Poverty in the midst of abundance: AI will make goods cheaper, but your labor will get cheaper faster

cousin_it

Poverty in the midst of abundance: AI will make goods cheaper, but your labor will get cheaper faster — LessWrong

138 Poverty in the midst of abundance: AI will make goods cheaper, but your labor will get cheaper faster

by cousin_it

26th Sep 2026

1 min read

138

Very simple idea, but I thought it'd be worth making a post on this.

Some people are saying AI will make all goods cheaper, so you'll be able to afford a nicer life by working. Without any redistribution, just by market mechanisms. These people are wrong.

AI will lower the price of goods you need to survive, and also the price of your labor. The question is which will get cheaper faster. Let's use energy cost as a proxy. A day's worth of labor equivalent to yours can be done by AI for just a few cents in electricity on tokens. But feeding you with e.g. apples for a day will cost more energy than that, because growing apples can only be energy-optimized so far. So selling your labor at market price will leave you unable to afford apples.

This means a future with economic AI might look like "poverty in the midst of abundance": all goods are cheap, but somehow you can't find a job paying even that much. In economic terms, many previous inventions were complements to human labor, while AI is a general substitute for human labor, lowering its exchange rate vs other goods.

Maybe the problem can be solved by redistribution, or by everyone having investments, or something else - that's a bigger discussion. In this post I just want to make the basic point above.

AIWorld Optimization

Frontpage

138

New Comment

34 comments, sorted by

top scoring

Click to highlight new comments since: Today at 8:07 PM

[-]Dagon1mo5130

There are arguments to be made that TODAY is poverty in the midst of abundance. There are lots of people unable to find a job (or a "good" job), struggling with money, expecting never to be financially secure.

[+]marquis_de_sod1mo-100

[-]jmh1mo152

I think a bit deeper dig into the complement versus substitute might be merited. I question the claim that many (perhaps any) prior inventions were completements to human labor in the sense of not displacing some humans labor/work activity. Similarly, AI is a substitute for some human activities but complementary to others. I think one has to look at both sides before trying to conclude anything on this front.

I think looking into the history of some of the key inventions, how they disrupted (substitute) and how they enabled (complement) as well as just where we see both sides with AI will allow a much more sound assessment of the future. It might also point to what paths we might see evolving from here that seem to have a mitigating effect one whatever bad outcomes/impacts people are worried about.

I think you're correct in that, and on an increasing basis, we will see AI (and automation/robotics) performing jobs and tasks that people currently do and any person seeking to compete on the margins of that output somewhere will not be getting well paid. But that doesn't mean there will not be a sufficient number or other jobs AI and robots cannot do -- or that humans will not pay to have them do -- that human wages collapse to pure poverty for all. So some of the complementarity of AI and human works may be in the form of opening up activities that we don't have time for now (too much time and effort put into just making clothes, food, transportation, housing...) but that would be something people are willing to pay for.

[-]cousin_it1mo2317

AI is a substitute for some human activities but complementary to others

I think AI will be a pretty general substitute for human labor. It learns more jobs every year, faster than a human can learn. I haven't seen any plausible story which new human-only jobs would be able to support everyone.

[-]Mars_Will_Be_Ours1mo92

There might be a path for human survival under dismal conditions if it takes less energy to produce a human (GMOed to not need vitamins and get by on cheap nutrient slurry) along with their requisite food supply and use human labor than to accomplish the same feat using the energetically optimal combination of artificial intelligence and robot labor. Alternatively, if human level artificial intelligence proves too hard for a singleton to control and alignment problems persist, the singleton could use humans for low grade cognitive labor because we might be easier to control.

I cannot emphasize enough how dismal of an existence this would be, since a life under a misaligned superintelligence maximizing energy efficiency would be as impoverished as the poorest lives in sub-Saharan Africa while simultaneously far more totalitarian than the most oppressive regimes in human history thus far. If there's ever a life not worth living, it'd be this.

[-]jmh1mo30

I think you might be extrapolating from how things are today with specific tasks or activities over time (we don't see people manually tilling the soil any more but we do still see people doing farming -- and this is not an example to support my my suggestion but just an example of displacement that I suspect fits your view).

But I think if you start from a pretty basic truth, there is clearly one thing AIs cannot do that humans can: be human, then you might consider how the entire economy evolves and whether or not human labor might still remain a valuable economic good even it it's not doing what we currently think of income generating work. Add the, what I think is an obvious economic take, that AI output will be very large relative to what human output can be so the relative value of that human output compared to the AI output should be pretty large.

