AI forecasts span a range of potential futures, from economic stagnation to explosive growth. The divergence traces to three specific assumptions—each generating predictions we can already test.
Excellent article. Missing one very key component. Ai agents and robots will soon be widely available for purchase. Lending will spring up in lock step. Many of us already have multiple agents working for us providing orders of magnitude more leverage than anyone planned for.
Financial will soon allow for one to purchase a robot or agent teams and send it to work for you. Not just knowledge work. Robots are already doing an increasingly large number of jobs once thought to be the sole province of humans. For the first time one can own and control the means of production personally—not through share ownership.
Open Ai, Anthropic, Google, Boston Scientific, et,al are building the work force of the future.
Many people would rather not work. Others would like to work less. Some want to get more done.
Welcome to the world of Ai leverage and the new slave class.
The framing here is useful. Most of the quadrillion-dollar estimates depend on assumptions about deployment speed and labor market absorption that AI governance frameworks have not yet caught up with. The disagreement is not really about the number. It is about whether institutions can absorb that change without breaking. Nobody has a credible model for that part.
Regarding Acemoglu's easy/hard task dimension ("distribution"), and even the task-level analysis itself seem unfortunately misguided to me. I'd rather analyze how much the *jobs* themselves (not tasks!) require "offline apprehension", which includes actually doing something (nurse, construction worker) as well as interactions or understanding that is not easily moved online (investigator inspecting the crime scene, journalist interviewing people at an event). I think purely online jobs (and all the tasks within them) will be crushed, but jobs that require "offline apprehension" will be automated much slower, even if they are not complex. I would not overestimate the speed of massive displacement of workers with robots, and there are big doubts in whether that will be economically rational in the majority of places and for the majority of jobs. The restructuring of workflows around people who will do narrow remaining "offline apprehension" pockets of the jobs and the core intelligence part happening online (e.g., an investigator assistant who wears smart glasses and follows AI commands at the crime scene) will happen on 10-20 year horizon, not on the few years horizon, as more aggressive economic forecasts assume.
Another "key question" that I would add if whether consumption/demand could match the production capacity's (projected, potential) explosive growth, especially considering the pessimism related to job displacement and general economic and political uncertainty. Citrini Research has described a relatively radical scanario ( https://www.citriniresearch.com/p/2028gic ), but even in "milder" scenarios, demand could still limit the supply growth. This "key question" is moot if a growing share of economy becomes "AI endogenous", that is, AIs consuming goods and services and making 'conscious' computations for their own welfare, but this prospect deserves a whole separate discussion.
Excellent article. Missing one very key component. Ai agents and robots will soon be widely available for purchase. Lending will spring up in lock step. Many of us already have multiple agents working for us providing orders of magnitude more leverage than anyone planned for.
Financial will soon allow for one to purchase a robot or agent teams and send it to work for you. Not just knowledge work. Robots are already doing an increasingly large number of jobs once thought to be the sole province of humans. For the first time one can own and control the means of production personally—not through share ownership.
Open Ai, Anthropic, Google, Boston Scientific, et,al are building the work force of the future.
Many people would rather not work. Others would like to work less. Some want to get more done.
Welcome to the world of Ai leverage and the new slave class.
There will be two groups owners and owned.
The framing here is useful. Most of the quadrillion-dollar estimates depend on assumptions about deployment speed and labor market absorption that AI governance frameworks have not yet caught up with. The disagreement is not really about the number. It is about whether institutions can absorb that change without breaking. Nobody has a credible model for that part.
Congratulations
Great overview, very clearly written.
Regarding Acemoglu's easy/hard task dimension ("distribution"), and even the task-level analysis itself seem unfortunately misguided to me. I'd rather analyze how much the *jobs* themselves (not tasks!) require "offline apprehension", which includes actually doing something (nurse, construction worker) as well as interactions or understanding that is not easily moved online (investigator inspecting the crime scene, journalist interviewing people at an event). I think purely online jobs (and all the tasks within them) will be crushed, but jobs that require "offline apprehension" will be automated much slower, even if they are not complex. I would not overestimate the speed of massive displacement of workers with robots, and there are big doubts in whether that will be economically rational in the majority of places and for the majority of jobs. The restructuring of workflows around people who will do narrow remaining "offline apprehension" pockets of the jobs and the core intelligence part happening online (e.g., an investigator assistant who wears smart glasses and follows AI commands at the crime scene) will happen on 10-20 year horizon, not on the few years horizon, as more aggressive economic forecasts assume.
Another "key question" that I would add if whether consumption/demand could match the production capacity's (projected, potential) explosive growth, especially considering the pessimism related to job displacement and general economic and political uncertainty. Citrini Research has described a relatively radical scanario ( https://www.citriniresearch.com/p/2028gic ), but even in "milder" scenarios, demand could still limit the supply growth. This "key question" is moot if a growing share of economy becomes "AI endogenous", that is, AIs consuming goods and services and making 'conscious' computations for their own welfare, but this prospect deserves a whole separate discussion.