Epistemic status: my first solo post here, so critiques of both substance and form are welcome. Also, note that, while I wrote this and all of the content is mine, it was lightly edited by AI (Fable 5) for clarity before posting.
For calibration: the future described in AI 2040: Plan A sits around the 95th to 99th percentile of how well I expect things to actually go. If that's the future we get, I'll be grateful. This post is about the one part of it I think we should refuse to treat as an acceptable default.
The part I disagree with is the Epilogue (and the picture in the Space Governance Supplement behind it), in which the pre-AGI wealth distribution ends up reflected in permanent property rights over cosmic resources, for instance with some people buying up others' galaxies, and a permanent underclass, albeit one that is rich by 2026 standards, is presented as a tolerable cost of an otherwise wonderful outcome. The Epilogue acknowledges this outright:
"Many people aren't interested in the space lottery, so when they receive the tickets, they sell their tickets for money on the open market to people who value control over space. Somewhat uncomfortably, this leads to the wealthy having disproportionate control over cosmic resources. But it's hard to avoid: if people are allowed to trade their control over the stars for Earth assets, then people wealthy in Earth assets inevitably end up disproportionately influential, and proposals for extreme redistribution of Earth assets have already been rejected as politically infeasible."
Note the load-bearing conditional: the wealthy end up in control because the claims can be traded for Earth assets. Tradability is a design choice, not a law of nature, and that choice is one of the key things this post disputes.
I think that's incorrect on the merits, and I think publishing it as tolerable makes it more likely to happen. Policymakers read these documents; in fact, Scott Alexander explicitly mentions that the team moved the space material out of the main text partly out of worry about how Washington readers would take it, and it is well-documented that prior reports like AI 2027 were read at the highest levels of government. To be fair, moving the material out of the main text softens this concern somewhat too; but it remains published, linked, and already being written about. What documents like this treat as acceptable shape the eventual Overton window of what policies are more likely to get implemented.
(Some people disagree with this entire premise because they argue property rights won't survive the transition at all; see On Owning Galaxies. Here I'm taking the Plan A world at face value and arguing about what we should aim for within it.)
The stakes are lifespans, not mansions
It's tempting to read cosmic inequality as "some people get more luxury goods than others." But 1000 galaxies contain on the order of 1000 times the negentropy of one galaxy, and negentropy is what ultimately buys years of subjective existence, compute for your own cognition, and, if you're a utilitarian, the total happiness you can realize in your slice of the lightcone. Luxury consumption mostly runs into steeply diminishing returns: the marginal value of mansion square footage falls off fast. Years of life are not obviously like that, and the number of years on offer scales close to linearly with the resources you control. There's some chance superintelligence finds a loophole in the second law of thermodynamics, but I wouldn't count on it.
So the Epilogue's arrangement is less "you get fewer toys than the rich" and more "you are permanently sentenced to die roughly 1000x sooner, because of how wealth happened to be distributed in the particular decade AGI arrived." Yes, the one-galaxy person still lives an extremely long time. But permanent, unalterable inequality is still permanent, unalterable inequality.
Rich underclasses still die of poverty
The developed world today is unimaginably richer than the world of 500 or 1000 years ago. Poor people in rich countries mostly don't die of the old causes: mass famine, plague. And they genuinely do have it better than the poor of past centuries. But people still die of poverty in rich countries; it just looks different, because removing the old bottlenecks to survival exposed new ones (for example, rationed medication, homelessness, treatable illness left untreated).
I expect the same pattern after AGI. Future poverty won't look like today's, and it will be milder. But the permanently poor of that world will, as discussed above, still face real suffering, and real deaths, that more resources would have prevented. Locking people into permanent, unalterable relative poverty, forever, is something to avoid, not something to strive towards.
The qualifier gets dropped
There's also a memetic problem with "a permanent underclass is fine as long as everyone is rich by 2026 standards." Arguments get compressed as they travel, and the qualifier is the first thing to go. In real political contexts, I think there's a real risk this degrades into "a permanent underclass is fine, period." The political world does not operate in the same way as the rationalist world, and we don't control how a message evolves once we put it out there.
Lock-in creates bad incentives right now
People really, really do not want to end up in a permanent underclass, and they will work very hard to stay out of one. If the accepted picture of the future is "the music stops at AGI and you keep whatever you're holding," the individually rational move is to grab as much as possible before the music stops: take reckless bets, cut corners, fight anything (including safety measures) that might delay your payday. That's close to the opposite of the behavior we want in the run-up to AGI, and I think we're already seeing some of that with the "escape the permanent underclass" memes going around.
To their credit, the supplement's authors see this failure mode: they point out that if people expect the future to be divided up according to bargaining power, they'll spend the intelligence explosion jockeying for position instead of cooperating to reduce risk. My claim is that an Epilogue in which pre-AGI wealth ends up translating into cosmic holdings does exactly that.
The Epilogue does gesture at a correction: a plan where 10% of resources go to people who behaved altruistically. I don't find this reassuring. Faced with the prospect of permanent, frozen inequality, people's natural response is to accumulate, not to be good and hope for a prize. Any "reward the altruists" scheme also seems unusually easy to Goodhart. And as a prediction, this resources-for-altruistic-behavior part feels like one of the least likely parts of the piece.
