This article is remarkable for two reasons. First, three of the economists it interviews work for an MIT-affiliated body called The Institute for the Future of Work. It is their job to be global leaders in thought on this issue. They are the best of the best the academic discipline of Economics has to offer. Secondly, this article puts the central question to them unequivocally. - “What happens after AI/robots are capable of doing all work (even new as yet uninvented jobs), but are much cheaper to employ than humans?”
The central argument they’re replying to is that an economic concept called comparative advantage means jobs are safe from AI. The argument is that as computers will remain a scarce resource, super-intelligent AI won’t want to waste resources on doing “lesser” work, and will leave that to less capable humans. It hasn’t occurred to the economist proposing this idea that computing won’t be a scarce resource as it’s always getting cheaper, and more powerful.
One, David Autor, says AI won’t ever be better than humans, instead, it will give all humans new skills, so that everyone will have the economic advantages the highly educated now have. Another, Ethan Mollick, is at least honest in admitting economists don’t know what the future will be like when AI can do all work. One of the IFTFW economists, Pascal Restrepo, agrees with the idea of comparative advantage. He says AI will create vast wealth, but even the crumbs the owners of that AI give to humans to do the lesser work AI is not interested in will make us all richer than we are today.
So in summary. One economist who isn’t aware computers are getting cheaper. Another who doesn’t think AI will get better. Another who doesn’t know what will happen (the most honest of the bunch). And another who’s cheerful about the prospect of our future economy being a new feudalism where AI’s owners benignly let the peasants (everyone else) subsist on the leftover crumbs they don’t want.
The premise thrown to those economists is slightly different - that AI is more productive than a human at everything, but not necessarily cheaper, then sends their different interpretations, which i’ve ended up summarizing as:
Noah Smith: Compute will remain limited, so it’ll be directed at whatever is most profitable.
David Autor: AI can make the big-bucks jobs available to everyone, but they still need to Git Gud and figure out how to harness AI for them… Until everyone does. Pray “everyone” gets small enough by then so there will still be enough bucks left to live.
Daron Acemoğlu: Rest easy. AI won’t cover everyone’s jobs any time soon. Things will still suck when it does, though.
Ethan Mollick: AI won’t cover all jobs, but it’s about to cover all the “good” jobs. I don’t want to think what happens next.
Noah Smith on the earlier comments: Society will hopefully sort itself out and still leave humans “some” role. Relax.
Pascual Restrepo: You’ll still have “some” role, buuut it will become irrelevant anyway as AI will play a colossally larger role in the overall economy. Pray AI’s profits get distributed to everyone.
I think all of this misses the point completely. Compute can still be limited, but i think it’s more likely to happen through the supply-demand curve. Even if it remains expensive forever, there will likely be a point where the value of AI will equal its compute cost - once that is reached all AI jobs pay the same per compute time AI spends on them. Under the comparative advantage umbrella, the best-paying jobs for a human would turn out to be the jobs AI is the most inefficient at relative to the human regardless of whether they’re fullfilling to the human or not. but this still misses a much bigger problem.
Production needed land, labor, and starting from a certain technology level, capital. There used to be a social balance here in that everyone had their own labor and society always managed to figure out the price they’d sell their labor for - even if that price was different over time and over different societies, they’d still manage to sell it anyway because there was someone willing to buy it in the first place.
With AI, capital can now become labor instead of only amplifying production per labor. The industrial revolution was merely an inflection point where the amplifying factor of capital made labor without an amount of capital out of reach for most individuals uncompetitive. The ARA revolution is merely an inflection point where capital is making human labor uncompetitive.
On old societies, everyone had access to enough land to survive(or thrive) on their own labor with a capital multiplier small enough to be achievable by oneself - you could sell the labor to yourself and get a more enjoyable life in exchange. As societies got larger, the land available per individual got smaller, but capital could still allow for success. People with large amounts of land/capital figured out they could get more by lending their land/capital to others in exchange for labor, and society could figure out the exchange rate for which they were willing to trade those because labor was scarce.
ARA is bringing a post-scarcity economy, but the post-scarcity is coming to labor first, society has grown large enough that there’s not enough land for everyone to survive under their own labor without access to capital, and the people with land/capital now have enough that they don’t need to buy labor anymore. Worst of all, they’re now worse off by letting others labor with their resources for free, because the cost to keep them alive is now higher than the benefit their labor would bring to them. Society doesn’t like this, but what can it do about it?