So while I don't rule out the possible outcome I don't see it as some forgone conclusion without making some, to me, rather unreasonable assumptions about the dynamics of the system. A bit what I was waving at with the comment about "or that people will buy".

Alternatively you might end up in a dual economy world where everyone gets some UBI that are used to buy the AI goods and then the pure human economy continues using whatever money was in place. People value live performances, human sports, human humor, circus, whatever show act -- or things like watching someone build the sofa you want to buy for your house and want to be sure it was made by a human. Mass production by humans goes away and you find a lot of local work becomes available because something made far away runs into a certification of "human made" issue.

[-]cousin_it1mo41

In the post I assumed no redistribution. Under these conditions, I don't see a plausible story which new jobs can support everyone. Be specific.

Maybe to demonstrate a bit more what we're up against, let's take your example of making sofas. Say AI can do this job for 1/10th of human survival wage. Then human-made sofas will have to cost 10x more than regular ones if the human is to survive. It'll be a bit hard to sell such sofas.

[-]zebrask1mo10

Hand-made artisanal goods already sell for dramatically more than mass produced equivalents, bespoke work is probably easily a 10x markup.

It is unclear if that kind of market can sustain all humans in an AI world, but it is not unheard of even today.

[-]dan.parshall1mo60

Restrepo's "We won't be missed" paper predicts that human wages will fall to the price of equivalent compute; combined with Binder's post about those costs... we may be below rice wages https://www.nber.org/books-and-chapters/economics-transformative-ai/we-wont-be-missed-work-and-growth-agi-world

[-]cousin_it1mo20

Yeah, totally. Binder's post is a great companion to mine.

[-]RogerDearnaley1mo60

In general, European countries have historically been more willing to do redistribution than the US. It may or may not be a coincidence that most major companies, including all the ones likely to profit massively off an AI transition, are headquartered in the US rather than Europe. As the importance of labor compared to capital decreases, the US labor market will be of less interest to companies, and being headquartered in a corporate tax haven will become more attractive. So don't be surprised if AI ends up nominally running out of a small tax haven.

Of course, physical control over the data centers may become important, and those are not generally located in small tax havens, at the moment.

[-]Garrett Baker1mo5-15

Increased productivity should mean stocks increase in value, so why not just invest?

[-]Karl Krueger1mo5443

A significant fraction of humans don't have surplus to invest, and have traditionally depended on a regular wage for their labor to cover their living expenses. This is typically called "living paycheck to paycheck" and affects about half the population of a wide swath of developed countries. People in this situation do not have the option to switch from living off their labor to living off investments.

[-]Davidmanheim1mo209

...and that's only people in developed countries. Most of the global population is far worse off.

[-]Garrett Baker1mo40

I agree with this, however the original post's conclusion was:

So selling your labor at market price will leave you unable to afford apples.

And I, and I believe most people on LessWrong, have the opportunity to invest, since we typically don't live "paycheck to paycheck". I agree that if the original post said "the average person's" or even "most people's" instead of "your", my response would not make sense. But it did say "your" and insofar as it is talking about "your labor" and 'your ability to afford apples' I think its wrong (given the property-rights and econ-as-normal assumptions its using).

[-]cousin_it1mo72

Hmm, I'm the OP and this wasn't the intended interpretation :-) My post was intended as a reply to breathless articles like this one, saying "you the typical person in the world will have a super-helpful AI friend and be happy". I'm basically saying "you TTPITW will be utterly screwed by the new order, don't listen to the lies".

[-]Garrett Baker1mo109

I think it is good form to clarify this is what you're saying in the OP then. As it is, given its assumptions, it is literally not true, and so just sounds like rehearsing a doom argument (which incidentally is why I think it has such a high upvote-count) rather than presenting a new argument. Maybe I'm being picky & overly literal, but I do think its important to make sure you're saying what you mean.

[-]cqb1mo25

I thought it was pretty clear that the collective you was referring to the typical/average/median person in the world.

[-]Rochambeau1mo3-7

The fact that about half of the population is "living paycheck to paycheck" means there are a great deal of people who do have the option to invest but choose not to. It simply cannot be the case that half the population's non-discretionary expenses exactly match their their earning potential (how could that possibly be the case across the almost continuous variables of pay and spending possible); instead their motivation to work and earn more/spend less creeps up as they go into debt and they let themselves spend more whenever they have a pay increase/as debt decreases.

How that motivation is changed is another question. I am not very comfortable with forced saving, but one solution would be required roth/401k contributions (or at the very least have the social security trust (and other country equivalents) be investing in productive companies rather than government bonds).

