I find this piece persuasive but was hoping for more specifics about how we might shore up unemployment insurance. Should we dramatically extend how long unemployed workers can take UI, for example?
It also didn’t address the possibility that AI completely replaces human labor eventually.
I think you are underestimating the problem and overestimating our current unemployment system.
First, AGI is simply not like past technological disruptions. Automating some particular set of jobs allows the affected people to eventually find new jobs, so ideally they just need some assistance during that transition. When you invent AGI smart enough to do literally any job humans can for far less than it takes to feed and house a human, people will not be finding new jobs, and you will need a radically different structure to have this not be catastrophic.
As to unemployment insurance, it of course does help in some cases, but many people fall between the cracks with our current system, and even those who are eligible can only get it for about half a year and only while applying to jobs every week. If AGI has eliminated all the jobs so there's nothing to apply to, that's a problem. Suppose you removed the time limit and the application requirement, now you have a permanent underclass based on how your last job ended. Oh, you were a college student who didn't get their first job before AGI hit, or you quit a job to try some contract work? Sorry, you get nothing. Okay, so let's remove that criteria... hey look we just reinvented UBI.
I think the point being made here is that the evidence does not support your AGI scenario as the likely outcome right now. If that happens, UI is not going to work. But it’s not at all clear that this is happening.
I like the suggestion of automatic stablizers for unemployment insurance. While were at it, lets add them for social security too. I don't get why automatic stablizers aren't more popular, especially with a Congress that doesn't want held responsible for making hard choices.
Gimbel is right that we shouldn't get cute with solutions when the impact trajectory remains uncertain. Shoring up unemployment insurance is sensible and actionable.
But here's what troubles me. The METR benchmarks on AI task completion tell a different story than the labor market research. All the labor market research is backwards looking, so it's anchored to the AI capabilities as they exist. We're watching these models move from completing tasks that take humans 30 minutes last year to tasks that take 4-5 hours today. https://www.linkedin.com/posts/metr-evals_we-estimate-that-on-our-tasks-anthropics-activity-7408005652431847424-H5iH/
The curve is bending upward fast. The application layer is still getting built out, which explains why we don't see mass displacement yet. Companies are still figuring out how to integrate these tools into workflows. But the underlying capability is compounding in ways that don't show up in BLS data until they do.
The Industrial Revolution comparison cuts is helpful but AI's compression of this timeline is precisely what makes it such a unique technology. The question isn't whether the economy will eventually absorb displaced workers into new occupations. The question is what happens in the interim. The uncertainty itself is the argument for thinking bigger, not smaller. Strengthening UI is necessary but not sufficient. What we need is a framework for asking the harder question: how does a democratic society ensure that the gains from compounding intelligence are distributed broadly rather than captured by those who control the models? That's not a question UI can answer. It's a question about power, ownership, and what kind of economy we want to build on the other side.
"But the disruptions didn’t cause us to ban factories, and here we are, two centuries later: richer and more fully employed than ever. (I feel sure I am much happier as an economist pontificating on the internet than I would have been as a weaver.)"
I feel like this is yada-yada'ing tectonic upheavals in social systems across europe/america and decades of labor rights movements, some of which resulted in actual revolutions happening in some countries.
This is a great reminder of the virtues of our UI system, and of the importance of having a good UI system in a healthy society. Credit to socialists within the Democratic party way back when, for establishing this in a number of states during the early 1930s, and to our great socialist former president, FDR, and the socialist Democrats in Congress at the time, for codifying it nationally shortly thereafter.
We absolutely need this, and we need it to work well. I totally agree with this essay.
However, there's no reason not to *also* adopt a universal basic income, in addition to UI. Saudi Arabia, of all places, has had such a system in place for decades. Saudi citizens earn a substantial share of their nation's oil revenues. The amount was something like $40,000 per person, per year while I was there during the early 2000s. I'd have to look up how that has changed to mention anything about how it is now — but here I'd just like to note how well it worked for them 20 years ago. It absolutely revolutionized the quality of their lives, and is something all wealthy nations should consider, in my opinion.
UBI should work by replacing the remaining manual labor government jobs with mandatory 10 hrs of community service per week and a reduction of full time benefit qualifications to an additional 20 hrs.
The biggest danger is our work becomes a place to inefficiently hang out where extroverts are having a blast and introverts want to die.
I find this piece persuasive but was hoping for more specifics about how we might shore up unemployment insurance. Should we dramatically extend how long unemployed workers can take UI, for example?
