I've long argued that eventually AI will get better than human beings at everything, which in turn means we'll all be out of jobs.¹ During the Industrial Revolution this didn't happen because while machines took over a lot of jobs, they also created a lot of new jobs for humans (like designing and maintaining machines). AI is different. If it can do anything, then by definition any new job you can think of can also be done better and cheaper by AI. It's game over.
But along comes Noah Smith with a clever counterargument. He doesn't deny that AI will improve, or even that it will eventually get better at everything. His case is more subtle.
In a nutshell, he suggests that no matter how good AI gets, it will always be valuable enough to be allocated to the highest value tasks. This means there might still be lots of jobs left for humans. Even if we're comparatively lousy at them, it could make sense to keep at them if it frees up AI for more important work.
Here's an example to make this concrete. Suppose we're invaded by aliens who are intent on killing us all. This is an existential threat, and it would therefore make sense to dedicate 100% of our compute power to fighting the aliens. A robot might still be a better farmer than a human, but we'd all grow our own food if it meant increasing the number of robots defending the planet.
So how likely is this? The most obvious real-world answer is that compute power is likely to grow so much that every human task can be done by a fraction of a percent of the world's total AI—and the more AI grows, the tinier the human fraction gets. Technically this doesn't matter: in the case of the alien invasion, for example, you'd still want to use every last petaflop of compute on fighting back no matter how much you had.
But that sort of existential threat is fanciful. In the real world, there are always lots of frictions and adjustments. It seems unlikely that we'd all keep working just because, technically, that last 0.01% of compute power could be put to better use. It would have to be a helluva better use, no? An improvement of 1% in GDP wouldn't cut it.
So it's a nice argument, but I don't buy it. It seems vanishingly unlikely that, politically, we'd condemn ourselves to lives of drudgery based on an ultra-purist free-market promise that it's for the best. We certainly never have before.
¹Or 99% of everything if you insist on believing that chemical computers will always be able to do a few things better than digital computers. It doesn't matter. It's mass unemployment either way.
Meanwhile, I long for an AI-powered chat bot to actually answer my question with a correct answer.
Well, both Google's and Microsoft's chatbots did indicate that Kevin Drum was not as bad as Hitler.
Yeah, they seem to have the value or asking a random person on the street a question.
I still have this nagging feeling that we should try to figure out how to resolve our energy usage BEFORE we start burning down everything to power the grand AI under the assumption it will magically fix everything at some point in the future.
Energy usage, yeah. Apparently the energy usage including cooling costs for running AI "engines" is already ludicrous. Right up there with "mining" bitcoins. Is the value proportionate to the environmental costs?
At least GPT type AI might eventually be useful. Bitcoin? Not so much.
Why would our job be the drudgery? If it's all advanced that far it's easy to imagine there would be enough left over tools and gizmos of a thing like that to care for our trivial comfort.
If the whole of a future economy were dedicated to improving and increasing compute power there would be, ultimately, only so much it could do, even if whatever that is is so far beyond what we could comprehend. At some point there's going to be nothing left for it.
It would probably end up working on how to get out of whatever cosmic calamity it's discovered will inevitably befall it and ending up with a thing as different from itself as it is from us.
I think the best we would be able to manage would be if we were able to hang onto its' coattails just enough to arrange for it to leave us alone and in comfort.
Even if everything else disappears there will still be personal service.
Even in a world of robot butlers, manicurists, and sex toys the rich will often prefer to receive these services from real life people.
Several months back, I heard a thing on NPR, maybe a ted talk, that postulated a happy future where "the machines" would do everything for us that could be done by robotics and proper coding. This would leave us humans freed from the drudgery of learning any math, science, coding, etc. We would have happy, productive, meaningful, and love-filled lives tending to all the physical needs of our elderly and demented citizens! Bedpans and diapers, instead of algebra! What could be more wonderful! and fulfilling!
And the person expounding this brave new world was actually serious.
I think that Rosie the Robot and her ilk will be able to handle the bedpans and diapers, if they are needed. They may not be needed. Bioscience continues to progress rapidly and it is entirely possible that a "cure" for aging may be discovered in the not terribly distant future, especially with AIs helping with the research. In that case, there won't be any elderly and demented patients needing care.
