On my understanding of Korinek and Suh, wages don't go to zero, they get set equal to the return on capital. The assumption is that capital has low returns because of Malthusian constraints: it can just keep duplicating until the returns equal the cost of duplication.
But if a larger capital stock also makes technological progress go faster, which seems reasonable if we're talking about AI that replaces all human tasks, then you can never duplicate capital fast enough and the returns to capital grow forever. Thus, wages also grow forever.
There is another faster way for wages to fall to zero. The firm becomes a worker-owned and controlled coop--one worker, one share, one vote. Managers are picked from the workers themselves, and no worker can sell their share except to another incoming worker. Now look at the books: there are no wages paid. The worker-owners get a monthly draw on their anticipated annual profit shares, evened up or down at the end of the year. The worker-owners are no longer wage workers, but profit-sharing owners. Mondragon in the Basque Country is the case in point, doing well and growing for 70+ years now. The worker-owners there, by the way, do not fear robots. One of their coops, Itsthus Engineering of Madison, WI, makes robots.
I think this is missing the point, and would likely make workers more in danger if anything. The return to capital is the return to saving, which we can assume is spread evenly. (If it’s in a bank account, say, it doesn’t much matter who the bank lends to, because the interest rate will be dragged up by AI companies). Turning workers into part-owners of a cooperative doesn’t protect them against the company as a whole being outmoded. (As an aside, I don’t think labor cooperatives are the ideal way to organize a firm, and everything done to fix them is just to turn them into a normal corporation with bells and whistles).
> "If the rate of technological change persistently outstrips that of This is the logic of Anton Korinek and Donghyun Suh’s paper “Scenarios for the Transition to AGI”."
The way I try to explain this to people is that there is a rate of return on investment in human capital, for institutions, governments, and firms investing in employees leads to more productivity and compounding growth.
However, with AI we see a reversal in the marketplace for agentic labor, inhuman capital investment will eventually yield better returns than investment in human capital.
When an agent (either organic or inorganic) is trained and the costs of agentic labor are of concern, it's the replicability and scalability that is important. In humans you see compounding network effects in cities or academic institutions where human capital breeds human capital development in other agents, but it still requires long timeframes for returns on initial investment. You can't develop the human capital of a graduate student and then just have them engage in mitosis.
But inhuman capital investment does not require this, it is replicable and scalable as long as compute resources exist. In the past it required joint human development to receive the full return on investment, but now that is likely no longer going to be the case.
Provocative, timely, great read. Thanks.
On my understanding of Korinek and Suh, wages don't go to zero, they get set equal to the return on capital. The assumption is that capital has low returns because of Malthusian constraints: it can just keep duplicating until the returns equal the cost of duplication.
But if a larger capital stock also makes technological progress go faster, which seems reasonable if we're talking about AI that replaces all human tasks, then you can never duplicate capital fast enough and the returns to capital grow forever. Thus, wages also grow forever.
You’re right, I conflated decline with “go to zero”.
There is another faster way for wages to fall to zero. The firm becomes a worker-owned and controlled coop--one worker, one share, one vote. Managers are picked from the workers themselves, and no worker can sell their share except to another incoming worker. Now look at the books: there are no wages paid. The worker-owners get a monthly draw on their anticipated annual profit shares, evened up or down at the end of the year. The worker-owners are no longer wage workers, but profit-sharing owners. Mondragon in the Basque Country is the case in point, doing well and growing for 70+ years now. The worker-owners there, by the way, do not fear robots. One of their coops, Itsthus Engineering of Madison, WI, makes robots.
I think this is missing the point, and would likely make workers more in danger if anything. The return to capital is the return to saving, which we can assume is spread evenly. (If it’s in a bank account, say, it doesn’t much matter who the bank lends to, because the interest rate will be dragged up by AI companies). Turning workers into part-owners of a cooperative doesn’t protect them against the company as a whole being outmoded. (As an aside, I don’t think labor cooperatives are the ideal way to organize a firm, and everything done to fix them is just to turn them into a normal corporation with bells and whistles).
> "If the rate of technological change persistently outstrips that of This is the logic of Anton Korinek and Donghyun Suh’s paper “Scenarios for the Transition to AGI”."
outstrips that of capital accumulation?
Yep yep. Was revising how I phrases things, and didn’t tie up all the loose ends.
The way I try to explain this to people is that there is a rate of return on investment in human capital, for institutions, governments, and firms investing in employees leads to more productivity and compounding growth.
However, with AI we see a reversal in the marketplace for agentic labor, inhuman capital investment will eventually yield better returns than investment in human capital.
When an agent (either organic or inorganic) is trained and the costs of agentic labor are of concern, it's the replicability and scalability that is important. In humans you see compounding network effects in cities or academic institutions where human capital breeds human capital development in other agents, but it still requires long timeframes for returns on initial investment. You can't develop the human capital of a graduate student and then just have them engage in mitosis.
But inhuman capital investment does not require this, it is replicable and scalable as long as compute resources exist. In the past it required joint human development to receive the full return on investment, but now that is likely no longer going to be the case.