My thoughts:

If the mechanism of private property tends to allocate capital to its most productive uses, then incentives are being aligned to put many people to work for common benefit

Much of your later arguments depend on this statement, but I don't think the conclusions follow from the premises. I make a long argument here that an optimal market as described by neoclassical economics does not utilize available resources in a utility-maximizing way, and that it is nearly always possible to improve the "common benefit" by allocating resources AWAY from their most productive uses. This is because from the market's perspective, the most productive use of a resource is the use that will command the highest price on the market. If a starving man has 1 dollar, but a rich man has 100 dollars to buy a sandwich to bury in the ground for his amusement, the more productive use of two slices of bread, a slab of teriyaki tofu, kale, tomato, and a spoonful of vegan mayo is to use it to make a sandwich for the rich man. 

This would in fact be ESPECIALLY true if all inequality resulted from differences in the productive capacities of different individuals' property. Because there are people who do not own anything productive (labor-power or otherwise), then those people would never benefit a cent from the most productive allocations of capital. 

You say:

From a global utilitarian perspective, having much more than others is not on its own a reason to transfer wealth to them. Instead, you should expect the return you can get on reinvesting your wealth into profit-yielding enterprises to frequently be higher than the return they can get, so you might be able to make a more important gift to the future than to the present.

I disagree. A hundred dollars to the poor is worth more than a thousand dollars to the rich. Higher returns in terms of money does not entail higher returns in terms of utility.

Even when there is a large enough market failure to justify philanthropy, some amount of paternalism is warranted, because your wealth advantage corresponds to a way in which you know better than them.

Again, market failures are not necessary for the market's distribution of resources to be suboptimal. And the latter statement doesn't seem to logically follow from anything. Just because your labor/capital is more productive, why should that mean you understand other people's preferences better than they do? 

Then, you say:

Under the means of production hypothesis, the answer was straightforward: when I buy a good, I am sending a price signal which causes some combination of reallocation of resources to produce more of that good, and the reallocation of that good and its inputs away from those with the least productive use for them. On balance I should expect such price signals to enrich those alleviating scarcity by improving the efficiency with which scarce goods are produced

Creating price signals by buying goods does enrich people, but only because it enriches yourself. If you consume more, you're redistributing productive resources away from other "less productive" uses, and towards your own uses. This is net-positive according to the economistic utility function, but there's no reason to believe it's net-positive from a utilitarian perspective, and it's pretty easy to think of situations where the reverse is true. For example, let's say that you're a billionaire and you want to spend your entire fortune on a pet project of yours: paying one million people a hundred thousand dollars each to build the world's largest pyramid in your honor, without being allowed to use any technology or sunblock. Not even one other soul in the world cares a single bit about this pyramid, but they want that hundred thousand dollars, so they agree to build it for you. This may intuitively strike you as magnanimous, maybe even altruistic. Everybody involved benefits, right? Thanks to your generous consumption, everyone is a whole lot richer! But this is an example of the broken window fallacy. The analysis ignores opportunity costs. If one million people are now spending their time carrying blocks of sandstone in the desert, that's one million fewer people free to practice medicine, or make smoothies, or grow mushrooms, or wash dishes, or make cartoons, or umbrellas, or fleshlights. The amount of labor available to produce all those things has gone down, and so their price will go up, and real consumption will go down. Resources have been redirected away from their "less productive" use (medicine, food, sex toys) and towards their "more productive" use (the vanity of one obscenely rich asshole). 100 billion dollars worth of consumer goods have been lost to produce a 100 billion dollar pyramid. They are both equal in monetary price, but the former has vastly more utility than the latter. If the billionaire had burned his money instead of building pyramids, everyone except him would be better off. This is something that I think is often missed in criticisms of "zero-sum" thinking. It's true that the economy is not zero-sum, but this does not imply that there are no tradeoffs.

It follows analytically that under the oppression hypothesis, since the enrichment of producers doesn't happen, any price signals I send do not reallocate resources to produce more in-demand goods on net

Let's examine a historical society whose inequality everyone agrees to be the result of an oppressive, extractive class: the antebellum south. In that society, the producers of goods were the slaves, but they didn't own their own labor, so the product of their labor went to their owners, minus the bit that was necessary for their subsistence. I don't think it was the case that price signals in the antebellum south allocated resources any differently than they did in any other society, and I don't fully understand why they would. If a slaveowner can make more money by redirecting the activity of his slaves away from cotton production and towards tobacco production, than he will do so, and spending a lot of money on tobacco will still send a price signal that resources should be allocated towards tobacco production. This is because although the slaveowner is not a "producer", he still has an incentive to utilize his property (in this case human beings) in the most profitable way. This is no different from owners of land, or companies, who have the same incentives to maximize the productivity of their property, despite merely owning productive goods, and not personally producing goods with their own labor. I don't think the price mechanism works differently between the two hypotheses.