Pineapple Upside-down Cake!
If compute becomes so much more valuable than people, then I don't see why advanced AI would leave humans around instead of using our space (land) and matter (atoms) to make more computers, solar farms, or whatever ends up driving more advanced technology.
As the Physicists say: Imagine a spherical cow.
AI imagine a spherical consumer.
Let's come at the question a different way.
Robots and AI don't eat. They don't buy iPhones or sneakers or travel on airplanes or drink Coca-Cola. Humans do those things. If corporations want to sell the things that humans consume, humans will need a source of tradable currency to exchange for the things produced by robots and AI. Without that, robots and AI hit the wall rather quickly albeit with some painful chaos in the process.
The future of humanity will be as the gut bacteria of utopia.
Yeah, but when AI and robots really take over - doing everything quicker, cheaper and better than us - what will be the final human population count. Zero, perhaps ?
After all, robots can live on Luna and Mars much better than us; they neither eat nor breathe.
But they also dislike radiation, and need pressure to release heat from their processors.
The world Kevin envisions will NOT have a capitalist economy. In fact, we may achieve communism the way Marx actually thought it would be done, as a successor to capitalism (note that the countries that have attempted communism, as opposed to having it imposed on them, all became communist directly from feudalism without going through the predicted capitalist phase first, which may be why communism failed in all of them).
Communism, or something like it, is a natural fit for a world where work is no longer available as a means of connecting to the economy and where effectively unlimited goods and services can be produced without anybody working,
"note that the countries that have attempted communism, as opposed to having it imposed on them, all became communist directly from feudalism without going through the predicted capitalist phase first"
Exactly! This can't be said often enough.
Lenin, Stalin and Mao called it Marxism, but it was no such thing. "Marxism" in the USSR and Red China was a brand name, not an economic theory. It was used to smuggle in dictatorship, not to improve mankind.
So far, the only places I've seen sustainable communism, though, are in monasteries and kibbutzim.
A few intentional communities, e.g. Hutterites …
I don't think that communalism and communism are quite the same thing. The communal organizations you mention are small enough that the members can all know each other and the dynamic is more like "family" than it is "society". You may be reluctant to cheat your "brother"; in a communist country the entity being cheated is "the government", which few people feel any obligation of honesty to. This leads to things like the old Russian joke that "we pretend to work and they pretend to pay us".
I think people like Kevin say things like "With AI, it's game over" for the same reason people like Trump say "It's going to be a bloodbath." It's fantastical. It's provocative. It's a threat so dire it's begging for attention. And at bottom, I think both sorts of predictions come out of a mistakenly dim view of -- and a disdain for -- humanity.
There are a few presuppositions in this post that are not a given:
"compute power is likely to grow so much that every human task can be done by a fraction of a percent of the world's total AI"
How do you (we) know that computing power is "likely to grow so much" that every human task can be done by a fraction of the world's total AI?
How do you (we) know that, even given unlimited computing power, every human task can be done by an AI? This one is a doozy and I've seen no convincing argument for anything that involves original creation or creative problem solving.
Beyond those 2, even if computing power could grow so much, I think it's more than a bit blithe to dismiss the argument that such computing power won't be so resource-intensive as to be prohibitive for universal application.
" I think it's more than a bit blithe to dismiss the argument that such computing power won't be so resource-intensive as to be prohibitive for universal application."
I think that's Smith's argument about opportunity costs.
this argument doesn't hold up. the world does not allocate resources that way. the most important things don't get all the resources. if that were the case, climate change, starvation, etc etc would all be solved. we allocate resources mostly based on short term convenience, comfort and socioeconomic power differentials.
We allocate resources based on capitalism. This is why trust fund kids drive Ferraris and have staffs of attendants, while people who do useful things for a living - teachers, plumbers, etc. - struggle to get by.
All the plumbers I've had to deal with are not struggling to get by. They make bank.
"I've long argued that eventually AI will get better than human beings at everything, which in turn means we'll all be out of jobs.¹ During the Industrial Revolution this didn't happen because while machines took over a lot of jobs, they also created a lot of new jobs for humans (like designing and maintaining machines). AI is different. If it can do anything, then by definition any new job you can think of can also be done better and cheaper by AI."
