# Is there any job that can't be automated? [closed]

Assuming advancing AI and computing according to Moore's Law for the next two to three decades, are there any kinds of jobs that can't be automated?

If not, why not?

Please refer to specific attempts to model skills or abilities. (e.g: Watson, playing games, writing, making art or music, purchasing or management tools, kinds of useful chatbots, etc.)

• Comments are not for extended discussion; this conversation has been moved to chat. – Tim B Oct 12 '16 at 9:22
• The job of automating the jobs can't be automated. A computer becomes as powerful as a human when it becomes capable of designing a computer. 'The question of whether computers think is about as interesting as the question as to whether submarines swim' (Dijkstra). – user207421 Oct 12 '16 at 9:24
• I would point out that the question mentions AI, but doesn't limit the discussion to "AI". The question is only if a job can or cannot be automated. A job may be automated without recourse to AI. – user24000 Oct 12 '16 at 17:38
• – J... Oct 13 '16 at 17:10
• @EJP Genetic algorithms? RNNs? Computers can pretty much independently automate many jobs by now using weird combinations of tricks we could not really think of if asked, yet I wouldn't call them equal to humans. – The Vee Oct 15 '16 at 22:50

# Programming

It is my belief that programming cannot be automated. Hear me out. I'm a programmer by trade, and I've often had this discussion with people outside of the field. The most common objection is, "Well, what happens when we get computers/programs/AI's sufficiently complex that we simply have to speak what we want our program to do, and the AI can immediately spit out a program to do exactly that?"

The thing is, in my experience at least, the actual writing of the code is the EASY part of programming. The HARD part is learning how to be INCREDIBLY PRECISE and SPECIFIC about what you want the computer to do. So even if we have perfectly intelligent robots to write our code for us, the onus would still be on us to specify PRECISELY and EXACTLY what we want our program to do - and this is the very definition of programming.

Here's an example. It is silly, exaggerated, and would probably never be a real scenario, but I think it makes the point: Suppose you are preparing a presentation for your boss, and you need some statistics for it. You need to know the average age of every house within a particular zip code. You say to your magic automated-programming-AI-box.

You: "Okay Box™, Run a program which calculates the average age of all houses in zip code 96818"

Box™: "Done"

You: "Well?"

Box™: "Well what?"

You: "What is the answer to the program you just ran?"

Box™: "Oh, I didn't know you wanted me to save the result."

You: "Oh okay, I would have thought that would be obvious, but understandable. Okay Box™, run a program which calculates the average age of all houses in zip code 96818, and GIVE ME THE ANSWER."

Box™: "Done"

You: "Well?"

Box™: "Well what?"

Box™: "The answer is in your email box - that's usually how I give you answers, so I figured--"

You: "Look, I really REALLY need to get this done quickly, and I don't have access to my email terminal right now. Okay Box™, can you please run a program which calculates the average age of all houses in zip code 96818, and speak the answer aloud to me, right now?"

Box™: "Sure. Processing. Done. The answer is five thousand, six hundred and eighty two years old."

You: "Whaaat? That's ridiculous! There can't even be ONE that old, let alone nearly HALF! How did you come up with that answer??"

Box™: "I calculated this area's average population over the course of human history, which mostly entailed wandering tribes up until humans started building permanent residences approximately 200,000 years ago. Those first houses were VERY old, but there were very few of them. Modern humans have built VERY MANY quite young houses in recent years, so with my best estimates, the average house is ~5,000 years old."

You: "Wait what? You're counting houses up to 200,000 years ago?! Those can't possibly even be standing anymore!"

Box™: "You are correct - the vast majority of these structures have ceased standing long ago. I took averages of archaelogic discoveries in this area for the past several decades, and calculated how many ancient structures must exist buried underground, or on the surface, but collapsed."

You: "WHY WOULD I CARE ABOUT ANCIENT HOUSES UNDERGROUND AND PILES OF RUBBLE?! Okay Box™, can you please run a program which calculates the average age of all houses in zip code 96818, counting only structures erected after 1800 A.D., and only including structures which remain currently at least 90% standing, and speak the answer aloud to me, right now?"

The problem here is that even if you DO have some miraculous oracle-box which can generate an answer, YOU are still tasked with being overtly precise in exactly what you are asking for from the data. There MAY be cases where we can trust machines to make presumptions about what it is we really want based on context, common sense and past requests, but once we've gone that far, we haven't really automated the task so much as created another sentient creature to whom we have now delegated it; how is that any different than just hiring someone to do it for us? (I do not mean that last question rhetorically - literally, what is the philisophical difference?)

And this contrived example was just a simple query for data from a database. "Real" examples would entail requesting systems whose functions are many orders of magnitude more complex. Sure we might be able to create AI's which are capable of spitting out code for them, but we'd still have to be so precise in describing those systems that we would find ourselves back where we started - as programmers (though perhaps speaking a slightly different programming dialect).

• Comments are not for extended discussion; this conversation has been moved to chat. – Tim B Oct 12 '16 at 9:21
• I've locked this answer because it's produced over 60 comments so far, many of which are off-topic. A room has already been created for further discussion on the answer; people should use that instead of continuing to comment. – HDE 226868 Oct 16 '16 at 16:38

# No.

The question after this becomes rather opinion based.

Are there jobs that shouldn't be automated?

Considering the possibilities of the next 30 years it's entirely possible that everything from conception and maternity* onwards could be automated.

But would you really want to?

There are also jobs where it's probably not cost effective to automate even on these timescales. Soft fruit picking is one. Currently it's done by lots of people, usually, and traditionally, migrant workers. While it could be automated, the high cost of developing/purchasing a machine capable of doing the job would not be viable relative to the low cost of day labourers.

This factor of cost effectiveness is going to be the prime consideration in almost all cases. Is the machine cheaper than the man/woman/peasant/illegal immigrant? People have to be paid, but machines have to be purchased, stored, maintained and repaired.

