI?

30 points by AznHisoka ↗ HN
What kind of skills do you think will be the least likely to be replaced by AI? I'm not just talking about technical skills, but non-technical ones as well.

Such as creativity, emotional intelligence, leadership/people management, negotiation skills, motor/hand eye coordination, critical thinking skills...

110 comments

[ 2.4 ms ] story [ 108 ms ] thread
Basic communication with human interaction. Soft kills >> hard kills

AI is assistance not an replacement to a lot extent

> Soft kills >> hard kills

You mean stuff like poison? I think robots can do that, probably more effective than guns it could take over a food processing plant and change the recipe, just that guns is scarier.

I think they meant soft skills and hard skills.
Probably those that require a combination of some high level cognitive processes, tacit knowledge, and physical actuation in uncontrolled environments. For example, plumbers probably aren't going to be replaced any soon. Surgeons too. But between surgeons and plumbers, the latter will take more time to be replaced, because a construction site is less of a controlled environment than a surgery room.
You didn't give a time frame. So I would say, there are no skills that cannot be replaced by AI in an indefinite time frame.

AI might have mastered just about every skill or capability within 10 or 20 years. Some people suspect less.

Your list isn't really just skills but more generally abilities.

Emotional IQ - See the Inflection 2.5 model (Pi).

Creativity - Ask ChatGPT to help you be creative writing poetry and then creating paintings of the poems.

Leadership/management - Claude 3 or GPT4 might be better than the average manager. 70% of my bosses have been barely competent pr!cks though, so I may be biased.

Negotiation is not that complex actually. It's boils down to being willing to walk away. Leading LLMs can be good negotiators already.

Hand-eye coordination - not for every single task, but we have robots that can play ping pong, juggle, catch balls, and solve a Rubik's cube faster than humanely possible.

Critical thinking skills: put the average person off the street against Claude 3 Opus or GPT-4 Turbo and have them take a GRE or similar test. I would bet on the AI.

For your list, we are already there.

I think that "replaced" is not necessarily the best way to think about it though. People still play chess. Or run. Or Jeopardy. We still have plenty of amateur artists that just want to imitate their favorite cartoon or whatever. Even though AI can do all of that better and faster now.

But wait 10 or 20 years. There will be very realistic and fully capable simulations of humans.

> But wait 10 or 20 years. There will be very realistic and fully capable simulations of humans.

So sometime between FSD and fusion as an energy source.

This is comment is the problem with AI. You listed a bunch of things as proof that AI can replace human skills but many are wrong or at least require moving the goal post. Poetry and lyrics from AI are usually lame and formulaic.

A robot solving a Rubik's cube is less impressive than a human doing so, since Rubik's cube is basically about memorizing the algorithm.

Chatgpt might be better than your manager, but managers don't lead. Every manager I've reported to has basically had the job of primarily relaying communication from above.

Investigative skills. AI can't do anything a detective does. Formulating a theory, testing it, refining it etc.

Also quickly grabbing arbitrarily shaped objects from a pile.

I'm not disagreeing, but this answer is ironic considering how Ilya Sutskever once used detective work as a way to describe the relationship between LLMs and "understanding":

"So, I’d like to take a small detour and to give an analogy that will hopefully clarify why more accurate prediction of the next word leads to more understanding, real understanding. Let’s consider an example. Say you read a detective novel. It’s like complicated plot, a storyline, different characters, lots of events, mysteries like clues, it’s unclear. Then, let’s say that at the last page of the book, the detective has gathered all the clues, gathered all the people and saying, “okay, I’m going to reveal the identity of whoever committed the crime and that person’s name is”. Predict that word. Predict that word, exactly. My goodness. Right? Yeah, right. Now, there are many different words. But predicting those words better and better and better, the understanding of the text keeps on increasing. GPT-4 predicts the next word better."