I don't think these hypotheses of inequality contradict each other. Going back to the slave society example, inequality between the slaveowners and the slaves could be explained entirely due to differences in "productive capacity". The slaveowners owned lots of means of production (including the labor-power of the slaves themselves) and the slaves owned no means of production (not even their own labor-power). This is consistent with the hypothesis that wealth corresponds to productive capacity, because the slaveowners owned far more productive capacity, and it is also consistent with the hypothesis that wealth is the result of one class exerting force on another to maintain a claim on their property, because the slaveowners' rights over the products of their slaves' labor was enforced by violence. 

It's odd to see a post that uses the phrase "means of production" this often without mentioning Marx, who explicitly believed and repeatedly said that oppression was caused by, and equivalent to, distinctions in ownership of the means of production, rather than the two concepts being competing hypotheses. 

I'm very glad some people are finally working on this. This is one of the best suggested interventions I've seen in effective altruism to date. I correctly predicted that my biologist tumblr mutual nunuisancenewt would have a skeptical take though, and since he doesn’t have an account here, I will present his thoughts in his absence:

First, he’s very skeptical of the estimates for how many wild animals are affected by the new world screwworm. The only source given in your google doc is a paper which provides a statistic on the prevalence of myiasis in feral swine, which is really not enough information to conclude that myiasis is equally prevalent among all warm-blooded animals in South America. It's plausible that screwworms are significantly more likely to target large mammals like pigs and cows to lay their eggs in, and that screwworm myiasis is rare among the small rodents and bats who make up the majority of the wild mammals in South America, since those mammals spend a lot more time in burrows and caves, and are more vulnerable to predation (making them less likely to die from parasites, and also making it a worse idea to lay eggs in them because they don't live as long). KB Mathias cites a paper that cites another paper from 1937 that looked at 298 wild rabbits in North America prior to the eradication of the screwworm in that region, and found that 4% of them (12) were infected. This datapoint is evidence that screwworm myiasis is nearly as prevalent in smaller and more numerous wild animals than in larger and slower ones, but it is quite weak evidence. The EV estimates could easily be overstated by multiple orders of magnitude. Screwworms could conceivably affect closer to a million animals than a billion. 

Secondly, there would be negative consequences for wild-animal-welfare to growing the cattle industry in South America, which this plan hinges on accomplishing. I’m extremely certain that total cattle welfare would be improved by eradicating the screwworm, despite resulting in higher cattle populations, but the expansion of cattle farming in South America would hasten the deforestation of the Amazon, which has profoundly negative ecological and climatic consequences. The Amazon is a carbon sink, and destroying more of it will worsen the effects of climate change, which is bad for wild animals on a large scale. Brian Tomasik argues here that rainforest beef production is net positive because it reduces the population of wild insects, but nunuisancenewt strenuously disagrees, (primarily) on the grounds that stable ecosystems are largely populated by specialists, who are larger, fewer in number, and practice K-selection, whereas disrupted ecosystems are largely populated by generalists, who are the reverse, and can be expected to experience more suffering. I’m not sure if this rule applies to a hectare of rainforest specifically, because of how unusually dense the wildlife is in such areas, but I would assume that habitat loss in the rainforest probably means more ecological disruption generally, affecting species across a much larger area than that which is directly affected. 

He's not concerned at all about the direct ecological effects of eliminating the screwworm, or any other parasite, since they have small and predictable effects on the ecosystem, and they represent a nutrient transfer away from large slow-breeding animals towards small fast-breeding animals, which means more animal suffering, even discounting the very serious direct harm they cause.