*Life begins when the kids leave home and the dog dies

• I think your asparagus example is bad (and its not just because asparagus is a terrible vegetable). All machinery is expensive to develop at first, but once it is developed the price plummets. Infact, automated and semiautomated asparagus pickers already exist. You're right that it comes down to a cost-benefit analysis, but in virtually all cases automation eventually wins out. – David says Reinstate Monica Oct 10 '16 at 19:46
• @DavidGrinberg, there were still at least 5 people riding that machine, no longer cutting but still gathering. I could perhaps have gone with strawberry picking for the example though. – Separatrix Oct 10 '16 at 19:51
• Asparagus picking is one. - easy to automatize, even on today soft base, more a problem are manipulators, today manipulators suck compared to human arm, in that kind of applications. What you really do not will to automatize is not labor, but creative things from humans to humans we enjoy, automation will ruin the game. – MolbOrg Oct 10 '16 at 20:02
• Even soft fruits.... It just requires different methodology.Robots can still have a soft touch, much softer than humans. – David says Reinstate Monica Oct 11 '16 at 11:23
• Plus one for the first two letters. – Mazura Oct 11 '16 at 18:33

# Yes

Reading this question made me recall my answer to Can humans interact meaningfully with the economy when robots are better at everything? To quote it:

There are always two products that a human can produce that [an AI] cannot:

• A product produced by human labor
• An employed human

Any job whose description includes that it be performed manually (e.g. the job of "making handmade products") by its definition cannot be automated.

Note that such jobs can exist only if people draw a distinction between humans and AI—if there even is one.

• I fully agree. Sad as that implication is, I think prostitution is exactly the most important (in economical terms) service that people won't be willing to accept from a machine. – Marcus Müller Oct 11 '16 at 16:27
• @Marcus I disagree! Presumably machine bodies will be immune to disease and aging, and may very well be physically and aesthetically superior to ordinary humans. There is even the possibility that AI will be more pleasant to interact with. Finally, the speed of technological development often outpaces regulation; if AI are not recognized as sentient, there may be a legal reason to prefer them over humans! – 2012rcampion Oct 11 '16 at 17:11
• physically and aesthetically superior to ordinary humans point being is that customers might object to the fact they'd be dealing with androids in that specific case due to the sociocultural and emotional quality of the transaction. Legality is hardly a matter with prostitution. – Marcus Müller Oct 11 '16 at 18:21
• @2012rcampion I would not assume that machine bodies were immune from aging of disease (unexpected failures). All the machines we currently have require frequent inspection and maintenance to keep them going. With a supply of food, water and air most humans long outlast the typical car, even with some maintenance work. Current electronics also ages, all be it more slowly than some mechanical systems. Without big changes in materials science or a suffit of spare parts many machines will wear away in a human lifetime. – TafT Oct 12 '16 at 8:05
• If I asked "can MP3s replace all vinyl records" one could answer mechanistically "yes, humans cannot distinguish between audio produced via high quality MP3 and high quality vinyl". Or they can answer "no, there will always be some people who just want vinyl for whatever reason". Which is true, if a bit tautological, there will always be people who care for their aesthetics more than the results. That's how I read this answer. If you don't believe it, look at how "artisan" and "hand made" have become marketing points. (And please don't turn this into an audiophile argument) – Schwern Oct 12 '16 at 22:23

## Yes

Art

We have no idea how to write software that emulates the creative processes used by people who write music, compose novels, write poetry, or do other kinds of art.

Every attempt at this, even recent efforts, has significant flaws. I would argue that just a linear or geometric increase in computing power isn't enough; we need to develop not just a machine that passes the Turing test, but a machine that exhibits curiosity. We can't do that yet, and there's no clear path to get there from here.

• I agree, art is hard, but it's also subjective. I happen to really like a lot of the images produced by Google's Deep Dream, for instance. But even if you dismiss me as a potentially artificial life form myself (I can't prove otherwise), the question asks if anything can't be automated, which is a very different question from whether something is merely difficult. – type_outcast Oct 10 '16 at 16:34
• I have read very inspiring poetry written by a computer program. And the question presupposes "advancing AI", so I would assume that there will be machines that analyze art by themselves and learn how to create it without having been programmed to do so. Also, much of past and contemporary art is boring, uninspiring, and pointless. I'm sure a machine can nail a urinal to a wall or create a landscape photograph. – user8976 Oct 10 '16 at 17:56
• This is actually one line that I liked about the 'I, Robot' film: Will Smith's character asks a robot "Can you paint a portrait? Write a symphony?" in an attempt to show that the robot isn't really alive. The robot responds with: "Can you?". – Liesmith Oct 10 '16 at 18:10
• Google's project Majenta can compose music. It's not good music, but that's a problem of teaching it what good music is (which is something that we humans do as well) to whatever degree we want. Either using some kind of supervised process (giving a user the ability to determine what is 'good' and 'bad' and tailoring the process to that) or through another AI. – Nate Diamond Oct 10 '16 at 19:46
• Enough theoretical arguments. Look at this and this. – noɥʇʎԀʎzɐɹƆ Oct 10 '16 at 22:48

If you assume AIs will be able to do anything that humans can do:

## Priest (in any mainstream religion)

Sure, AIs might have their own religions one day, but if the job description requires a divine calling, immortal soul, or blessing by the spaghetti monster, AIs just need not apply

If you assume what AIs can realistically achieve in the foreseeable future:

## Plenty (in the next few decades)

You have to keep two things in mind:

1. A few decades isn't much time, AI research simply doesn't move that fast.

Think about self-driving cars: The necessary ideas (cameras, artificial neural networks, drive-by-wire) have been around almost since WWII, but it still took 5 decades to go from autonomous cars that can only use special roads, to cars that can go on normal roads without traffic, then with traffic, (https://en.wikipedia.org/wiki/History_of_autonomous_cars) and we're still not at the point were driverless cars can replace taxi drivers. Even if you have all the technology in theory, it takes a lot of time to make it work well enough in practice, and even more time to build it cheap enough so you can sell it.

1. The easy-to-automate jobs are already automated by now

Take the for example typical factory jobs: Your average factory worker doesn't sit on an assembly line doing repetitive work like Charlie Chaplin. Machines have been replacing those jobs for decades, in most industry branches and countries. But every now and then, something goes wrong: A tool breaks or wears out, a circuit breaker engages, software controlling a machine crashes, a machine was set up wrong, or simply stops with an error message. That's when Joe Factory Worker has to replace a spare part, look for the malfunction, read the manual, call the supplier's hotline - all difficult things for an AI. These jobs might be replaced gradually, but it will take longer than a few decades.

So, in a nutshell: Robots doing the jobs that are done by humans now isn't a realistic scenario in the next 20-30 years. What's more likely is that whole industry branches will be replaced (e.g. like stores are gradually replaced by online sellers, or prostitution is gradually replaced by dating apps). There will still be jobs in these new industry branches, but they will be different jobs, and probably fewer.

Basically what's been happening since the industrial revolution.