This is a policy choice, not an inevitability
I agree with the authors that massive wealth redistribution is, and will likely remain, outside the Overton window. But avoiding a permanent underclass does not require redistribution. Two ideas, from a few minutes of thinking:
The first idea is to make the shares inalienable. Give everyone their allocation of space resources and prohibit transferring it, directly or indirectly: no sales, no pledging it as collateral, no thousand-year leases. The supplement already floats a waiting period before people can sell, and describes even that as somewhat paternalistic. I'd go further: for the resources that determine how long you get to live, don't allow the sale at all. There's plenty of legal precedent for inalienability: spendthrift clauses in irrevocable trusts, bans on organ sales, the non-assignability of Social Security benefits and most pension rights in the US. In contrast, the Epilogue's design (distribute claims broadly, then let them be sold freely) has precedent too, but in a bad way: it is the design of post-Soviet voucher privatization, where ordinary Russians received shares of state companies, largely sold them cheap, and watched control concentrate into a few hands within a decade. The obvious objection is that this blocks mutually beneficial trades. But we already accept inalienability where what's being sold is a person's own future; you can't sell yourself into slavery either, however voluntarily. And I doubt the resources would lie fallow, at least not in the medium or long term: I strongly suspect nearly everyone would find some use for their share, even something as simple as "mine it and send the output home to Earth."
Another option (not mutually exclusive with the first) is to simply not freeze everything at t=AGI. If political reality demands that some initial tranche be allocated in a way that reflects existing wealth, fine, but leave most resources unallocated, to be distributed by mechanisms that stay open over time. My version: allocate the first 1000 light-years now, and let each later shell of resources, (1000n, 1000(n+1)] light-years out, be assigned by some future competition or civic process. They themselves float something a bit similar, distributing 10% of resources now, 10% in a century, 10% in a millennium, and so on. If people can augment themselves with resources they already hold, you'd need an adjustment mechanism to keep later rounds fair, and you'd want the contest itself to be positive-sum and avoid Moloch-type dynamics. But both feel like solvable design problems rather than impossibilities, especially since, with multiple rounds of allocation, any issues can be patched between rounds, unlike a one-time reward, which cannot be corrected once granted.
These are the first two ideas I came up with; better ones surely exist. The point is that permanent lock-in is not inevitable. It's a preventable policy decision.
To be clear one final time: if we get the future in Plan A, Epilogue and all, I'll count us lucky. But "I'd take it over most futures" and "we should consider it acceptable" are different claims. A permanent underclass, however "rich" by today's standards, is worth preventing if we can prevent it. And we can.
Epistemic status: my first solo post here, so critiques of both substance and form are welcome. Also, note that, while I wrote this and all of the content is mine, it was lightly edited by AI (Fable 5) for clarity before posting.
For calibration: the future described in AI 2040: Plan A sits around the 95th to 99th percentile of how well I expect things to actually go. If that's the future we get, I'll be grateful. This post is about the one part of it I think we should refuse to treat as an acceptable default.
The part I disagree with is the Epilogue (and the picture in the Space Governance Supplement behind it), in which the pre-AGI wealth distribution ends up reflected in permanent property rights over cosmic resources, for instance with some people buying up others' galaxies, and a permanent underclass, albeit one that is rich by 2026 standards, is presented as a tolerable cost of an otherwise wonderful outcome. The Epilogue acknowledges this outright:
Note the load-bearing conditional: the wealthy end up in control because the claims can be traded for Earth assets. Tradability is a design choice, not a law of nature, and that choice is one of the key things this post disputes.
I think that's incorrect on the merits, and I think publishing it as tolerable makes it more likely to happen. Policymakers read these documents; in fact, Scott Alexander explicitly mentions that the team moved the space material out of the main text partly out of worry about how Washington readers would take it, and it is well-documented that prior reports like AI 2027 were read at the highest levels of government. To be fair, moving the material out of the main text softens this concern somewhat too; but it remains published, linked, and already being written about. What documents like this treat as acceptable shape the eventual Overton window of what policies are more likely to get implemented.
(Some people disagree with this entire premise because they argue property rights won't survive the transition at all; see On Owning Galaxies. Here I'm taking the Plan A world at face value and arguing about what we should aim for within it.)
The stakes are lifespans, not mansions
It's tempting to read cosmic inequality as "some people get more luxury goods than others." But 1000 galaxies contain on the order of 1000 times the negentropy of one galaxy, and negentropy is what ultimately buys years of subjective existence, compute for your own cognition, and, if you're a utilitarian, the total happiness you can realize in your slice of the lightcone. Luxury consumption mostly runs into steeply diminishing returns: the marginal value of mansion square footage falls off fast. Years of life are not obviously like that, and the number of years on offer scales close to linearly with the resources you control. There's some chance superintelligence finds a loophole in the second law of thermodynamics, but I wouldn't count on it.