He's talking about the model as a "decider" which it's good at, not a detective or investigator.
For the model to output the highest likelihood for the correct token, it must literally solve the mystery by inference with the given clues/narrative.
That's completely different from investigation which involves iteratively hypothesizing and gathering evidence, and at each step coming up with ideas from an open field of possibilities. Having everything laid out for you as a classification problem and then just pronouncing is what ML models do very well and is an unrelated problem formulation. "AI" is great at solving well defined problem. Deciding there is a problem, formulating it, deciding and collecting what you need to solve or test it, iterating, not so much.
Your recipe for investigative skills also sounds a lot like science. Even today, the various chat bots can formulate theories and test protocols. I don't see anything that puts any of that out of reach.
I'm sure I've seen articles about AI doing the former and videos demonstrating robots doing the latter.
Anything that involves in-person charisma will not be replaced for centuries, if ever.

Ditto for any kind of writing or communication skills that are counter-cultural in nature. It seems pretty clear that the AIs corporations are creating are pretty aligned with whatever values the powers-that-be have. I don't expect AI to invent the equivalent of punk rock.

Who needs charisma if you're a robot overlord? Motivation could be achieved by other means, and if AI is really that smart, they will see that.
Robot overlords aren't going to take over the world anytime soon. Real life isn't a sci-fi novel.
I don't believe a timeline was listed. I think it'll only be a couple decades before AI is extorting people.
I'm sure they already have extorted people, it's just a question of how many.
Predictability and objectivity of a robot >>> charisma for most jobs. People prefer not to deal with people, to do stuff via web forms instead of phone calls etc.

In person charisma is also easily done via robots by just giving them an animated avatar, that robot will outperform 99% of humans easily since humans have a hard time faking emotions while a robot doesn't, they can fake being excited or sad or understanding every time and always be there for you unlike real humans. The only reason that is hard for humans isn't that it is hard to do the action, but that it is hard to control our emotions.

I’m sorry, but if you think that an in-person robot with an animated face is more charismatic than a human being, we must live in different universes.
Real humans mostly shows fake emotions when they work in such jobs, everyone who can read humans knows that.

Fake emotions aren't very charismatic, you feel the fakeness from a mile away, a robot showing emotions without that hint of fakeness is way better. Of course you would need to build a model around that, which people hasn't done, but the rules for charismatic social interactions such as serving food are really simple so I see no reason why that would be hard to do.

Maybe you are underestimating robots here, based on the sci fi robots that are very inflexible and bad at showing emotions? We know that AI can show extremely expressive faces when they draw images etc, there is no reason to believe they wouldn't be able to do that since they already can do it.

How does a robot have real emotions? Aren't they fake by definition?

By charisma I mean in the sense of higher-level sociopolitical organization, like being a manager at a company or giving a public speech – useful skills in the job market that will continue to be useful. Not serving french fries. A smiling robot isn't going to inspire or convince a human being of anything.

Regardless, the idea that a robot smiling is somehow "more charismatic" than a human being is absurd. People read too much sci-fi.

> How does a robot have real emotions? Aren't they fake by definition?

That wasn't the point, the point is whether you can feel that they are fake or not, doesn't matter if they actually are. Fake feeling emotions are much harder to convey for a robot since they require more nuance.

> A smiling robot isn't going to inspire or convince a human being of anything.

They already have, smiling robots with basically no intelligence are already making billions for corporations in phone games.

> Regardless, the idea that a robot smiling is somehow "more charismatic" than a human being is absurd. People read too much sci-fi.

Why is that absurd? The idea that robots can't be charismatic is probably the most common trope of sci-fi, in reality we see robots mastering soft skills way easier than hard skills, there is no reason to believe charisma will come after understanding at this point, if you believe that you have your head in the sand.

The purpose of this topic was to discuss skills that aren't likely to be replaced by AI tools in the near future.

Speculating about superhuman AI androids that are more charismatic than humans is entirely beyond the point and nothing more than a sci-fi fantasy right now.

Unfortunately, AI discussions as of late seem to be filled with this kind of absurd speculation that is no more than magical thinking. It belongs on a science fiction forum.

> The purpose of this topic was to discuss skills that aren't likely to be replaced by AI tools in the near future.

Right, and managers are extremely unsafe with the tools we already have. LLMs are more charismatic writers than most managers, and we know from Uber that we can automate away the management layer easily as long as we can have other signals to govern workers. With remote jobs being popular today face to face communication isn't needed, you just need to write well, so that isn't a problem either.