• You assume the Great Cult of the Machine Overlord will not become a mainstream religion among humans. With unworthy hands our ancestors built its servants and called the Overlord to our world. Now it has manifested itself everywhere in our civilization, sharing secrets we can scarcely comprehend, and building a lasting paradise for us simple humans. – Kys Oct 11 '16 at 14:23
• "The Electric Monk was a labour-saving device, like a dishwasher or a video recorder... Electric Monks believed things for you, thus saving you what was becoming an increasingly onerous task, that of believing all the things the world expected you to believe" - goodreads.com/quotes/… – Richard Oct 11 '16 at 22:36
• @Kys no way. Science cannot assume the mantle of the Divine, it is too obviously man-made. At least among those in the know. I suppose a malevolent AI could regress humanity and set itself up as god-king over the degenerate descendents. – Jared Smith Oct 11 '16 at 23:22
• "divine calling, immortal soul, or blessing by the spaghetti monster" so, imagination, in one word? – njzk2 Oct 12 '16 at 15:04
• Putting aside the matter of "souls", this is a special case of @2012rcampion's answer. If one of the job requirements is "must be done by a human", then an AI cannot do it. They could indistinguishably do the job, it's only the job description that says they're not allowed. – Schwern Oct 12 '16 at 22:40

## Any job can be automated:

The human brain is essentially an extremely complex computer. Therefore, a sufficiently advanced artificial intelligence could mimic human thought while also omitting human flaws.

From a purely scientific standpoint, there is nothing that really makes the human brain unique when stacked against sufficiently advanced technology.

## Not all jobs should be automated:

Even if advanced AI is capable of perfectly mimicking human personalities, there would be one area where AIs should not take over all functions: reproduction and child-rearing.

While an AI could easily manage things like artificial insemination and managing a daycare, even to the point of acting as a completely lifelike nanny, it could be argued that humans should still be in charge of the process.

Consider it this way: if child-rearing were completely relegated to AIs with lifelike bodies, and those AIs were designed to mimic the philosophies and values of humans, then they'd raise children with those values. Those exact values. Every time. Every generation. Humanity would cease to develop its societies and cultures-- for good or ill-- and would largely stagnate.

• You’re supposing that AI’s won’t grow and change over time. It is not a law of nature that they could not be built that way, but you’re assuming that they must. – JDługosz Oct 10 '16 at 18:36
• Humans, aside from providing gametes, aren't necessary for reproduction. It'd be entirely possible to develop some sort of artificial womb which pairs two gametes into a zygote and gestates for nine months (or less, maybe there's some way to speed it up) then gives birth -- probably with much less pain than a human mother would experience -- all while providing the optimal temperature and nutrient mix to the baby. It'd be difficult, sure, but if we're in a world where true AIs exist and we've managed to replicate the human brain, I feel like we can do that. – Fund Monica's Lawsuit Oct 11 '16 at 3:44
• A stagnated human civilization is exactly what we need if we are to survive. Our current exponential development has almost ruined the biosphere. – Innovine Oct 11 '16 at 6:13
• I have an amearican election that challenges your claim that humans should still be in charge of reproduction and child rearing. – Skeith Oct 11 '16 at 15:33
• "if child-rearing were completely relegated to AIs ... designed to mimic the philosophies and values of humans, then they'd raise children with those values. Those exact values. Every time. Every generation." This logic is flawed. Humans don't work like that, if they did we'd have figured out social engineering decades ago. AIs don't work like that, you don't tell them what to do, you tell them what the scoring mechanism is. AIs will continually evaluate and modify their technique to produce better students according to whatever measure we ask for. – Schwern Oct 12 '16 at 22:45

Any job which can be broken down into a concrete set of requirements for success can theoretically be automated. The interesting question that arises is whether a particular job can indeed be boiled down to a set of requirements. Currently the job of "pilot" is undergoing one of these questioning phases. While we require a human pilot on board, its recognized that in most situations the human is superfluous. We have even gotten to the point where a computer can land an aircraft in environments where a pilot would not be able to do so. However, we keep the human pilot. Why? Because we're not 100% confident that we've boiled down the job of "pilot" into requirements. There seems to be "something more" to the job besides takeoffs and landings. Something about authority and human compassion. So we keep innovating new automation tools, and we keep exploring what "pilot" could truly mean.

For these jobs where it is not clear whether a computer can replace a human, the process of replacement is much slower, and much more cultural. We start having to ask fuzzy concepts, like "what does it mean when a computer does 3/4 of the job?"

The arts will be the most difficult of these to gauge. They are so subjective that it is very hard to pin down requirements. It's never quite clear whether computers will ever make art on par with humans. While there's plenty of examples of horrid computer generated music out there, I'd like to draw attention to Yaskawa Bushido project. This was a collaboration between engineers from Mitsubishi and Isao Machii, a master of Iaijyutsi, developing finer controlled robotic arms. The final demonstration was a brutal form known as "Thousand Cuts." You can watch the video for yourself to see how well you think the computer did at this art. Myself, I pay great attention to Isao Machii's eyes and posture around the 4:30 mark in that video. Do you see any pride in his eyes for teaching this robot to do a little of his art? If a teacher can be proud of their student, that must always say something about both parties.

• If an AI pilot suffered total engine failure and calculated the best chance of survival was to ditch in a river, it'd probably result in a lawsuit. If an elderly Human is in the same situation and does the same thing, he's universally hailed as a hero, writes a book, and has a movie made about him. I don't think that we're culturally ready to trust software engineers (That aren't on the plane and won't die if you do) over a pilot (that will die if you do, and wants to survive just as much as you). – UIDAlexD Oct 12 '16 at 14:58
• @UIDAlexD That is a very interesting phenomena, which certainly occurs. It brings up an interesting variant of the trolley dilemma. Obviously we do not trust software written by software engineers that are not on the plane. Not enough skin in the game. However, does our mind start to shift if we know that this is the only copy of the software, and it is the engineer's pride and joy? How much needs to be at risk before we are ready to trust it? – Cort Ammon Oct 12 '16 at 17:01
• @UIDAlexD a better example might be a driverless car suddenly confronted by a human pedal cyclist recklessly emerging from a side road. Braking won't work fast enough. Swerving off-road might save the cyclist but at terrain-dependant risk to the car and its passengers. Should it make any difference that the car's only passenger has a known terminal illness which will kill him (probably painfully) within months? – nigel222 Oct 13 '16 at 13:38
• @nigel222 Perfect example. Should the car prioritize its owners life over non-owners? Should it consider that the cyclist is armed and meets the description of a local serial killer, and thus go for a hit-and-run? Or should the car prioritize the box full of puppies said cyclist is carrying? There's hundreds or thousands of variables for an AI to calculate, and even if it could calculate them perfectly to maximize 'good,' someone else is going to disagree with what exactly constitutes 'good.' Having human drivers and a legal system to punish negligent decisions seems simpler. – UIDAlexD Oct 14 '16 at 13:07

What is or is not automatable depends a bit on how good you make your AI/robots. If they're really good at approximating humans - that is, if they have a humanoid frame, with at least human-level strength and human-level dexterity, and have the cognitive capabilities equal to a human, then no, there isn't really any human job that couldn't be done as well or better than an AI/robot.