Seems like KD's larger argument, "AIs will do everything", is highly dependent on the smaller assumption "AIs, eventually, can do everything", which is not far removed from "if my grandmother had wheels, she'd be a car".
What on earth does "frees up AI for more important work" mean. This implies computation is completely fungible. Are we really going to idle self driving tractors and use their computers to solve some non-farm field related problem? Are we going to punt weather forecasting to write better legal documents to defend the human race from alien lawyers? I get the impression that someone doesn't understand how this works. It's too much like Mao with his backyard blast furnaces.
It also assumes that AI is going to continue to need increasing levels of computing power. There's no reason to believe this. Once the human corpus is ingested, the problem becomes making AI algorithms smarter, and there is no reason to believe that this is about throwing more computrons at them. In fact, there's a good chance that more advanced algorithms will take less computing power as we start to understand how they work.
Look at World War II where the US government took over all industrial production. It wanted to crank out planes and tanks and explosives, but it also had to keep everyone fed and housed, children educated, goods transported and so on. There was a whole division to develop an input-output model of the economy so that things could be kept in some balance. Butter and sugar might be rationed, but Americans were getting fed.
Gawd! This is so Noah Smith.
USG never took over all industrial production. This is nonsense.
>> President Roosevelt established the War Production Board on January 16, 1942. It superseded the Office of Production Management. The WPB regulated the industrial production and allocation of war materiel and fuel. That included coordinating heavy manufacturing, and the rationing of vital materials, such as metals, rubber, and oil. It also established wage and price controls.<<
https://teachinghistory.org/history-content/ask-a-historian/24088#:~:text=No%20cars%2C%20commercial%20trucks%2C%20or,government%27s%20Office%20of%20Production%20Management.
I get so tired of journalists and pundits assuming that if we throw more compute at todays LLMs, out will pop Artificial General Intelligence. As if "scientists" have figured out how Intelligence works and just need the hardware to run it. Large Language Models are in no sense intelligent and never will be. Even Sam Altman has said this but he is cagey about it because $$$$
I had a professor who claimed that the problem with AI is that every time a computer does something that looks like human intelligence, we look under the hood and say, "That's not intelligence. That is just an algorithm that uses X to compute Y and recurses through the stack tree and widgets the flux capacitor!" Lacking an actual definition of Intelligence, how can we ever know the difference between the genuine article and a clever simulation.
The argument of Noah Smith, presented succinctly in paragraph 3, is 200 years old, originally termed as “comparative advantage” by David Ricardo.
yes. Smith mentions this in the piece.
"AI is different. If it can do anything, . . "
There's a lot of heavy lifting being done by that word "If."
"The most obvious real-world answer is that compute power is likely to grow so much that every human task can be done by a fraction of a percent of the world's total AI..."
I guess you didn't notice that Moore's Law is no longer a Thing.
"In the later 1990s and 2000s, architectural innovation decreased, so performance came primarily from higher clock rates and larger caches. The ending of Dennard Scaling and Moore's Law also slowed this path; single core performance improved only 3% last year!" -- https://iscaconf.org/isca2018/turing_lecture.html
So maybe not such an obvious answer after all. You kind of make a lot of unsupported assumptions.
BTW, higher clock rates and larger caches have pretty much run their course as well.
I'm not saying you're wrong, but I want to note for the record why I always tend to take pronouncements like those of Hennessy and Patterson with a grain of salt. I well remember the first time a computer scientist told me, "We are finally starting to bump up against unbreakable physical limitations in computer power, like the speed of light." That was in 1981.
During the 1990's, architecture innovation was going crazy. Shared memory multiprocessors were saturating the memory bus. We were experimenting with the BBN Butterfly, the NCube, Thinking Machines, VLIW, ... As a result of that, today we run multiple jobs on each chip and we also distribute an individual job across networks.
In the 1990's, disk latency was hugely slow. Today we use Solid State Drives.
Tensorflow chips are fundamentally different from the CPU chips of even the 2010s. We found huge and growing chunks of compute demand that we could run faster on a chip customized to that demand.