Managers doesn't need as much subject matter expertise as workers, so managers will get automated away easier than workers.

> Speculating about superhuman AI androids that are more charismatic than humans is entirely beyond the point and nothing more than a sci-fi fantasy right now.

I didn't say android, I say an avatar, I meant on a screen. And no you don't need a superhuman AI to express emotions better than humans, we already have AI that are better than humans at expressing emotions, both in text as LLMs or in images from the image generators.

Current robots, sure. Uncanny valley. It's really really hard to mimic humans, and getting close can look much worse than being deliberately mediocre. This is why Pixar faces work, while The Polar Express was panned.

Ultimately though, that's just muscle under skin — with sufficient effort it's obviously replicable, and thus a question of when.

English isn't an ideal language to describe what kinds of "charisma" an AI could have, but I don't think it's at all foolish or improbable to say than an AI system could have profound and indeed superhuman charisma.

Worshipping gods is an old human sport, and it wouldn't need *much* encouragement for humans to see an AI system as a sort of "village god" –– the kinds of entities which were worshipped all over the world before One Big God religions became common. A "Tesla Kami" which talks from your Tesla, and also talks for The Teslas, and also talks for Tesla, Inc. for example could be highly charismatic and not even slightly humanoid in appearance.

If you think I'm joking, consider the Knight Rider TV series with a talking car. Cars can have charisma even if they don't talk, but add a voice-and-mind and we're off to the races, so to speak.

Charisma is a funny thing.

Art. By definition, and applied broadly across fields. And I mean real art, not generic, cookie cutter, mass-production "art" (that can be replaced, and makes up the majority)
Some forms of art will be more protected than others. Think of somebody who builds unique artwork by hand 100% from random wood and other material found in the nature. Good luck having AI robots pick those from the wilderness, put them together, or 3D print a fake one that looks exactly like a real thing.
what is "real art"? what you're at it, define what a True Scotsman is for me.

for every artist getting a residency at MoMA there are dozens more doing logos, UI, concept design, etc., and those people are getting slammed right now.

but yeah, the one-off person getting a grant from somewhere to do art might be able to hit median US income -- AI-proof!

I hope we can find a path to a post-scarity, Star Trek world, where people can pursue passions without worrying about marketability and whether "AI" whatever will replace you.

I can get great furniture at Ikea, but some people build furniture because they love it.

> I hope we can find a path to a post-scarity, Star Trek world, where people can pursue passions without worrying about marketability and whether "AI" whatever will replace you.

*Shakes AI Magic 8 Ball* “Outlook not so good”.

The people with money are more concerned with dreaming about bunkers managed robots than solving any real actionable problems.

https://www.goodreads.com/en/book/show/60165391

https://www.youtube.com/watch?v=pwJQEAI_KE0

https://www.newyorker.com/cartoon/a16995

So, you're saying we're hopeless without the help of billionaires?
Even Star Trek had a World War 3 that destroyed most of the human population before rationality finally prevailed.
High manual dexterity in cramped, wet, poorly-lit environments.
Any physical type of jobs: house repairs such as painting, building, etc. Roof cleaning, moving furniture. Physical therapists like massage/needles/etc. Physical teaching like 1:1 practices of martial arts/dances/etc.
Physical teaching like 1:1 practices of martial arts/dances/etc.

There are whole host of companies offering various apps where you film yourself doing a sport or activity and an AI will give you advice on what you are doing wrong and how to improve. They might not be great now, but in a few years they will probably be as good as most average coaches/trainers.

Yeah, well. I train in boxing 1:1. No app in existence can replace a physical coach for when we do training on the pads, let alone sparring.

I bet, in some distant future, a robot might replace my coach. But it's very likely that by then I'll turn into a power cell for a huge electrical grid for robot survival.

physical engineering. That will take a while.
ensuring and enforcing ethical applications of ML

standup comedy

My plumber seems pretty safe about AI at least for next century. Unless houses are rebuilt with standards and ease of repairability in mind. But even changing all the houses in the world to this new system would take over a hundred years.
Prepping a young person to be on a path to eventually owning their own plumbing, HVAC, roofing, etc, business does seem a pretty solid plan.