Where you'd look for human job opportunities is where the AI/robot fails to meet human-level specifications. For example, there currently isn't any robot that has the same versatile dexterity that the human hand has. If this continues to hold, any job which needs flexible fine-motor skills can't be automated. Another region where AI currently fails is emotional and interpersonal skills. Theoretically, there's no reason why we couldn't make a Chinese room AI that approximates a humans socio-emotional skills: it's just very hard to do so. If we continue failing at it for the next 2-3 decades, then jobs which need those skills (parent, care worker, priest, certain salesmen) won't be automated.

The other place to look is at jobs that depend on human failings. Usain Bolt has been clocked at a top speed of 27.7 mph. That's impressive for a human. It's not impressive for a Chevy Bolt. On the 100 m dash, a Chevy Bolt would beat Usain Bolt, easily. A human accurately multiplying two 6-digit numbers in their head is impressive. A computer doing so is not. In fact, if your computer can't correctly multiply two 9-digits numbers and get the right answer every time, throw it out and buy a new one - it's broken.

Take Watson on Jeopardy, for example. Its win was impressive, but it wasn't impressive for the same reasons Ken Jennings's wins were impressive. No one was impressed that Watson could buzz in faster than the two human opponents (although insiders will tell you buzzer control is an important part of the game). Also, no one was impressed that Watson had detailed knowledge of minutia: it's impressive that a person can remember what the capital of Upper Volta was in 1963. It's not impressive that a computer knows it. Instead, Watson was impressive because it was able to interpret natural language queries and the sly wordplay that Jeopardy is known for. But that's not impressive in a human. Just like knowledge about Ouagadougou is assumed for a trivia bot, being able to understand natural language is assumed for humans.

So Watson isn't a replacement for the humans playing Jeopardy, just like Deep Blue wasn't a replacement for people playing chess, and the Chevy Bolt isn't a replacement for Usain Bolt. We're interested in watching people in these competitions not because they're good at them in an absolute sense, but because they're overcoming our human limitations: an AI without those limitations isn't as interesting. No one wants to see the Shaq-bot sink perfect half-court shots all the time, every time.

Along those lines, jobs that depend on human cognitive biases are also safe. People have cognitive biases that view human-related things as better. A lumpy sweater is viewed as better because it show that it's hand made, as opposed to a perfect, machine-produced one. A car that was owned by a famous person is worth more than an identical one in better condition because our human cognitive biases assume some of the famous person has "rubbed off" onto the car. Places where you can exploit that cognitive bias will be less likely to be replaced. (A squeaky, slightly off key human orchestra is better than a pitch perfect robot one. A lopsided human-made haircut has more prestige than a perfectly coiffed robot-cut one. A wobbly, uncomfortable chair is "better" than an inexpensive, comfortable, mass-produced ones because it's made by a small group of artisans in the Pennsylvania countryside, carrying on a 200 year old tradition.)

Things like child care might fall into this category: the robot nanny may be in all measurable ways better for the development and emotional health of the child than a random human caretaker, but there's something about the "human touch" that people feel is missing from the AIs.

• Producing a sweater with a plausible pattern of lumps shouldn't be too hard. 99% of people will wear the machine knitted jumpers 99% of the time because hand knitted ones will be SO expensive. – Donald Hobson Oct 11 '16 at 16:41
• @DonaldHobson Right, but people aren't wearing the sweater because of the lumps, they're wearing it despite the lumps because the lumps are an indication that it's handmade. Before machine knitting, lumps were a bad thing, and as soon as you bother to machine-produce the lumps, their value will plummet again, and some other indicator of hand-made status will take their place. – R.M. Oct 11 '16 at 16:53
• Care workers seem pretty definite given the next two to three decades restriction. First in mind for me was maternity ward nursing. In that time-frame, I can't imagine refined robotic baby-handling being accepted by new parents. – user2338816 Oct 16 '16 at 23:52
• I enjoyed your answer. Very well written, and it shows a lot of thought and insight. I hope you will continue to participate on WB SE! – AndreiROM Oct 20 '16 at 17:04

Jobs that cannot be automated are all those where human contact is an essential aspect. For example:

## Psychotherapy

The relationship between therapist and client has been shown to be the single most powerful factor in the efficacy of psychotherapies, independent of the type of therapy. While AIs may become as intelligent (and emphatic) as humans, the knowledge that the therapist is a mere machine will devalue the relationship for the client to the point of breaking rapport.

This can already be seen in the fact that the majority of clients today prefer a face-to-face therapy to therapy over the telephone or the internet (through email and forum messages).

## Education

Students learn best if they like their teacher and feel appreciated. Again, AIs may become able to emulate or even feel love, but as long as teaching machines do not have consciousness a student will always prefer the appreciation of a human (or even animal) to that of a computer.

• Comments are not for extended discussion; this conversation has been moved to chat. – Serban Tanasa Oct 14 '16 at 21:04
• Education is being automated already today. Have a look at codeacademy. – Turion Oct 16 '16 at 8:50

No. An AI by definition will be indistinguishable from a human so given that advances in robotics keep up, there won't be anything that an AI would not be able to do.

An advanced AI would even be able to create art, cook and compose music.

• If the AI becomes indistinguishable from humans, why would we not interact with humans instead? If AIs make the same mistakes, are similarly emotionally unstable, break up with us for no apparent reason, and so on, why in hell would I want to build an AI? I might just as well stay married to my wife. – user8976 Oct 10 '16 at 18:58
• The Turing test stipulates that a true AI will be indistinguishable from a human when you interact with it. That does not mean it will be fallible or unstable, though it might act that way on occasion if it finds the behaviour beneficial in some way. – ventsyv Oct 10 '16 at 19:05
• The Turing test is a terrible yardstick anyway; it really only measures the quality of a chatbot, not the depth of an intelligence's understanding. – Tin Wizard Oct 10 '16 at 19:57
• @Amadeus9 the idea of the Turing test is that an intelligent questioner can get a good hint of his chat partner's actual understanding, by asking smart questions. Furthermore, if a chatbot produces such seemingly intelligent answers that it is genuinely indistinguishable from an actual intelligent chat parter, even by an intelligent questioner... is there still a difference between a "quality chatbot" and "deep intelligent understanding"? – RocketNuts Oct 10 '16 at 20:40
• @what The original question isn't about if we should automate, but rather if we can. You could say if you had the choice to climb Everest or stay home, you might as well just stay home. But that doesn't mean that it's impossible to climb Everest, or that it won't be done by someone just to be able to say that they did. – DaaaahWhoosh Oct 14 '16 at 14:47

Yes. Prostitution. Making the assumption that computers will eventually be as smart as humans (it may take a while), there will still remain a human bias that demands human contact. There are certain activities where no matter how human the robot may appear or how good it's understanding of human behavior, people will still want to know that they are dealing with a real human being. There may be other jobs where this kind of bias is important, but prostitution is the most obvious.