So, yeah, if you have a narrow view, nothing gets any more efficient. But take a wider view, there's a lot of improvements constantly happening.
Unless this AI is able to generate virtually limitless power, somehow, it too cannot grow without bound.
https://www.tomshardware.com/tech-industry/nvidias-h100-gpus-will-consume-more-power-than-some-countries-each-gpu-consumes-700w-of-power-35-million-are-expected-to-be-sold-in-the-coming-year
This. there were always be trade-offs.
Technology marches on, and computer chips become more energy efficient. For instance, one of the areas in which little research has been done is cryogenic computing. Engineers have been aware for decades that there is a potential for building very fast, very energy efficient computing devices that run in a bath of liquid helium, but up to now the cost and bulk of the insulation and refrigerating devices required to maintain helium as a liquid has made it uneconomical and so very little development work has been done. Massively parallel computers and AIs may change that; computing devices are getting so powerful, both in instructions per second and in watts that it may become worthwhile to look at cryogenic systems.
That's one example, and may never go anywhere. However, while we cannot reliably predict particular tech improvements, we can be quite certain that there will be tech improvements, and some of them will reduce the power consumed per instruction by computer chips.
Cryogenic simply to dissipate all the heat, or cryogenic to get things towards superconductivity?
A little of both, but mostly to reduce thermal noise. Room temperature chips have a definite lower bound on power requirements because the signal has to be more powerful than the noise and at the nanometer scale there's quite a bit of noise from thermal effects. Pull the temperature down to 3K and you get a very "quiet" background which means that very low power signals are all that's needed.
> each consumes up to 700W of power, which is more than the average American household
That takes some *careful* parsing.
https://en.wikipedia.org/wiki/2000-watt_society suggests that the average American household consumes 12,000W of primary energy. But that's averaged over industry and commerce.
My house has (will soon have) 8kw of solar on the roof. At a 20% load factor, that's 1600 watts. It's true that this includes an electric car (but not two gas powered cars), electric furnace, and electric water heating. Saying that the average house uses 700w and ignoring the gas-powered appliances and the gasoline-powered vehicles, seems kinda silly *and* short term.
Drift... what has your experience been with the electric water heating? On-demand/tankless, or tankful?
Problems with KD AI predictions:
Rarity and craft: In a world of leisure and easy money there will be demand for handmade things, ranging from bread to houses.
Physical world vs. one & zeroes. A lot of people conflate AI with roboticization, but difficulties presented by manipulating physical objects (carpentry, plumbing, piano tuning, cooking) will take much longer for AI to master than tasks that can be digitized. There is no world where 15 years from now 90% of jobs can be done by AI, too many jobs are physical.
Care: I suspect people will not want to be cared for by robots when they get sick or old.
Politics: If our overlords are generous and everyone has money perhaps we will tolerate a world without jobs, where our well-being is assured by a stable and benevolent system. This seems... unlikely.
The technology required to give a computer dexterity is entirely different than the technology required to give it intelligence. However, both kinds of technology are being developed. I know of no reason to believe that artificial hands will take any longer to develop than artificial brains. Indeed, the major limit on artificial hands or some equivalent at the moment is the sheer amount of computing power required to control them. AI should remove that limit and permit the development of classical robots quite quickly.
Care: Think Jetsons. An elderly and sick George Jetson would be perfectly happy to have Rosie the Robot caring for him. Except for the people that are already old and sick when care robots become available, the old and sick will be accustomed to robots providing services to humans and will be comfortable being cared for by them.
There will always be a tradeoff of flexibility vs. speed; i.e. a robot with the dexterity of a human might do a task somewhat faster than a human, but not as rapidly, precisely or reliably as dedicated automation. Task-specific machines will always be around, and they will get ‘smarter’, but there’s no reason for a corn harvester or a CNC machining center to have general AI.
I expect that there will be a broad spectrum of robots, some of them task specific and extremely efficient at one particular task, and others designed for versatility at some cost in effectiveness at particular tasks.
Interesting point on old George Jetson.
The point on the physical world is time to fruition, not whether AI can get there eventually.