This article is remarkable for two reasons. First, three of the economists it interviews work for an MIT-affiliated body called The Institute for the Future of Work. It is their job to be global leaders in thought on this issue. They are the best of the best the academic discipline of Economics has to offer. Secondly, this article puts the central question to them unequivocally. - “What happens after AI/robots are capable of doing all work (even new as yet uninvented jobs), but are much cheaper to employ than humans?”
The central argument they’re replying to is that an economic concept called comparative advantage means jobs are safe from AI. The argument is that as computers will remain a scarce resource, super-intelligent AI won’t want to waste resources on doing “lesser” work, and will leave that to less capable humans. It hasn’t occurred to the economist proposing this idea that computing won’t be a scarce resource as it’s always getting cheaper, and more powerful.
One, David Autor, says AI won’t ever be better than humans, instead, it will give all humans new skills, so that everyone will have the economic advantages the highly educated now have. Another, Ethan Mollick, is at least honest in admitting economists don’t know what the future will be like when AI can do all work. One of the IFTFW economists, Pascal Restrepo, agrees with the idea of comparative advantage. He says AI will create vast wealth, but even the crumbs the owners of that AI give to humans to do the lesser work AI is not interested in will make us all richer than we are today.
So in summary. One economist who isn’t aware computers are getting cheaper. Another who doesn’t think AI will get better. Another who doesn’t know what will happen (the most honest of the bunch). And another who’s cheerful about the prospect of our future economy being a new feudalism where AI’s owners benignly let the peasants (everyone else) subsist on the leftover crumbs they don’t want.
The premise thrown to those economists is slightly different - that AI is more productive than a human at everything, but not necessarily cheaper, then sends their different interpretations, which i’ve ended up summarizing as:
Noah Smith: Compute will remain limited, so it’ll be directed at whatever is most profitable.
David Autor: AI can make the big-bucks jobs available to everyone, but they still need to Git Gud and figure out how to harness AI for them… Until everyone does. Pray “everyone” gets small enough by then so there will still be enough bucks left to live.
Daron Acemoğlu: Rest easy. AI won’t cover everyone’s jobs any time soon. Things will still suck when it does, though.
Ethan Mollick: AI won’t cover all jobs, but it’s about to cover all the “good” jobs. I don’t want to think what happens next.
Noah Smith on the earlier comments: Society will hopefully sort itself out and still leave humans “some” role. Relax.
Pascual Restrepo: You’ll still have “some” role, buuut it will become irrelevant anyway as AI will play a colossally larger role in the overall economy. Pray AI’s profits get distributed to everyone.
I think all of this misses the point completely. Compute can still be limited, but i think it’s more likely to happen through the supply-demand curve. Even if it remains expensive forever, there will likely be a point where the value of AI will equal its compute cost - once that is reached all AI jobs pay the same per compute time AI spends on them. Under the comparative advantage umbrella, the best-paying jobs for a human would turn out to be the jobs AI is the most inefficient at relative to the human regardless of whether they’re fullfilling to the human or not. but this still misses a much bigger problem.
Production needed land, labor, and starting from a certain technology level, capital. There used to be a social balance here in that everyone had their own labor and society always managed to figure out the price they’d sell their labor for - even if that price was different over time and over different societies, they’d still manage to sell it anyway because there was someone willing to buy it in the first place.
With AI, capital can now become labor instead of only amplifying production per labor. The industrial revolution was merely an inflection point where the amplifying factor of capital made labor without an amount of capital out of reach for most individuals uncompetitive. The ARA revolution is merely an inflection point where capital is making human labor uncompetitive.
On old societies, everyone had access to enough land to survive(or thrive) on their own labor with a capital multiplier small enough to be achievable by oneself - you could sell the labor to yourself and get a more enjoyable life in exchange. As societies got larger, the land available per individual got smaller, but capital could still allow for success. People with large amounts of land/capital figured out they could get more by lending their land/capital to others in exchange for labor, and society could figure out the exchange rate for which they were willing to trade those because labor was scarce.
ARA is bringing a post-scarcity economy, but the post-scarcity is coming to labor first, society has grown large enough that there’s not enough land for everyone to survive under their own labor without access to capital, and the people with land/capital now have enough that they don’t need to buy labor anymore. Worst of all, they’re now worse off by letting others labor with their resources for free, because the cost to keep them alive is now higher than the benefit their labor would bring to them. Society doesn’t like this, but what can it do about it?
tl;dr: George was almost right.