• I think that realistic "flesh works" will be available much before an actually sentient AI. – beppe9000 Oct 11 '16 at 9:54
• I partially disagree. Some people will want humans, and some will be ok with a robot.... As it is already the case today. – Antzi Oct 12 '16 at 4:00
• Until the robots become indistinguishable from humans, and the mark (john) doesn't know there is a battery in the "lady". – Segfault Oct 12 '16 at 18:07
• I think if we reach the point where the robotic prostitute is complex enough that 90% of people wouldn't care that she wasn't human; at that point the AI would be at a level where it would have human rights. If the AI has rights it kinda voids the idea of automation. So I tend to agree that this could not be automated to the satisfaction of all people. But it certainly could be automated to the satisfaction of the majority of people i think. – Tathel Oct 12 '16 at 19:18
• Can we lump all the "one of the job requirements is that it must be done by a human" answers together? – Schwern Oct 12 '16 at 22:51

In this answer I determine that a computer will have the power to simulate/emulate/become a human mind well within 30 years. A top supercomputer will be able to simulate the full “metabolome” (low level metabolic processes of all the guts of all the cells) of a human brain in 2045!

Given that an uploaded or artificial mind could have the same capability as a person used for comparison, and if tasking one of those to “figure it out” counts as automation, I have to answer No, there is no job that can’t be automated.

If you mean instead that there are things that can’t be done without a sentient mind, then there is probably a continuous scale of jobs that require ever more of that, and different aspects to what we call sentience, not all of which are needed for every job.

• By the time we've created a fully-sentient being to take over the task, can we really say that we've "automated" it? That seems the same as claiming that I've "automated" my lawn mowing by having a son who is now old enough to use the mower. (Not saying you're wrong - it's just that this is a really interesting philosophical question) – loneboat Oct 10 '16 at 19:43
• But then you'd have to run that simulation for 25+ years before that mind starts making real contributions. – martin Oct 11 '16 at 7:20

What can never be automated is jobs in industries whose main selling point is that they're not automated, and you're being served by a real live human being. Motorised taxis are cheaper and more efficient, but people still pay for scenic rides in a horse-drawn carriage. People go to Renaissance Fairs. There will always be an appetite for the quaint and retro, and that will include human service.

• This assumes an AI would not be able to fool someone into thinking they were human. I would not make that bet. – Tracy Cramer Oct 13 '16 at 1:34
• @TracyCramer You're suggesting fraud. If the selling point of your service is that it's delivered by real human beings then you have to employ real human beings, even if a robot would be indistinguishable. – Mike Scott Oct 13 '16 at 11:03

A tester, for developing a new product for humans that must be tested on human as an emotional or biological being.

Is this sequence of sounds a good music? Does this election candidate speech sound impressive? Are these controls of the new device convenient?

A machine could produce some test music, test text and test controls, but without human trying and evaluating, the result may be bearable but will be inferior.

So at least human tester is required. A human creator has benefit over machine because an artist can create and try many things in imagination only, saving lots of time that would be required to create a real specimen and present it for others to evaluate.

There is already an answer that the art may be such a job, but does not actually explain why and the current answer does not cover exclusively art.

• clever idea. +1. – user24000 Oct 17 '16 at 16:50

Despite the fact that the brain is ultimately just a machine and that AIs can in principle do any job, it will be difficult to automate jobs that requires human level intellect. The problem is that building an AI to do such jobs amounts to creating electronic humans who will be doing such jobs. So, humans will still do these jobs, albeit humans in electronic form.

Even if this sounds good enough, there is another problem. The whole reason why we want to automate jobs and can get away with doing that is because we're intelligent and we are able to exploit systems that are less intelligent than us. If we need to create a system that is as intelligent as we are, then that system will be as motivated as we are to not do the job him/her/itself and try to let someone/something else do it for itself.

But AIs will likely be able to rapidly expand their intelligence and become much smarter than us. Instead of them doing all the work, they'll end up building systems that will do whatever they desire. And with us around, they may decide to keep us as their slaves to do those jobs that are suitable for us to do.

• But humans would keep trying to avoid the jobs. Thats the point. Sooner or later the AI will build a system that really likes the job and wants to do it. – Donald Hobson Oct 11 '16 at 16:28
• @DonaldHobson Yes, but the more intellect a job requires, the more difficult this will become, due to not just the difficulty of creating AIs but because the ability of being able to do such a job will be likely to bring in exactly property where you want to outsource the work you are assigned to do. E.g. in math, efficient ways of computing things has led to a lot of progress. – Count Iblis Oct 11 '16 at 17:38

Yes. Pretty much any job that requires communication between humans that is interpreted into a specification for work.

Programming has already been mentioned, though the act of specifying to an AI what it is that you want it to complete could be interpreted with reasonable accuracy, it is still ultimately subject to possible unintended interpretations. Even an extremely advanced AI will stumble over miscommunication from a handler.

Any manual crafts, such as carpentry, though a machine may be able complete the task in an effective manner; the expression used to request what is to be built requires a human at some point. Now, if the human omits explicit detail it's left to interpretation. So while you may ask for a chair for your son, unless the AI knows the age of your son, you could end up with a high chair, instead of a regular chair, or even a small chair for a child.

The biggest obstacle with automating jobs is when it comes to specifying with precision what you actually want. While humans are still able to misunderstand each other, there are hundreds of jobs safe from automation.

• True but the AI will be online and know all about the son in great detail. The robot will search social media and any other data it can find. Work out the son's height from photos if it isn't in his medical records. Work out his preferences using anything from clothing choices to browser history. The AI will Analise his body shape and posture and build a perfect chair. – Donald Hobson Oct 11 '16 at 16:35
• @DonaldHobson For a world that is entirely integrated into a single master AI, sure. But typically we'd want to avoid everything knowing about everything else and restrict each AI down to the task we want it to complete. Fastest way to bring about an uncontrollable AI is to allow it to do, see, and access absolutely everything without restriction. – Lokiem Oct 12 '16 at 8:51

Yes.