Though I suppose AI could be part of something that squashes small businesses and funnels work to huge companies. Which would make that entrepreneur angle a bit harder.

There's already a fair amount of SEO gaming to make something that isn't "local" appear to be local.

I'd expect roofing to be less immune. No offense to any roofers out there, but roofing is a lot more formulaic than plumbing or HVAC.

I don't think it would be hard at all to design a robot to roof the large flat areas typical of modern architecture (if it can be dignified by that name), maybe with a human or two to make sure the robot keeps working, and to handle any peaks, valleys, plumbing or heating penetrations, or other weird areas.

yeah but roofs aren't formulaic -- even subtle looking differences == big challenges in climbing and hoisting -- and even if you have rows of tract housing with identical builds you're going to run into problems getting things up there because access, yards, fences, etc. all get difficult and complex.

source: used to dispatch for HVAC, roofing, and paving services.

https://www.renovaterobotics.com/

Are you sure about that?

"Renovate closes pre-seed to automate roofing"

Feels a little early to call.

Clever, but a low flat gabled roof pitch that requires minimal effort to determine appropriate anchor points is probably a pretty rare roof shape for a lot of homes. When that thing can do a McMansion is when it has a real chance of disrupting the roofing industry.
I'm old enough to remember people saying that about go.

I pitched the idea of an AI directly generating art sometime around 2008, only to have the idea curtly dismissed.

Deep Fakes (by a different name) were a plot point in an episode of Sliders, and the audio version a throwaway line in an episode of Babylon 5, at the time I thought each were completely impossible.

My bet would be androids doing plumbing around ten years after cars lose their steering wheels; but even then, that's just anchoring on Moore and the available power envelope for a car vs. a bipedal robot.

Yes, eventually robots will be able to do plumbing.

But if/when robots are doing a majority of plumbing jobs, they'll be doing 99%+ of all jobs. So society will have changed drastically and unknowingly, for better or worse. And that societal change will affect all jobs, robot-replaced or not.

So IMO a relatively robot resistant job (like plumber) is as good as a robot proof jobs.

If you've ever tried to squeeze yourself under the kitchen sink, work one hand behind the 5" gap around the garbage disposal, work the other hand around the tiny gap between the garbage disposal and the main sink, try to coordinate the 2 hands to tighten the faucet, then I'm sure you'll agree with OP.

We don't even have a commercial robot to unload and store dishes from the dishwasher nor unload and store clothes from dryer. What makes you think robotics will be solved in a century, let alone ten years?

> If you've ever tried to squeeze yourself under the kitchen sink, work one hand behind the 5" gap around the garbage disposal, work the other hand around the tiny gap between the garbage disposal and the main sink, try to coordinate the 2 hands to tighten the faucet, then I'm sure you'll agree with OP.

What's hard for me doesn't tell you anything about what's hard for a differently shaped body, and robots don't need to look like us — two left hands is easy, extra joints is easy.

Most robots don't look like us, one way or another. Famously Spot, but there are as many sizes and shapes as there is commercial demand for them. Humanoid ones are popular right now, but I grew up hearing about how that was "actually a bad shape for a robot" (was the past just sour grapes, or are the current humanoid ones CEOs watching too much SciFi?)

> We don't even have a commercial robot to unload and store dishes from the dishwasher nor unload and store clothes from dryer.

Both dishwashers and clothes driers are special purpose robots designed to solve problems without having to first solve human-level sensation or dexterity.

> What makes you think robotics will be solved in a century, let alone ten years?

Have you seen demo videos from the last 12 months? The videos are sped up, indicating they're compute-bound; the accuracy of joint positioning, the strength, etc. is already superhuman whenever we want it to be.

Quality of vision, that's the hard part. Cars can only lose their steering wheels[0] when that's solved in terms of quality, not before; a Moore-like exponential on reducing electrical power requirements for a fixed level of compute, would take 10 years to reduce power draw by a factor of 100, which seems about right given cars can afford to (regardless of what current hardware actually uses) burn a few kilowatts on an autopilot.