So the Epilogue's arrangement is less "you get fewer toys than the rich" and more "you are permanently sentenced to die roughly 1000x sooner, because of how wealth happened to be distributed in the particular decade AGI arrived." Yes, the one-galaxy person still lives an extremely long time. But permanent, unalterable inequality is still permanent, unalterable inequality.
Rich underclasses still die of poverty
The developed world today is unimaginably richer than the world of 500 or 1000 years ago. Poor people in rich countries mostly don't die of the old causes: mass famine, plague. And they genuinely do have it better than the poor of past centuries. But people still die of poverty in rich countries; it just looks different, because removing the old bottlenecks to survival exposed new ones (for example, rationed medication, homelessness, treatable illness left untreated).
I expect the same pattern after AGI. Future poverty won't look like today's, and it will be milder. But the permanently poor of that world will, as discussed above, still face real suffering, and real deaths, that more resources would have prevented. Locking people into permanent, unalterable relative poverty, forever, is something to avoid, not something to strive towards.
The qualifier gets dropped
There's also a memetic problem with "a permanent underclass is fine as long as everyone is rich by 2026 standards." Arguments get compressed as they travel, and the qualifier is the first thing to go. In real political contexts, I think there's a real risk this degrades into "a permanent underclass is fine, period." The political world does not operate in the same way as the rationalist world, and we don't control how a message evolves once we put it out there.
Lock-in creates bad incentives right now
People really, really do not want to end up in a permanent underclass, and they will work very hard to stay out of one. If the accepted picture of the future is "the music stops at AGI and you keep whatever you're holding," the individually rational move is to grab as much as possible before the music stops: take reckless bets, cut corners, fight anything (including safety measures) that might delay your payday. That's close to the opposite of the behavior we want in the run-up to AGI, and I think we're already seeing some of that with the "escape the permanent underclass" memes going around.
To their credit, the supplement's authors see this failure mode: they point out that if people expect the future to be divided up according to bargaining power, they'll spend the intelligence explosion jockeying for position instead of cooperating to reduce risk. My claim is that an Epilogue in which pre-AGI wealth ends up translating into cosmic holdings does exactly that.
The Epilogue does gesture at a correction: a plan where 10% of resources go to people who behaved altruistically. I don't find this reassuring. Faced with the prospect of permanent, frozen inequality, people's natural response is to accumulate, not to be good and hope for a prize. Any "reward the altruists" scheme also seems unusually easy to Goodhart. And as a prediction, this resources-for-altruistic-behavior part feels like one of the least likely parts of the piece.
This is a policy choice, not an inevitability
I agree with the authors that massive wealth redistribution is, and will likely remain, outside the Overton window. But avoiding a permanent underclass does not require redistribution. Two ideas, from a few minutes of thinking:
The first idea is to make the shares inalienable. Give everyone their allocation of space resources and prohibit transferring it, directly or indirectly: no sales, no pledging it as collateral, no thousand-year leases. The supplement already floats a waiting period before people can sell, and describes even that as somewhat paternalistic. I'd go further: for the resources that determine how long you get to live, don't allow the sale at all. There's plenty of legal precedent for inalienability: spendthrift clauses in irrevocable trusts, bans on organ sales, the non-assignability of Social Security benefits and most pension rights in the US. In contrast, the Epilogue's design (distribute claims broadly, then let them be sold freely) has precedent too, but in a bad way: it is the design of post-Soviet voucher privatization, where ordinary Russians received shares of state companies, largely sold them cheap, and watched control concentrate into a few hands within a decade. The obvious objection is that this blocks mutually beneficial trades. But we already accept inalienability where what's being sold is a person's own future; you can't sell yourself into slavery either, however voluntarily. And I doubt the resources would lie fallow, at least not in the medium or long term: I strongly suspect nearly everyone would find some use for their share, even something as simple as "mine it and send the output home to Earth."
Another option (not mutually exclusive with the first) is to simply not freeze everything at t=AGI. If political reality demands that some initial tranche be allocated in a way that reflects existing wealth, fine, but leave most resources unallocated, to be distributed by mechanisms that stay open over time. My version: allocate the first 1000 light-years now, and let each later shell of resources, (1000n, 1000(n+1)] light-years out, be assigned by some future competition or civic process. They themselves float something a bit similar, distributing 10% of resources now, 10% in a century, 10% in a millennium, and so on. If people can augment themselves with resources they already hold, you'd need an adjustment mechanism to keep later rounds fair, and you'd want the contest itself to be positive-sum and avoid Moloch-type dynamics. But both feel like solvable design problems rather than impossibilities, especially since, with multiple rounds of allocation, any issues can be patched between rounds, unlike a one-time reward, which cannot be corrected once granted.
These are the first two ideas I came up with; better ones surely exist. The point is that permanent lock-in is not inevitable. It's a preventable policy decision.
To be clear one final time: if we get the future in Plan A, Epilogue and all, I'll count us lucky. But "I'd take it over most futures" and "we should consider it acceptable" are different claims. A permanent underclass, however "rich" by today's standards, is worth preventing if we can prevent it. And we can.