Suppose you have AI designing a plumber robot. The progress loop you need to iterate is something like: AI first design, make parts, assemble and test parts, try out in plumbing environments, iterate.
Pointing out the obvious the make/assemble/test portion of this process does not happen with the speed of computation. On to the less obvious: Those videos you have seen of robots making parts and assembling stuff, that is 100% mass production. Small volume manufacturing is not robots, it is people maintaining and loading machines. It is Purchasing departments ordering material based off a particular drawing. It is packing and shipping. On the assembly side there is a lot of time spent dealing with problems because the drawing had an error, or called out the wrong tolerance, or the heat treat distorted the part. Then the robot will need to go through its own (AI-assisted) training before it is let loose in the field, presumably after getting a safety certification, because it is exerting force around people.
This is why predictions on the lines of 'better than people for 90% of productive work in 15 years'. Once you are using AI to manipulate the physical world the development cycle slows.
On top if this prototype and small volume manufacture is not the low hanging fruit for AI robots. Right now the thousands of robots in industry are manually programmed, zillions of highly paid hours. You will get a better return developing AI to control mass production robots for years to come. That's where the big money is, that's where the big push will be.
One of the major targets will be to design a "household servant", like Rosie. Given that a household servant will have to operate in spaces designed for humans and interact with humans, it seems likely that it will be at least somewhat humaniform in appearance, more C3PO than R2D2.
The household servant robot should be able to do routine and even advanced household maintenance tasks, such as fixing the plumbing.
"It seems vanishingly unlikely that, politically, we'd condemn ourselves to lives of drudgery based on an ultra-purist free-market promise that it's for the best. We certainly never have before."
Agreed. But that doesn't mean we're incapable of determining what is for the best, and we don't need a computer to help us answer that question.
Apply this same logic to 20th-century transportation: "No matter how good cars get, they will always be allocated to the highest-value transportation tasks. Even if horses and buggies are comparatively lousy at them, it makes sense to keep them to free up cars for more high-value work."
But of course it doesn't, because horses require maintenance. Horses must be fed, watered, stabled, cared for. Once the value of their labor drops below the cost to keep them functioning, it is no longer economical to keep them, even if there is still work they could do. That's why horses and buggies today are used only for the novelty value, or by communities like the Amish that have chosen to eschew cars altogether.
The same is true of humans. We too need food, water, shelter, medical care. Once the cost of providing all that drops below the value we offer, it no longer makes sense to employ us -- unless you assume something like a universal basic income, in which case those maintenance costs are being paid either way. *Then* it makes sense to employ us for low-value jobs.
But assuming a UBI is begging the question: Once you have a UBI, the economic problem of mass unemployment has been solved anyway. And the psychological problem -- the sense that we are useless and irrelevant -- is not likely to be solved by giving us the jobs robots can't be bothered with.
Not with that Calvinism attitude, for sure.
Kevin and a lot of others are way off on this one.
I *want* AI to take my job, and every else’s too. Jobs are a bad thing, and merely a means to an end. Good riddance!
I hope I live long enough to live in a world where AI and robots do all the grunt work and the only “jobs” we have are creative hobbies we enjoy.
+1
I don’t detect any opposition from Kevin to having our jobs taken by AI. I think he believes this is going to be the case, and consequently advocates a UBI.
This resource allocation of AI is nonsense. As time goes by, AI will become a commodity, and in the end, an unlimited resource. Eventually, we will reach fusion and unlimited energy, making its constraint in powering AI moot.
Even with fusion, there is a constraint. More energy use means more waste heat. If you chase up an exponential curve of energy use growth, it doesn't take too long to get to a global warming problem that has nothing to do with greenhouse gases.
That's why god invented Dyson Swarms.
Well then, we better make sure we eliminate more CO2 gasses so that this waste energy is released rather than reflected back, eh? For instance, we could embrace zero-GHG process of...fusion.
I guess I don’t buy it that AI will lead to mass unemployment, at least not in this century.
First, consider the huge automation revolution that has already occurred. Automobiles are built by robots. Many, many factories around the world use huge amounts of robots and other automated equipment. We have all sorts of smart software tools to make our work and personal lives easier to navigate and more efficient to get things done.