It also didn’t address the possibility that AI completely replaces human labor eventually.
💯 This calls for a follow-up, for sure.
Same. I assume the piece was going to tell us more about UI reforms, but then it just ended.
I also would’ve liked to see more policy specifics.
I think you are underestimating the problem and overestimating our current unemployment system.
First, AGI is simply not like past technological disruptions. Automating some particular set of jobs allows the affected people to eventually find new jobs, so ideally they just need some assistance during that transition. When you invent AGI smart enough to do literally any job humans can for far less than it takes to feed and house a human, people will not be finding new jobs, and you will need a radically different structure to have this not be catastrophic.
As to unemployment insurance, it of course does help in some cases, but many people fall between the cracks with our current system, and even those who are eligible can only get it for about half a year and only while applying to jobs every week. If AGI has eliminated all the jobs so there's nothing to apply to, that's a problem. Suppose you removed the time limit and the application requirement, now you have a permanent underclass based on how your last job ended. Oh, you were a college student who didn't get their first job before AGI hit, or you quit a job to try some contract work? Sorry, you get nothing. Okay, so let's remove that criteria... hey look we just reinvented UBI.
I think the point being made here is that the evidence does not support your AGI scenario as the likely outcome right now. If that happens, UI is not going to work. But it’s not at all clear that this is happening.
I like the suggestion of automatic stablizers for unemployment insurance. While were at it, lets add them for social security too. I don't get why automatic stablizers aren't more popular, especially with a Congress that doesn't want held responsible for making hard choices.
Gimbel is right that we shouldn't get cute with solutions when the impact trajectory remains uncertain. Shoring up unemployment insurance is sensible and actionable.
But here's what troubles me. The METR benchmarks on AI task completion tell a different story than the labor market research. All the labor market research is backwards looking, so it's anchored to the AI capabilities as they exist. We're watching these models move from completing tasks that take humans 30 minutes last year to tasks that take 4-5 hours today. https://www.linkedin.com/posts/metr-evals_we-estimate-that-on-our-tasks-anthropics-activity-7408005652431847424-H5iH/
The curve is bending upward fast. The application layer is still getting built out, which explains why we don't see mass displacement yet. Companies are still figuring out how to integrate these tools into workflows. But the underlying capability is compounding in ways that don't show up in BLS data until they do.
The Industrial Revolution comparison cuts is helpful but AI's compression of this timeline is precisely what makes it such a unique technology. The question isn't whether the economy will eventually absorb displaced workers into new occupations. The question is what happens in the interim. The uncertainty itself is the argument for thinking bigger, not smaller. Strengthening UI is necessary but not sufficient. What we need is a framework for asking the harder question: how does a democratic society ensure that the gains from compounding intelligence are distributed broadly rather than captured by those who control the models? That's not a question UI can answer. It's a question about power, ownership, and what kind of economy we want to build on the other side.
"But the disruptions didn’t cause us to ban factories, and here we are, two centuries later: richer and more fully employed than ever. (I feel sure I am much happier as an economist pontificating on the internet than I would have been as a weaver.)"
I feel like this is yada-yada'ing tectonic upheavals in social systems across europe/america and decades of labor rights movements, some of which resulted in actual revolutions happening in some countries.
This is a great reminder of the virtues of our UI system, and of the importance of having a good UI system in a healthy society. Credit to socialists within the Democratic party way back when, for establishing this in a number of states during the early 1930s, and to our great socialist former president, FDR, and the socialist Democrats in Congress at the time, for codifying it nationally shortly thereafter.
We absolutely need this, and we need it to work well. I totally agree with this essay.
However, there's no reason not to *also* adopt a universal basic income, in addition to UI. Saudi Arabia, of all places, has had such a system in place for decades. Saudi citizens earn a substantial share of their nation's oil revenues. The amount was something like $40,000 per person, per year while I was there during the early 2000s. I'd have to look up how that has changed to mention anything about how it is now — but here I'd just like to note how well it worked for them 20 years ago. It absolutely revolutionized the quality of their lives, and is something all wealthy nations should consider, in my opinion.
UBI should work by replacing the remaining manual labor government jobs with mandatory 10 hrs of community service per week and a reduction of full time benefit qualifications to an additional 20 hrs.
The biggest danger is our work becomes a place to inefficiently hang out where extroverts are having a blast and introverts want to die.