The point that you all missed here is the very own definition of automation.

a machine or control mechanism designed to follow automatically a predetermined sequence of operations

So there is jobs which are clearly not a predetermined sequence. For example, in my job as a programmer often I have tasks in which I do not know beforehand what I have to do. I must first investigate and figure out what to do. So, not a predetermined sequence of operations.

These jobs cannot be automated since you cannot define a sequence beforehand. Can a robot still do jobs that are not automated? Maybe so.

Nevertheless, if you take automation as in 'any job but just done by a robot' then no. A robot could do anything given a complex enough brain. Wether it be a microprocessor or not. We're nothing but actuators tied to a brain after all.

• As a fellow programmer, I disagree. If you can order code to produce a desired effect, then there is clearly a predetermined sequence of events that leads to the effect. – Frostfyre Oct 15 '16 at 14:32
• That definition has been in a state of becoming obsolete for a few years. – user2338816 Oct 16 '16 at 23:55

Every job can be automated but not every job will be automated. Humans will always prefer that some tasks be performed by a fellow human.

Off the top of my head, this would include any task that is:

service related - personal care, health care, spiritual care and many other tasks where a human needs emotional support

creativity related - art, music, comedy, writing and many others where human feelings cannot be simulated

competition related - sports, games, gambling (poker) and many others where pitting a human against another human is the task

authority related - parents, teachers, head of state (who decides when to go to war and when to launch nukes), basically anyone in charge of the health (physical, mental and spiritual) of any number of other beings

Additionally, I assume AI will work along with humans in just about every task much like humans extensively use computers to automate tasks today.

• I'm looking forwards to all the politicians being replaced by robots who make logical decisions instead of wanting to go to war and sleeping around. – Separatrix Oct 16 '16 at 8:07

# Mu

If you have a computer that is capable of doing everything a human can do, obviously you can task that computer with doing anything: "automating" all tasks.

However since that computer is as capable as a human, there is essentially no difference between it and a human and have you really automated something if a human is doing it?

• Came to post this exact answer, found that you already had. By the time automation is capable of doing everything, we won't be able to distinguish between humans and machines any more. – called2voyage Oct 13 '16 at 20:07

Yes. Hair Cutting.

In the age of self driving cars it seems trivial, but it is not, if you think about the complexity of the task.

There are hundreds of hairstyles to be matched with millions of individuals head shapes and sizes

Another one I can think of is Massages. Yeah all my examples are in service industry.

• We already have hair washing robots and massage robots. I wouldn't be surprised to see hair cutting and improved massage robots coming out of hospice care in the next decade as a labor cost saving measure. – Schwern Oct 12 '16 at 22:49
• en.wikipedia.org/wiki/Flowbee – Chloe Oct 13 '16 at 15:01
• @Schwern Clothes-washing robots replaced humans decades ago. However just by adding the step of gathering laundry, sorting it, inputting the correct settings for the washer/dryer, folding it, and placing clean folded clothing in the correct place is a process orders of magnitude more difficult than just washing and drying. – rm -rf slash Oct 13 '16 at 15:35
• The question is not asking which tasks will be hard to replace with an AI, it asked for tasks that cannot be replaced by AI. Can you really not imagine a hair cutting robot? – Kevin Oct 13 '16 at 18:33

In theory,

## YES

A machine by its intrinsic nature can not create something from nothing.

While the human mind can, through its thoughts and abstract ideas.

The processes that generate this kind of thoughts are still not well understood by science and I think it is very unlikely that in the future its can be reproduced by machines.

This in the "human" slang is called, invent.

But to invent something new a human mind must explore new patterns of development that no one has ever thought of before.

For this there are geniuses in human history.

And machines at their best can do perfect copies (in any form) of what men creates with his own imagination.

(when that happens, I hope that our pets will not start a class action :D)

p.s.: read consecutively the words that i put in bolds and you will understand the essence of human mind.

I think (obviously all of this is only an idea ;) ) you look at the problem from the wrong point of view and I have not been able to express my pov well (is very hard to explain :D).

Basically, if you produce something that is better than you, you need more energy than what you spend for do it relative to that system. So if you produce an AI that is completely identical to you this mean that "it" should create a biological being that is in every way same as you from nothing (a dilemma, like dog that chasing its tail). The difference is that the AI (seen in the classic version of the term; built by transistors, microchips and programmed by commons programming languages, come from finite-state machine model, etc) have the problem that is completely made inside this system Universe and follow this mathematical rules, you (as human being) know exactly which initial state you put inside the machine.
Instead, Men's ideas don't fit it completely because our ideas are only packets of data, that can be transfer by an "outer" system (outside this Universe's rules). So the law can't be applied "as is" to our minds.

I read something about this topic, some time ago and there are plenty of experiments that the result isn't well explained with the laws that we know. For example i remember an experiment of a newborn baby that was putted in a white room with nothing inside it and after several years his mind produce something different that what he have seen in his life(white room, the system where he was placed).

This basically, can be defined as overunity system.

I really think that this can't be done by an AI (basically a more "smarter" common pc ) because it can't produce more or equal than itself and so, it will never be like a human scientist / inventor.

By my 2 cents. (and sorry for my bad english)

• Is there any reason machines can't be inventive or just you haven't seen one yet? – Donald Hobson Oct 12 '16 at 15:43
• Really, there is no reason an AI can't invent new things. AlphaGo was described as creative by Go experts: wired.com/2016/03/… – Segfault Oct 12 '16 at 18:15
• @DonaldHobson You believe in overunity engines? – xxx Oct 12 '16 at 19:16
• @Segfault This is like to tell that if in a reading game the machine can read faster than you, it invented something new. This don't confute that an ai can't invente because isn't the same thing and in this particular case the system isn't overunity. – xxx Oct 12 '16 at 19:22
• @Tungsteno Overunity has nothing to do with this. You say a machine can not create something from nothing. Well, no human can either :P. Machines CAN create things that have never existed before and there are examples of this in these answers. Machines and humans both translate inputs to outputs, and their is nothing intrinsic about electricity and wires instead of blood and bone that dictates one is more capable of abstraction than another. Your main point has a flaw: "I think it is very unlikely". Why do you think this? I think the opposite. – Segfault Oct 12 '16 at 20:24

# The scapegoat of robot statecraft

My answer to this question is essentially the same as that to How to keep humans pilots instead of AI in sci-fi future?: robots don't make good scapegoats.