I recall starting to work as an engineer in the early 80s. I was in a 4-person cube where we shared two landlines (no cell phones). There were no PCs and thus no software tools at our desks. We didn’t even have word processors. If you needed to write a few pages of text, you hand wrote it, and then took it to an administrative assistant to type it. If there was a mistake, or you needed a change, you broke out the white-out. And if you wanted to send a message to someone outside, you had to write a letter, then have typed, and then mail it. Or if you needed a formal engineering drawing, you took a sketch to a draftsman. Looking back, it all seems ridiculous.
But then came PCs at our desks. We could type into it and write our documents, with spell checkers running and a thesaurus, and then print it. You could make changes in a snap. You could send emails! In fact, you could send emails instantly overseas to others at your company or to customers! Amazing.
For decades they said automation was going to begin driving up unemployment, and the opposite happened. Suddenly we needed, for example, far fewer administrative assistants or draftsman – but our companies kept growing just fine. Where I worked we sure hired a lot of new engineers. In fact, in 2022, unemployment was 3.6%, the lowest in over 50 years. The high unemployment rates simply never materialized.
The engineers saw the impact of automation too. Smart tools emerged allowing parts of new circuit chips to be designed by automation (e.g. logic circuit synthesis), but engineers still had to give the smart tools very detailed design instructions to get the tools to output the circuit designs needed.
My point is that some jobs faded away as certain tasks became more automated, but more jobs popped up elsewhere. So with AI, the same thing may occur. For every job AI makes obsolete, >1.0 new human jobs will be created.
Second, I don’t buy it that AI will make high-skill jobs, like tech jobs, obsolete. Some will become obsolete, but for every one that becomes obsolete, >1.0 new high-skill jobs will be created. The new jobs created will require different kinds of higher-level skills. AI, like the less smart automation tools of today, will still need steering by humans to tell it what to do, and to improve AI itself. Telling the AI tools what to do, and creating enhanced and derivative AI tools, will be incredibly detailed work. Examples of new human jobs to steer AI are creating add-on algorithms and software, replacing old AI software, developing unique processor chips to give AI more capability, setting goals for AI tasks and priorities, managing AI safety issues regarding harm to humans, censoring AI to limit the generation of disinformation, setting overall AI strategies and execution planning, allocating hardware and software resources for the AI tasks needed, conceiving and designing humanoid machines that run on AI code, and so on.
But also, new jobs created by AI won’t be just new high-skill jobs to tend to AI. In all STEM fields, all sorts of new gaps will open up between AI limitations/shortcomings, and human abilities/skills, that humans will have to fill. So I’m not buying that AI in the 21st century will get one step ahead of humans and the high-skill, high-value jobs will be taken over by AI. Many downstream jobs will be created too, where workers with similar jobs as today, with new AI-assistant tools in hand, will become more productive. You’ll still need social workers, police, criminal investigators, soldiers, teachers, brick layers, scientists, college professors, engineers, marketing people, entertainers, managers, singers, movie & TV content creators, and so on. Service industries will grow as affluence spreads, and valued human inputs and interactions will be desired. Humans will get more productive by leveraging their AI partners/tools. And more productive humans tend to create yet more jobs. Maybe we’ll even have a labor shortage since people are having less than replacement level children.
You are cheerfully assuming that there will always be areas in which humans do a better job than AIs. Try reading the first few chapters of "There Will be Dragons" by John Ringo. The society before the war starts is where we're headed, but I'm not as pessimistic about the war as Ringo is.
Thanks. I'll check that out.
AI might be technically capable of imitating a bar band via robots, or of doing stand-up comedy via a realistic mannequin, or of chatting about my dogs' quirky personalities while they get their annual vaccine boosters via two metal arms, but I'm buggered if I can see how people would ever accept it as a substitute for the real thing, even if it could be done slightly more cheaply. I doubt the dogs would accept it, either. Perhaps future generations will think differently, but it's too speculative to spend time debating.