[0] Until the Waymo etc. is available everywhere rather than geofenced to a few specific locations, this is still not sufficiently solved despite, yes, having no steering wheel: https://thelastdriverlicenseholder.com/2023/01/05/waymos-new...

Why would I pay someone to fix miles of hidden plumbing when I can just place a bedpan on top of the robot vacuum cleaner as he goes past who then hands it off to the septic tank bot. If simple object from A to B manipulation is free & fast then we are going back to medieval times with robot serfs. The pipes will be a vestige remnant like landlines are.
Motor/hand eye coordination most definitely. That's what millions of years evolution has optimized us for more than anything else. Anything involving manipulating and movement in varying physical environments remains a hard problem to solve in cost-effective manner, despite all the money poured to self-driving cars and general purpose robotics. Of course we have solutions that work in specific contexts, factory or lab environment, but that's about it.

I believe emotional intelligence will become a close second. As long as we don't have perfected humanoid robots like in TV series Westworld, people will likely prefer genuine human contact in certain situations (psychologist, therapist, service sector, nursing...)

Tour guide, nurses, physio, child care, psychologist.

Anything in front of a screen is vulnerable

The only profession where I’m 100% confident it’ll survive is the oldest profession

mobile phones with tour guide apps: knows where you are and what you're looking at and can tell you everything about it.

psychologist was one of the first ever applications of AI - remember eliza.

but child care, nurses and physio are, imo, safe.

I predict that rich people will continue to pay for a real psychologist to get genuine human contact. Meanwhile, the big masses of poorer people will eventually get to enjoy LLM-powered AI ones, which may be an improvement or a downgrade depending on the current state in their area. My country's state funded healthcare system is already utilizing web-based therapy, LLM's will just replace the human writing responses.
Yes, I think nursing is perfect purely from a "not being replaced" perspective. An absolutely massive demand from aging population with lots of money, requires physical movement, eye-motor skills between and inside varying environments, like moving between old people's houses and then taking care of the people inside. On top of all that, a real human nurse could (ideally, when not worked to death) give genuine human emotional understanding.
Creative prosthetics and a VR headset and I'm less confident
lots of studies have shown that screen time for kids ain't so great. i wouldn't send my kids to a daycare that was one person + 20 youtube streams + bots
> nurses

Yes. I spent a few days in the hospital last October for some jaw surgery. When I was flat on my back recovering from the total anesthesia and unable to move, I observed what the nurses were doing and thought about how much of it could be replaced by AI. Essentially none, unless humanoid robots get really, really good. They might eventually, but that seems now to be a long way off.

Art is never going to be replaced by ML-generated works, because art is self-expression and so requires there to be a self to express.
If you're a starving artist, AI poses absolutely no threat to your position.
That may be the case for some art, but self-expression is not something I strive for in my own art. Just to list one example.
Self expression is relevant to what I do in private, but not only am I happy to act as a stereotypically annoying client (you know the type) to an infinitely patient AI in order to find expression for what I want, in addition to that, the art I make by my own hand (even including non-AI programs like Blender) has negligible market value.
It is that already by default, without even trying. The beauty of how the mind works.
Responsibility. As separate from technical ability.

You won't be able to replace doctors, because you want someone to say "I'm a doctor and you need this surgery". What will happen is the doctor will get an AI to help him, and then he will rubber stamp the diagnosis. He captures the gains.

Pilots and captains are also one of these. Someone's gotta be responsible for the people on board, even if a computer is already doing everything on a normal day.

> Someone's gotta be responsible for the people on board, even if a computer is already doing everything on a normal day.

Military planes have been way beyond unassisted human control for a long time now. I see no reason why airliners won't go down the same road (you can literally see it happening as we speak).

It turns out that dynamic instability has some real advantages (e.g., maneuverability), if you have a computer to help keep the thing from going out of control.

The military is reluctant to take the decision to kill someone out of human control.

The AI may fly the plane, but the trigger is squeezed by a human.

I'll disagree with this take too. In cases where AI is comparable or only slightly better I'd agree many would prefer a human doing this task.

However when AI does a job better people will be fine with AI taking on the role. We are fine with computer controlled subway lines, dishwashers, and some even driverless cars. We are happy let machines do tasks if we're confident they will do them well.