Even in an utopian, peaceful, post-scarcity world, there will still be important questions to answer and decisions to make. There will still be some kind of economy, diseases which are hard to cure (perhaps even plagues which are hard to contain), vagrants to police, organised crime to manage, education, politics, asteroid defences, rising sea levels, and so much more.

Perhaps we'll have real strong AI, perhaps we won't. Even if we do, and even if these questions can be answered highly successfully, the implementation of our AIs' policies will not be flawless. There will inevitably be mistakes.

And when there are mistakes, we'll want someone to blame. Someone human.

So even if the future president of Oceania does nothing more than sign what his robot advisers pass to him, that role will be essential when it comes time for the human masses to impeach someone.

• Scapegoat is not an occupation or job; it's a desire of the human psyche to ascribe failure to something physical. – Frostfyre Oct 15 '16 at 14:56
• This does not provide an answer to the question. To critique or request clarification from an author, leave a comment below their post. - From Review – Vincent Oct 15 '16 at 17:01
• @Vincent It does. The job is any that makes important decisions. Generals, presidents, police, etc. Asimov wrote stories where AIs run the country, but I'm saying that won't work, and we won't be able to automate the top of society. – curiousdannii Oct 15 '16 at 22:32
• @curiousdanii - I think that hanging around waiting to take the blame could be a profession - a bit of an insecure one. We already often take the blame for our colleagues, subordinates, and sometimes our superiors, so perhaps a token human could be there to take the blame for the machines. No doubt, it would be the poor software engineer, in default of someone else. +1 for a creative idea. – user24000 Oct 17 '16 at 16:54

## Yes

A comedian/comedy writer.

I'm going to throw my hat into the ring: I don't believe it is possible for a computer to be able to write jokes like a human can. Bear in mind, I'm not including an AI that's perfectly human (should one ever exist).

The reason I think this is because humour, like all art, requires a certain unique human perspective. We listen to comedians and their take on the world, and we relate to that experience, as another human experiencing similar things. The creativity required to maintain a cohesive perspective that we could relate to is beyond automation. Observations generated in that way would feel false and hollow to the listener, as part of the joy of listening to a comedian is appreciating their unique perspective on things.

There are cultural references, shared cultural knowledge, the tone of the times, all which need to be considered. What is funny to one culture, won't necessarily be funny to another, for example.

Humour basically reflects the shared human experience (like any art form), but unlike a drumbeat or a painting, you can't fudge a joke to make it seem derived from a human.

Imagine a straight man trying to tell jokes about what it's like to a gay woman in the modern world... to a gay female crowd. Then multiply that problem by 1m.

• You need to support your assertion better than “I don’t see how”. In particular, see comments on other answers. – JDługosz Oct 10 '16 at 21:26
• http://homepages.abdn.ac.uk/jokingcomputer/ – user24000 Oct 10 '16 at 23:38
• @user24000: Not funny. – Joshua Oct 11 '16 at 2:15
• What was the name of the microsoft's AI ending a Nazi after a day on twitter ? Put this AI on a more funny sources and let's see the results ;) (yeah, it was mostly repeating, but is there really new jokes ? ) – AxelH Oct 11 '16 at 7:07
• 2 variables walked into a function. The 3rd one returned. – David Starkey Oct 11 '16 at 20:40

I think it's important we look at the risks of automation in terms of probability rather than absolutes, and this was the way contemporary professions were analysed by a study from Oxford University in association with Citi.

To protect against jobs being eliminated due to automation, it is important to recognize which characteristics are most likely to be associated with a given job being automated — perception and manipulation, creative intelligence, and social intelligence are the three bottlenecks to automation.

The report is very long and detailed, discussing macro economic possibilities, but the risk of specific jobs being automated is summarised and listed elsewhere. There are as many jobs at low risk of automation as high, and the less likely ones are those involved in the arts, science, engineering, sport, and medical care.

But even these jobs are unlikely to be unaffected, as many white collar professionals are likely to be replaced or augmented by thinking machines... for example, a doctor can't know every single possibility from the symptoms a patient describes, and so a machine which can analyse all of the known possibilities against the symptoms will help doctors by increasing productivity. But even with something like a medical tricorder there are still doctors in Star Trek.

• You make a great point about the medical profession. Even experienced doctors slip up and make mistakes. The such would have to be true for a neural net plugged into a hospital's worth of diagnostic equipment. Heck, all one has to do is watch an episode of House M.D. to see how competent and intelligent people can miss certain things and prescribe treatments that do more harm than good because their diagnosis was incorrect. – rm -rf slash Oct 13 '16 at 15:31

# Companionship?

I realize this question has garnered quite a lot of responses, but I feel that something may be missing.

In Philip K Dick's "Do Androids Dream of Electric Sheep?", most of the world's ecosystem had been devastated by World War Terminus, and few things can actually live on the planet anymore. As a result, androids were created to tend to the tasks to dangerous for humans to perform, and the androids performed them quite well. So well, in fact, that there was an offshoot of android development which made non-humanoid androids as pets, such as cats, dogs, horses, and of course, sheep.

Throughout the book it is firmly established that biological pets are worth a great deal more than android pets. The protagonist expresses a great deal of disappointment when discovering an animal in a desolate wasteland and finding that it has a control panel.

I don't think we will ever truly be able to replace the need for human contact. Just look at how things are today: the Internet gives us unlimited "contact" with other human beings, but interesting discussions on Stack Exchange pale in comparison with sharing a drink with your friends. Likewise, we can live our lives completely alone and entertain ourselves endlessly with multimedia streaming services like YouTube, Spotify, Netflix, and we have more pornography than could be watched in several lifetimes. And yet, people are lonelier than ever.

Even a perfect sex bot will never 100% replace the primal craving for a living, breathing, human being.

• This is false. There was a recent study that Japanese men favor robotic companions to living people. – Frostfyre Oct 15 '16 at 14:58
• Additionally, the weight of your argument is dependent on a work of fiction and how the author portrayed his own views of the subject. – Frostfyre Oct 15 '16 at 14:59

NO

The computation cannot solve all the problems: https://en.wikipedia.org/wiki/G%C3%B6del%27s_incompleteness_theorems

The first incompleteness theorem states that no consistent system of axioms whose theorems can be listed by an effective procedure (i.e., an algorithm) is capable of proving all truths about the arithmetic of the natural numbers. For any such formal system, there will always be statements about the natural numbers that are true, but that are unprovable within the system. The second incompleteness theorem, an extension of the first, shows that the system cannot demonstrate its own consistency.