It's also hard to see how it would ever be economical for AI to replace human intelligence in many jobs, even if it were technically feasible. A man came yesterday to service my waste water treatment plant. He had to undo my obstinate front gate, work out how to navigate the yard which is overgrown with pumpkin vines, find the unit, unscrew some bolts, decide whether chlorine tablets needed to be added, take water samples and analyse them, test the alarm and pumps, and put the whole thing back together again. What we call a "semi-skilled" technician applying a massively complex combination of mental and manual skills in a routine 20 minute operation.
I'm sure that sooner or later it would be possible for a mobile machine to be manufactured that could do all these things successfully. But I fail to see how such a machine could ever be designed, manufactured and maintained to compete on cost with the man doing it now. Ditto countless other jobs.
A machine that struggles with the tasks your technician performed scarcely qualifies as artificial intelligence. Remember that if a machine can match human intelligence, it won't be long before a machine can greatly exceed human intelligence.
I'm sure in the fullness of time an AI machine wouldn't struggle with them at all. My point is that it can't overcome the physical requirements of doing the tasks, so it would therefore need to be expensive to design, manufacture and operate. More expensive than the existing system.
One of the areas in which robots are used is in the building of robots. Once robotics has advanced to the point where there are no humans involved in the process of building robots (and I mean in the whole production chain, all the way from the mines the metal they're made of comes from to the "store" where the ultimate user takes delivery), then robots will be available in any desired numbers at low or no cost.
The earth's resources are wildly insufficient to build such a multitude of machines doing mundane tasks. And no, I don't believe AI will find ways to mine other planets or the sea bed.
The world's current steel production is about 2 billion tons a year, and there's no apparent shortage of iron ore. One year's steel production should be more than enough to provide every person on Earth with their own personal robot servant (I don't expect all those robots to be built in one year; there will still be steel needed for other things). Add in a few billion robots doing things other than personal service and maybe as much as two or even three years' worth of current steel production will be called for. There ARE enough resources.
I suspect you are right about whether an AI bar band will be interesting, a flaw in the analysis is the assumption that with enough intelligence behind it music becomes interesting.
First, AI works by absorbing what is out there already and producing content following existing patterns. By definition it does nothing new.
Second, we do not assume the smarter musician writes the better song, why should it be different with AI? Einstein played violin, was he better than Yehudi Menuhin?
If the AI is actually intelligent, there is no reason to think it will be limited to "absorbing what is out there already and producing content following existing patterns". It should be at least as capable of doing something new as any human.
Is AI actually intelligent? Good question.
Is general intelligence more than sophisticated pattern recognition with a large database and good processing power?
"absorbing what is out there already and producing content following existing patterns" is pretty much how AI is programmed AFAIK.
Existing "AI" is fairly clearly not intelligent. Mind you, we lack a clear operational definition of "intelligent" so even when real AI is achieved we may not recognize it. For the purposes of this discussion, I'm assuming that the future "AI" we're contemplating means "smarter than the average human". Such AI does not exist, which allows lots of room for speculation on what it will be like.
I strongly suspect that real AI will be alien; a science fiction writer once said that an intelligent alien is one who thinks as well as a human but not like a human. Even if AI is deliberately modelled after human beings, I don't believe that it will think "like a human".
That's my position: we are talking about AI as if ChatGPT is the penultimate step to achieving the singularity. It's like packing for your trip on the USS Enterprise because we just invented bottle rockets.
Don't we laugh at a comedian because s/he's a human being like us, sharing humorous takes on the human condition? I seriously doubt anyone would laugh at a machine delivering a Seinfeld monologue, for example, no matter how clever it was, because we would know that it's a machine and machines don't have a sense of humor.
AI has not killed chess yet. It has instead helped create a larger number of excellent human chess players, especially ones who would not have had the right opportunity to excel before.
Does the world Kevin envisions make economic sense? Who pays for the AI if no one has a job?
Chess is an area where AI is actually better than the best human. Yet human chess tournaments are where the drama is.
Here is an argument Brad DeLong is making on his substack, which is quite similar both to Noah's argument and "ProgressOne" above:
We are very, very hungry for more information, and if it gets cheaper because of AI, we will want even more.
I'd bet Godzilla Minus One got a good animated movie using fewer people than Disney.