Myself, I’m about 90% sure the EV of expanded cattle farming in South America from reduced mortality rates is negative, and also about 90% sure that that negative EV is smaller in scale than the positive EV of reduced suffering from animals that counterfactually would have suffered myiasis and instead die of something else. But the scale of the second number is really, really uncertain without a lot more information. There’s a not-super-implausible timeline where the number of animals saved from myiasis is lower than the number of animals burned to death by rainforest clearing. I’m very interested to read the results of the Wild Animal Initative’s currently ongoing investigation on the effects of the New World Screwworm on wild animal welfare, but before then the available information seems very incomplete. I can’t find any better source on the number of animals affected by screwworm myiasis than yours. Information on wild animal suffering is frustratingly incomplete generally, and the complexity of ecosystems makes predicting the sign on wild animal interventions very, very hard. This case is no exception. 

Organizations like the Shrimp Welfare Project or the Fish Welfare Initiative have high utilitarian returns, but are difficult to sell to people outside of EA, so it makes sense for dollars coming directly from the core of the community to go to these niche but highly effective cause areas. 

I think this post is unreasonably optimistic. A repeated theme in your post is that exponential growth in technology and the animal advocacy community could result in drastic pro-animal change, but it ignores the fact that there could also be exponential growth in the animal agriculture industry and anti-animal attitudes. There has been a rise in veganism, but there has also been a rise in meat-eating. AI could be used to convince people to become vegan, but it could also convince people that eating meat is okay and they shouldn't worry about it. AI could be used to come up with welfare-improving interventions in animal agriculture, but it could also be used to come up with profit-maximizing interventions, and I suspect the latter tools are far more likely to be applied. I don't see a persuasive reason why the animal welfare movement will gain an advantage over competing interest groups. 

Your claim here is that a tax on the lowest income tax bracket is "less costly" than a tax elsewhere which raises the same quantity of revenue. I don't understand how or why this is supposed to be the case. First of all, even with your small transfer example, I'm not sure if I consider the incentives to be that minimal. If a student would have stood to gain 800 dollars a month if they got a part-time job, and now they only stood to gain 700 dollars, I'd expect them to be less likely to get a job. But that's not even relevant. Compare the two situations which you claim differ in cost:

Both situations, as you describe them, involve a UBI of 100 USD per person per month. In both situations (we will assume), part of the cost is paid by a flat income tax (this isn't how I think a UBI should be paid for, but it's easier to understand this way). In America A, there is also a 100% marginal tax rate on the first 100$ of everyone's monthly paycheck (making it equivalent to a 100$ monthly GMI). In America B, there is only the flat tax, which has to be higher. In both Americas, the tax system must raise roughly 30 billion dollars a month. In America A, every wage earner pays 100$ a month in taxes from the first tax. Let's say there are 150 million wage earners. That's 15 billion dollars raised by that tax. The flat tax must be at a rate to raise another 15 billion dollars then. Let's say the total amount of labor income in the US is a trillion dollars a month. To raise 15 billion dollars from a flat income tax, you'd raise it by 1.5%. America 2 is the same but the flat tax needs to be raised twice as much, to 3%, because wage earners are no longer all paying 100$. I'm oversimplifying here but don't nitpick me. The point is that it is not at all obvious to me that a 100% tax on the 0-100$ income bracket is "less costly" than a 1.5% raising of the income tax across the board. In monetary terms it's obviously exactly as costly, just much more regressive, and If there were some reason that it WAS in some way less costly, then I would expect it to be even less costly to raise all 30 billion of the dollars from that same tax bracket, by putting a 200% tax on the first 100 dollars. And this would apply if you wanted to raise a tax for anything, not just a UBI. No one would propose such a taxation scheme to pay for a 15 billion dollar investment in green energy for example. It has no advantage. 

Let me explain my first point more clearly. 

Yes, guaranteed, means-tested incomes have some costs due to substitution effects that UBIs don't. People near the income threshold might work less or even quit the labor force to avoid losing benefits, potentially increasing program costs.