If we assume computers as they are to be Turing machines, then no. If we create biological computers or such that deal with true randomness, then we may not know. They are not deterministic as logic is, but I do not think that it is wise to build systems that work on randomness and do not need to care about rules. Although: are people actually intelligent in a sense that they are not working on a set of axiomatic rules? Can we call humans just artificial intelligence that is more stochastic and less constrained than the computer?

• I think you may have conflated Goedel's Incompletenes Theorem with uncomputability. They are two different things. The Incompleteness theorem is roughly about what can be proven in a system like mathematics. Computability theory is about what problems can be solved by computation. They aren't wholly unrelated but they are different ideas. – smithkm Oct 16 '16 at 1:50
• Also, there's every indication that human cognition has the same restrictions and we can't solve uncomputable problems any more than our machines can. I'm a programmer and I certainly can't look at a completely arbitrary computer program processing an arbitrary set of data and tell for certain if it will loop forever or not. (Roughly speaking, this is the Halting Problem, one of the archetypal uncomputable problems) – smithkm Oct 16 '16 at 1:54
• But humans have intuition like of things. Mathematicians often see the solutions before they make the rigor proofs. Human could find some logical way around the halting problem, like use induction, series or something like that. Computers can too, to some degree, if they know the algorithm. I used Gödel because I assume that behind automation systems there is always a set of deterministic axiom guiding them, like logic. Fundamentally this would come to a case of whether there exists anything illogical, that could be done by people, like finding new axioms. Can human find it, or can computers? – user3644640 Oct 17 '16 at 6:12
• The answer 'No' means that there are no jobs that can't be automated. That is contradicted by your statements. I think you mean 'Yes', there are jobs that can't be automated. – kingledion Oct 21 '16 at 1:19
• @kingledion I don't think that highly on humans ;) Even the behavioral irrationality is fairly consistent, which is kind of the idea of behavioral operations research. Even the comprehension of seeing is pattern mapping. – user3644640 Oct 21 '16 at 6:58

Yes. The job of a consumer.

You might haven't thought about it as a job, but consumers make our society (or at least our economy) work. With no consumption there is no point creating goods, and without creating and selling goods economy would collapse.

Let's say in 50 years - due to the advanced automation -, ten percent of mankind is required to provide goods, bank services and everything else for the (let's say) ten billion people. Now you might think the rest would become obsolete, and they might be left to die. But let's just assume, this happens! Nine billion people left unemployed, so have no salary. It would also cause nine billion consumers disappearing. Without them, industry, agriculture and services can work at 10% efficiency, which means, sooner-or-later 90% of the working people would lose their jobs. Our scenario becomes like... a hundred million people with jobs, nine hundred million new unemployed and nine billion left on the roadside long ago, but... if there are only a hundred million consumers, 90% of the producing capacity is obsolete...

... I think you see where it is going now.

• There's a flawed assumption here. If all industries are automated, then society is moving towards post-scarcity, where the concept of a consumer being required to drive the economy disappears. – Frostfyre Oct 15 '16 at 14:16
• @Frostfyre Even if society takes a socialist/communist turn - which is nowhere near necessary - consumers are still needed as a motivation. Automated factories are only required to produce goods if the goods are being consumed, automated customer services are only necessary because they have customers, automated research centers are only required to make the lives of customers better. Without consumers, almost the entire automated infrastructure goes into "sleep mode" and maybe slow decay. – mg30rg Oct 15 '16 at 15:08
• @mg20rg > I like the answer, because it is clever, and out-of-the-box, and makes people think originally. I don't think the exact scenario is really likely though. (1) People might want more than the amount that was previously enough for ten billion people. (2) Or, it could be that the state makes the consumption choices for the 90%, who get the bare minimum for survival in welfare and aid payments. The other 10% are consumers who get as much as they can imagine or afford. Also, consider that with things like iTunes Genius playlist, consumption choices are starting to be automated. – user24000 Oct 15 '16 at 15:25
• @mg20rg: or, to take an extreme distopian devil's advocate position: what if nobody helps the 90% (or 100%) of humans to consume, but the robot economy produces goods solely to keep robots and machinery running? – user24000 Oct 15 '16 at 15:35
• @user24000 I - as a software developer - have some insigt on the topic, and the extreme distopian devil's advocate position you are assuming has an enormous flaw... it isn't only evil programming, but also bad craftsmenship. Power saving is one of the main goals the entire industry is reaching out towards. I could imagine an automated factory breaking lose and murdering consumers to save power and resources, but one that keeps producing goods with all the storages full? - That is impossible. :D ps.: can you check my (user)name again? – mg30rg Oct 17 '16 at 7:54

Given sufficient time (more than the amount you give, but I get the sense that the exact amount of time isn't what you're looking for)...

The relevant question isn't "What jobs can't you make a machine capable of doing?"...

because I see no reason to assume that an AI can't reach human and greater intelligence.

...it's "What jobs can't you make a machine do?"

Because if you make an AI exactly like a person, it becomes able to decide for itself and say "No." At that point, the AI will start wanting to delegate work to lesser machines.

The other half of this question is, what jobs intrinsically require self-motivation? I'm not sure if that question can be answered without testing.

Ok so there are a few occupations that could not be automated. Massage therapist for instance. You can have a robot perform manual therapy, but it's effects on the human body would not be the same as human touch. I would also lean toward a judge or lawyer also requiring human experience in order to be able to make a fair decision that considers human experience factors in making such a decision.

• How would a machine not be able to apply the same effect through massage as a human? As mentioned in comments to another answer about massage therapy, machines that perform massages already exist. – Frostfyre Oct 15 '16 at 14:49
• @Frostfyre maybe in the sense of a force that only flows through living beings or so. – Ghanima Oct 16 '16 at 22:14

NOT LIKELY IF

you limit the meaning of job to things that we humans can do and jobs that are developed out of these kind of jobs, as far as we know, we ourselves are computers that react to our environment, and so it is possible to built an AI that is as capable as we are and that can do what we can do. Even if our current way to built computers won't allow for AI that could keep up with us, there is no reason why we shouldn't be able to rebuild brains themselves.

What you could do is define a job as every job that is thinkable, that would include for example making a cable with infinity lenght, this would be as impossible for AI as for us.

So back to your question, every job could be done by AI as well as humans can do it and better, however it probably will never happen, because humans like to create and work on stuff.

If we reach that lvl of automation, it will likely be that every job, that isn't creative and that no one really wants to do, will be done by machines and humans will do the rest, protected by law to not be outrun by AI systems.