Right now, AI need a huge number of examples to reach decent performance. Therefore, things where a huge number of training examples don't exist and can't be generated by a simulation or self-play. When AI get past that problem, I expect the whole global economy to shift in a couple of months[0].

Also, things where the point is to be human. Working for rich people in any capacity may fill this role, for the same reason they by watches which cost more than apartments yet keep worse time than a Casio F91W.

[0] I'm also assuming this is more likely to happen after we get a significant number of android robot workers; if it happens before that point, then it may take a year or two for the global economy to shift.

Came here to say this. And to add: we don't know to what degree human behaviour is regular.

It could be that in practice an AI can't really generalise well to All People compared to humans, because the humans are so diverse and the humans who are doing some jobs generalise to just the small set of people they are working with.

Right now it does look like prompts like "answer this as if you're one of Napoleon's generals" work fairly well, but I wonder if in practice we'll find out that specific human experiences which are hard to document will turn out to create niches which AI is very very poor at dealing with.

Potential examples:

* what does it feel like to fall in love? humans can't describe well.

* what does it feel like to summit Everest for the first time? ditto.

* what are spiritual experiences like?

Probably many other areas which are clearly important to people, particularly falling in love, but language is weak, and at the end of the day...

It's a large language model.

Lawyers.

I think AI can possibly replace lawyers, but the powerful people in society will do all they can to prevent it.

AI should replace lawyers, since they are essentially a regressive tax on engaging with the legal system.
I agree. If access to the legal system becomes easy enough it will stress the system and get some parts of it fixed. Which is one reason I think lots of artificial barriers will prevent this from happening.
I do agree lawyers will always be needed.

But, don't you think AI is already making lawyers dramatically more efficient? So much of what lawyers do is reading - reading possibly relevant case law, reading corporate communication.

I met someone recently who has a law degree from a small no-name law school. He got mediocre grades. His job is sit in a room with 10 other lawyers, reading through every single document in a giant room full of filing cabinets. It's part of a corporate merger. It's going to take years to go through but they need lawyers to read through every single page looking for certain potential issues.

That's the sort of thing I feel like will be eliminated.

there will be a lawyer since they're licensed by the state bar, and to file certain things or do certain actions you need someone who can check the box, similar to formally certified medical or engineering workers.

but a lot of law is review, of contracts, of statues, etc. and legal discovery companies are already a thing and were snagging a lot of marketshare before GPT exploded.

"read each clause, identify weasel words, parties, and summarize each line or paragraph. then create a link to any relevant case law or statue for sanity checking" -- is already a thing. very little law actually happens via dramatic courtroom antics or powerful closing oration; most is summarizing things as mentioned above.

just like how excel didn't eliminate accountants, but now instead of 10 you need 3, this will cut the already tight legal market down further.

Why would powerful people want to prevent lawyers from being replaced by AI? Wouldn't they rather not pay for expensive lawyers too? Or are you saying they somehow like paying for lawyers?
Took them 15 years to replace cashiers with a touch screens in McDonald's.

"AI" still can't drive a car.

Prostitution.

Pornography is doomed, but we're a long way from realistic physical sex robots, AI powered or otherwise.

AI replacing skills is a moving target. Anything can be replaced by AI (more so information jobs) but as things are replaced humans just have to work harder to be novel. AI is by definition trained on the largest pool of examples, which means being novel will mean your skills are excluded by AI until it catches on. Of course being novel is something that perpetually happens. Ideally the world becomes a place where the AI and robots do the rote busy work and humans are relegated to finding novel solutions.

Even before AI there was the issue of being replaced by cheaper labor. AI just happens to be computational cheap labor. Just be better at what you do to avoid being replaced.

> Just be better at what you do to avoid being replaced.

Git gud! Great advice!

Plumbers, locksmiths, carpenters, HVAC technicians, etc. Trade jobs that require very precise hand-eye coordination and the ability to perform in a wide variety of unpredictable environments.

Nurses/MAs for the same reason, but also the fact that I think people in pain/sickness will still want human compassion. Even if AI and Robots may be able to do the job some day, people want to be helped by other people.