I think it's interesting to think about the timing of these events: it seems like AI will initially impact skilled labour a lot more, but will also initially be a complement rather than a substitute to skilled labour (for example, it seems like right now, the demand for software has gone up a bit, and while new software opportunities are low, employed software engineers are doing fine), and by the time we get the substitution effect, we will also start to see AI capabilities impacting unskilled labour, where I expect it to be more uniformly a substitute (since the demand for physical goods seems more fixed). Anyway, I found the initial result fairly counterintuitive, and thus found this to be a very interesting read and will continue to think about it more.
"Heckscher-Ohlin’s basic prediction is that countries will specialize in goods that intensively use what they have in abundance, while importing goods that depend on factors that they lack. In plain English, countries will cook recipes which use the ingredients sitting in the cupboard. Bangladesh, where most workers lack an extensive formal education, exports t-shirts; Taiwan, which is rich in highly educated engineers, specializes in microchips." And many countries like India that have workers with formal education & highly educated engineers, are unable to be competitive garment exporters nor can manufacture microchips – so much for the model!
Interesting. Is there a version of this model that accounts for the possibility of an AI-supercharged Robotics boom devaluing unskilled labor too? I think that in a maximalist prediction of AI capabilities (without ASI, that is), only second key question here will matter.
Yes, we also looked at the case where only unskilled work was automated (e.g., great bricklayer robots but not legal AI tools), and a case where both skilled and unskilled work is automated.
In the LLMs plus super-charged robot case, human wages are pinned to the cost of the machine/LLM which can do their job and convergence vs. divergence is decided entirely by who owns compute and whether its rents are shared. We don't model the timing, but with a delay between automating cognitive and physical work, there's a chance for wages to converge before robotics arrive. I'm very confused by robotics progress, so I don't know how long that window could be!
“If AI automates skilled labor that rich countries are relatively rich in, wages will converge between rich and poor countries—an unexpectedly progressive effect”
Is it? It seems like if this is happening because the price of skilled labour is dropping, not any increase in unskilled labour, except relatively.
The rest then goes to the rentier classes who own shares in AI companies.
I realise this model is flawed but even this progressive outcome is a disaster for skilled workers. Distributing money to share holders even in the global south (as a solution) isn’t all that progressive either. This all seems like a reversion to feudalism.
Wage convergence comes in part from unskilled wages rising in absolute terms, not just relative to falling skilled wages. This is due to a) falling prices of skill-intensive goods and services, and b) the unskilled sector maintaining its share of income as absolute production rises without bound (this falls out of Cobb-Douglas; it could move with more general substitution).
I agree that "progressive" is over-promising. In the model, convergence is in part because skilled workers in both North and South lose their skill premium. But the class of losers I'm mainly worried about is anyone without a claim on AI rents. Real wages would be up massively for everyone, but labor's share of income would collapse and inequality would also be up massively.
I think it's interesting to think about the timing of these events: it seems like AI will initially impact skilled labour a lot more, but will also initially be a complement rather than a substitute to skilled labour (for example, it seems like right now, the demand for software has gone up a bit, and while new software opportunities are low, employed software engineers are doing fine), and by the time we get the substitution effect, we will also start to see AI capabilities impacting unskilled labour, where I expect it to be more uniformly a substitute (since the demand for physical goods seems more fixed). Anyway, I found the initial result fairly counterintuitive, and thus found this to be a very interesting read and will continue to think about it more.
"Heckscher-Ohlin’s basic prediction is that countries will specialize in goods that intensively use what they have in abundance, while importing goods that depend on factors that they lack. In plain English, countries will cook recipes which use the ingredients sitting in the cupboard. Bangladesh, where most workers lack an extensive formal education, exports t-shirts; Taiwan, which is rich in highly educated engineers, specializes in microchips." And many countries like India that have workers with formal education & highly educated engineers, are unable to be competitive garment exporters nor can manufacture microchips – so much for the model!
Interesting. Is there a version of this model that accounts for the possibility of an AI-supercharged Robotics boom devaluing unskilled labor too? I think that in a maximalist prediction of AI capabilities (without ASI, that is), only second key question here will matter.
Yes, we also looked at the case where only unskilled work was automated (e.g., great bricklayer robots but not legal AI tools), and a case where both skilled and unskilled work is automated.
In the LLMs plus super-charged robot case, human wages are pinned to the cost of the machine/LLM which can do their job and convergence vs. divergence is decided entirely by who owns compute and whether its rents are shared. We don't model the timing, but with a delay between automating cognitive and physical work, there's a chance for wages to converge before robotics arrive. I'm very confused by robotics progress, so I don't know how long that window could be!
I am not well versed in robotics either but the headlines seem to point to the (possible) automation of skilled work first.
This post presents an interesting alternative to this post I read a week ago:
https://www.chat-gdp.org/p/the-staggering-welfare-consequences
EDIT: Skimming through it again, I can see that you are definitely aware of it.
“If AI automates skilled labor that rich countries are relatively rich in, wages will converge between rich and poor countries—an unexpectedly progressive effect”
Is it? It seems like if this is happening because the price of skilled labour is dropping, not any increase in unskilled labour, except relatively.
The rest then goes to the rentier classes who own shares in AI companies.
I realise this model is flawed but even this progressive outcome is a disaster for skilled workers. Distributing money to share holders even in the global south (as a solution) isn’t all that progressive either. This all seems like a reversion to feudalism.
Wage convergence comes in part from unskilled wages rising in absolute terms, not just relative to falling skilled wages. This is due to a) falling prices of skill-intensive goods and services, and b) the unskilled sector maintaining its share of income as absolute production rises without bound (this falls out of Cobb-Douglas; it could move with more general substitution).
I agree that "progressive" is over-promising. In the model, convergence is in part because skilled workers in both North and South lose their skill premium. But the class of losers I'm mainly worried about is anyone without a claim on AI rents. Real wages would be up massively for everyone, but labor's share of income would collapse and inequality would also be up massively.
Hi Oliver, KarthiK, and Joseph. Great post. Have you considered crossposting it to the EA Forum? I can do it myself if you prefer.
Please go ahead! Thanks!
Thanks. I plan to publish the post on Tuesday. I will let you know once it is live.
Published (https://forum.effectivealtruism.org/posts/oZy86oCmANNJzfwpu/what-will-ai-do-to-global-inequality).
Will AI eliminate illiteracy, or will it just put lipstick on a pig?