This is not what I mean. Observe Kevin’s graph describing the guaranteed minimum income in section 1.1. Notice how the first two income brackets have the same total income. An individual’s total income consists of their factor payments (in this case their wage) and their transfer payments. The GMI is structured so that the the wage plus the transfer will always equal at least a particular total income (I’ll continue to use 50,000 as an example). This means that if your wage is 10,000 dollars a year, your total income is 50,000 dollars, and if your wage is 50,000 dollars a year, your total income is still 50,000 dollars. This means in practice that you receive none of the first 50,000 dollars of your wage, making it indistinguishable from a 100% marginal tax rate on those dollars. This is very unintuitive to many people, so let me elaborate:

If you get a raise of one dollar, by how many cents does your total income increase? Your effective marginal tax rate equals 100 minus that number. I pay a tax of 25%, so if I get paid a dollar, I take home 75 cents. In the GMI world, if I got paid a dollar, I would take home 0 cents, because my transfer income would reduce by one dollar. 100 minus 0 is 100, so the effective marginal tax rate equals 100%. You say:

people near the income threshold might work less or even quit the labor force to avoid losing benefits

This is an inaccurate description of the problem. There is no income threshold. EVERY wage earner in this scenario is paying that 100% tax on their first 50,000 dollars. And there is no losing benefits. Going from a 40k yearly wage to a 50k yearly wage does not cause the earner to “lose” anything. It just doesn’t cause them to gain anything, so they have no incentive to take on the marginal disutility of the extra work for nothing in return. If they quit the labor force, it’s not to avoid losing benefits, it’s because it’s not worth it to work for no pay.

The same problem remains in the more realistic scenario where the GMI phases out slowly, so that rather than the first two bars in 1.1 being the same size, the second one is slightly larger than the first. This is how most means-tested benefits work in the real world. Let’s say that the GMI now works by paying a transfer of 25,000 for the unemployed, and for every dollar of wages earned, the transfer reduces by 25 cents. Now, if you get a raise of one dollar, your total income increases 75 cents, and pay an effective marginal tax rate of 25% (ignoring all other taxes) on all income between 0$ and 100,000, after which point there is no more transfer money to take away, and you start receiving one dollar for every dollar more you get paid. This is a better tax, but it’s still a tax. It’s not cheaper than a UBI. The cost is the same.

Compare phase-out world to a world with a UBI paid for by a flat tax. In phase-out world, there is a flat tax of 20% (arbitrary number), and the transfer payments phase out at a rate of 25 cents a dollar, meaning that the effective marginal tax rate is 45% between 0 and 100,000 dollars a year (assuming they phase out for each dollar of pre-tax income, not post-tax), and 20% for all income after that. In the second world there is a flat tax of 30%, and the effective marginal tax rate is 30% for all earners. The UBI is not “much more costly” than the GMI, because if you add up all the “labor dollars that someone doesn’t get”, you get the exact same number either way. The only difference is that in UBI-world, the rich receive less total income, and in phase-out world, the poor receive less total income. The same holds if the transfer payments are paid entirely through non-labor income taxes. 

Because the “phasing out” of the transfer payments would not appear on the IRS’s balance sheet as being a tax, it’s easy to mistake the GMI as somehow saving money compared to the UBI, even while accomplishing the exact same distribution. This is the mistake of thinking like a tax collector and not like a planning demon. In the same way that it’s easy to produce seemingly valid proofs that 1+1=1 by hiding a division by zero, it’s easy to make money appear out of nowhere by hiding a tax in the transfer system. The 50,000$ GMI world and the UBI+100% tax world have exactly the same distribution, and therefore exactly the same incentives and exactly the same amount of total wealth. There is no dollar that someone has in one world, but not the second. But the second world looks more “expensive” because the government chooses to give money and then take it away rather than withhold it, so it looks like there’s more distribution going on than there actually is. When we look at it from the demon’s perspective, we can see that there’s no difference between giving someone something and then taking it from them, and giving them nothing. This is what I was trying to convey with my example of the one million dollar UBI plus head tax. 

> However, empirically, these effects seem small.

The magnitude of the incentive effects is not relevant to the comparison between GMI and UBI. The key thing to understand is that the incentive effects of each are equivalent (if the GMI and UBI arrive at the same distribution). It’s the same effect both times: someone gains fewer cents when their wage is raised by a dollar. 

> real-world evidence suggests they do not seem nearly significant enough to make guaranteed income programs comparable in cost to full UBI systems.

They are comparable in cost because they are the same cost. Part of the cost is just hidden with a means-tested system by taxing the transfer payment instead of the wage, giving it the appearance of saving society money, even though everyone’s total income still adds up to the same number.

Two points:

Firstly, you say “In a labor economy, a guaranteed minimum income is a lot cheaper to implement than a universal basic income of the same amount”. I reject this. The cost of a transfer is the incentives it creates, not the total quantity of money moved around. If the US instituted a million dollar per person UBI, and funded it with a million dollar head tax, the “cost” on paper would be 300 trillion dollars, but the actual cost would be whatever the bureaucrats at the IRS are getting paid to write down all those numbers. No one’s incentives are changed, so society doesn’t actually pay any costs. It’s still producing the same amount of wealth as before, and distributing it in the same way as before.