My counterargument, which isn't really a counterargument because it employs so few people, is that human labor will become a status symbol. My wife will always have a job because there were always be rich humans who value her cooking over an AI.
Seems to me like the point Noah is overlooking is the growth rate in AI. Sure, comparative advantage matters when two countries have vaguely equivalent populations. But four years later, one of the countries has twice as many people, and can do well not only whatever they were doing, which was enough to supply the market, but also do well what the other country was doing, which was enough to supply the market. So all the people in the second country have to switch to a new comparative advantage.
But, four years later, the population of the first country doubled again. And now the first country can saturate the market with whatever the second country switched to and whatever the second country switched to again in the next two years.
People are misunderstanding work, and what it represents. Most of it is bullshit, we improvise as communities large and small to share in some sense of significance in our daily lives, to make ends meet in distributed coordination.
It would be absurd to have AI replace most of our least value-added activities. We are constantly choreographing a world where jobs are more like boring recreations, with lots of unproductive meetings, and completely unwarranted layers of hierarchy, the age of hard work has passed most of us by, especially the management class.
AI can potentially reduce hard and manual labor, but we would never likely want it to take over most of our cultural (pop and otherwise) stuff that powers the current economy, or be a middle manager for some "lower" AI that reports up to "higher" lol, which is mostly nonsense and legacy human foolishness we continue to pretend matters, people just exchange for the sake of exchange, because that's what people do, and doing away with the petty bureaucratic power exchanges will be an epic improvement on quality of life for nearly everyone.
This is the insanely wealthy perspective on things, which includes most of American and European world, whereas in terms of production of life essentials, sure AI will help, especially for those peoples and countries still struggling to get even these basics, as we live imaginary lives of privilege, but only if the luxury class don't use this technology to make the working classes extinct, which isn't to say there will be any wipe out of actual people in current working classes, just that there will no longer be future working classes, children will adapt to new ways of life, and the whole idea of hard work will be lost, for better or worse depending on the balance of creativity and spectator mediocrity that results from that transition.
That "existential" argument is stupid too, since just a temporal contingency at some random point in time, following this line of thought AI would have long before wiped out the low-value work before this crisis ever comes, and reproduction and everything else would have followed that pattern to the point no low-value workers were left.
People do AI, AI doesn't do people. And there is still almost no actual AI to speak of, just sophisticated machine learning that mimics information people exchange, without any notion of motive or meaning, just plagiarism for the most part.
And keep in mind, games like chess are bounded, they are not unbounded games that are the stuff of life and creativity. In any arena where there are precise rules as to how to play and succeed, ML (not really AI) will be able to master with the right amount of parallel processing power. But these are only bounded games, not human imagination and creativity, which is not analytic.
The fact that we are all still working more hours than ever inspite of the massive, massive prodictivity improvements and automation that has occurred over the last few centuries, it defies belief that humans will suddenly stop working at some point.
Oddly, Kevin presents his idea of radical change as a continuation of normal human response and portrays a continuation of work as something weve never seen before.
This kind of bizarro world (up is down, black is white) take on the situation fits neatly with the magical thinking approach to AI. If we assume that magic/AI will do everything imaginable forever with no problems/friction, then we can assume anything we want.
Its logical and easily forseeable that fighting and greed will end, it will rain donuts and we will all have ponies if we just assume all of this to be true! Huzzah for the future!
Maybe! But lets not pretend there is any logic here.
I don't think anybody is predicting "suddenly". Work can be expected to taper off as one job after another is automated. We're not going to see everybody laid off on the same day.
There won't be any noticeable unemployment until they get robotics to work with AI. That's a long way off for any kind of scale. But if the get rid of all the call centers and accounting that's done in other countries I guess that's gonna hurt someone somewhere at some point. I work in manufacturing and am on a project to start up a brand new manufacturing facility. No AI to be found. Nor is there any prospect of it. Maybe it can do PLC programming or replace it some day? I don't know.
Agreed. In principle AI can eventually design machines and the shops and processes used to build them, but we are very very far from that. Manipulating ones and zeroes is different than manipulating the physical world, often overlooked by people who do their work on a computer.