[-]RHollerith1mo*4-1

Because I don't want to help the AI labs or their suppliers in any way.

[-]nesiacel1mo41

This seems like an excellent argument for minimal UBI. Even if it's only nominal today, after enough singularity-driven deflation, everyone will be far richer in material terms than they are today.

[-]sudo-nym1mo32

This is the big thing the left-left worries about - the economics of labor currently do not look good, and if we don't get some sort of serious redistribution in place fast enough we could get genuinely destructive social unrest instead... and the fear is that governments will respond to destructive social unrest with increasingly draconian policing policies, the curtailment of basic freedoms, and AI-powered surveillance to enforce the previous.

You may have noticed that this describes a logical route to stable totalitarianism.

[-]esotericn1mo30

The general trend for the past few decades in my country has been a gradual devaluation of labour in comparison to capital assets.

What this generally looks like is not that people can’t eat or buy electronics or whatever but that scarce resources like land are very difficult to obtain.

I expect (short term) AI to accelerate this hugely such that food is still relatively inexpensive but that people who have pre-existing investments start to acquire more of the scarce resources.

[-]AnthonyC1mo20

Yes, this is a root of the reason arguments about comparative advantage don't help humans in a world of sufficiently advanced AI, for the same reasons they didn't protect horses in a world of trucks.

[-]Hastings1mo20

As an intuition pump, it may be easier to imagine a world where the vast majority of people don’t spend any money on goods that yield a wage for another person, than a world where the vast majority of people don’t earn any money.

A world where well off people buy human made goods but impoverished people only can afford to pay robots (and robots have no interest in human made goods- robot interest in human knickknacks helps a lot with this scenario) would be dangerously unstable.

One scenario:

Pressure A: people really don’t want their kids to be watched by robots, so a rapidly increasing fraction of all human wages come from childcare. (echoes of paying for daycare in Seattle)

Pressure B: whether or not they can watch a child well, near future robots will be able to keep a child alive and mostly sane for much less than it costs to pay a living wage to a nurse. (echoes of tablets)

As wages drop, the people who can still outcompete a robot have to work more and more hours, needing more childcare. For a long time they will be able to use some fraction of that money to pay a human childcare worker. That worker may be doing the same thing! Chains like this, of a parent providing expensive childcare while paying for cheap childcare, already exist and are weirdly long. When the pressure gets high enough, there could be a pretty sudden economic collapse as everyone runs out of human solidarity budget and switches to cheap robot childcare at once, up and down these chains.

[-]StanislavKrym1mo20

I wonder how similar the point is to The Intelligence Curse. As for the problem being solved, I doubt that I can think of a solution beyond redistributing the most basic resources and letting people give them to the robotic economy in exchange for directly usable goods...

[-]cousin_it1mo41

It'd need to be a lasting solution though. What if robots "buy out" a generation of humans, giving them +100 fun, and there are no more generations of humans.

That's why I prefer solutions like "everyone gets +50 intelligence". Slowing down AI enough that we can get IA. I think more and more people now are realizing that it's better, the only question is can we slow down AI enough.

[-]Dagon1mo20

the only question is can we slow down AI enough.

No, the other necessary question is can we actually amplify intelligence in any significant way. I don't know if there are even hints of how that might work, or how it'd be distributed among the (current and future) population of humans, AIs, and hybrid entities.

[-]cousin_it1mo20

Of course we can amplify human cognitive abilities a lot! Even without changing the physiology. John von Neumann was born to human parents; so was Mozart. And both were born randomly, so they were far from the peak.

[-]Dagon1mo20

Huh? I'm not asking if there's room to improve - of course there is. I'm asking if there is an achievable path to making this improvement much much faster and further than the Flynn effect.

[-]cousin_it1mo*20

I suspect we're missing some very low hanging fruit. Maybe some cognitive skills can be much improved even without genetic changes, just by pharma / biofeedback / electric stimulation of brain regions or something even simpler. For example, composers have vivid auditory imagination, but lots of people have vivid auditory imagination when they're about to fall asleep, so there might be an easy unlock. A lot of math skill seems to be about staying non-bored with math, so there might be an easy unlock too. In visual art, Betty Edwards' book gives an amazing one-day unlock of drawing skill just by making a certain change in how you look at things, I've done it myself.

[-]cubefox1mo12

Unfortunately intelligence amplification seems impossible without changing our brains. It would require selective breeding and gene editing and would take centuries.

[-]_rpd1mo20

distributed cognition just requires a performant neural interface

Moderation Log