Imagine two worlds: one in which there is a guaranteed minimum income exactly like the one you described, at a rate of 50,000 dollars. The second, where there is a UBI which gives everyone 50,000 dollars, and there is also a 100% marginal tax rate on all labor income less than 50,000 dollars (which doesn’t apply to the UBI). The first world seems reasonable, and the second crazy, but they are identical for all practical purposes. If you live in either world, and you have no job, and someone offers you to come work for their company to make boxes (or whatever) for 40,000 dollars a year, you’d have to be an idiot to accept! Because you’d be effectively taxed for 100% of that 40,000 dollars no matter what. You’d have exactly the same yearly income either way. A UBI funded by a tax on the poor is not cheaper than a UBI funded by a tax on the rich. 

You then say that a guaranteed minimum income would become more expensive as the labor force declined, but the cost of a UBI would remain the same. This is only the case because the effective tax has fewer payers, and is only paid by the tax bracket between 0$ and whatever the guaranteed minimum income is. World 2 would feel the same need to raise taxes elsewhere in the same situation, even though the quantity of money transferred would remain the same. However, if the UBI were paid for by a flat tax on labor income, the quantity available for transfer would stay the same assuming that the total quantity of labor income was the same, and if it were paid for by a progressive tax, the quantity available for transfer would increase! If your worry is that the total amount of labor income will decline, then that brings me to my second point.

My second point concerns financing a UBI. To start, I want to point out that automation increases “society”’s wealth, so if society suddenly has way more stuff in it, but everybody is poorer, something has gone wrong. It should get easier to pay people more if there’s more money to pay them with, not harder. 

To simplify, wealth is created by land, labor, and capital. People “earn” money by owning one or more of these things and selling it on the market. I agree that in a post-labor economy, you could not finance transfer payments by taxing the sale of labor, because the value of labor would go way down, and the value of land and capital would increase. But this is actually fortuitous, because land and capital, unlike labor, are alienable. Inequality between wage-earners is a result, at least partially, of nature. You can redistribute some of the income they earn through their skills, but you can’t redistribute their skills. Land and capital are very different. The same bond or share will pay the same return to whoever owns it. A post-labor society’s factor payments would all go to land and capital, which could be owned collectively. Land could be held in common and leased out to those willing to pay society ground rent for it (georgism), and governments could manage large (and/or numerous) social wealth funds on behalf of society, paying out dividends equally to every citizen (some kind of shareholder socialism? Property owning democracy? Really this should be called market socialism but that name is already taken by a different idea). Ideally, everybody would be a landlord who owned an equally valuable slice of land, and a capitalist who owned an equally valuable share of the world’s capital stock. In a world where we don’t need labor to make things, we also don’t need labor income to pay for things. 

I talk about all this here

Seems rather silly. Economic demand for data centers is very high. Even if you managed to garner some large political movement that supported stricter land use regulations for the construction of data centers (but were unable to garner a similarly sized political movement for pausing AI or something similar), AI companies would just pay more. NIMBYism can't price rich people out of homes, only poor people. When rents rise across the board people are pushed out from the bottom first. Trying to use zoning law and local development opposition to keep AI companies out of data centers is like trying to use them to make Elon Musk homeless. 

Also, just about any large building can be used as a data center, so it would be hard to restrict. Even then, if data centers in silicon valley became so scarce that google couldn't even afford them, they could just go elsewhere to build them. Location is very important for housing. I don't think that's as much the case with data centers.

Plus, we might think that privately allocated research effort is better matched to social value than whatever university scientists can get through the NIH committee.

Wouldn't we assume the opposite? As a rule, public dollars have more utility than private dollars. Most private patents probably have a negligible effect on social value, and some probably have a negative effect, even if I'm certain that more private patents tends to be a good thing. That 2.5 million dollars per patent via tax cuts could be paying for a patent on new modes of advertising, blenders, corporate interview software, and items on the taco bell menu, whereas the 3.1 million dollars per patent via the NIH is all paying for patents on vaccines, antidepressants, non-invasive surgery methods, and painkillers. Not all innovation is created equal.