Bentham’s Bulldog recently argued that AI won’t definitely make factory farms obsolete. I agree, but I’d go further and argue that by default AI won’t make factory farms obsolete. However, I think it’s possible (though not guaranteed) that AI could make factory farms a lot more humane. 

He throws out an 80% chance of cultivated meat being developed, and a 70% chance of it displacing factory farms contingent on it being developed. In particular, I think 70% is too optimistic. 

An 80% chance of cultivated meat being developed contingent on TAI seems reasonable to me, insofar as it's possible to forecast such things right now. We currently have no feasible technical roadmap for cultivated meat to be able to create all of the myriad kinds of animal products that people consume today. However, one could plausibly think that TAI means that any technology that’s consistent with the laws of physics will become possible. Cultivated meat will be possible in these worlds, and some additional superset of these worlds.

However, I’d put a much lower chance of it being adopted in a significant way. To make a very vibesy forecast, I think there's maybe a 40% chance that cultivated meat (or some other technical paradigm that doesn’t involve living beings, like bodyoids) is a majority of meat production in the future, contingent on it being developed. Multiplying these two likelihoods together gives a 32% chance that AI will make factory farms obsolete.

Here, I’m conditioning on humans as we know them continuing to exist. I take no position on how likely this is to happen, but it seems to me like this is the outcome that EAs are pushing for when trying to reduce existential risk from AI. 

My disagreement with Bentham’s Bulldog stems from the idea that the worlds in which cultivated meat are possible are also ones in which animal protein is significantly cheaper. Therefore, people will choose whichever they want more, and we have no evidence people will want cultivated meat more.

The worlds in which cultivated meat are possible are also ones in which animal protein is significantly cheaper 

The order of events for AI takeoff is roughly as follows: 1) recursive self improvement, 2) country of geniuses in a datacenter / automate all white-collar work, 3) industrial explosion 4) material abundance / post-scarcity 5) singularity. 

Given that cultivated meat is an industrial technology in the world of atoms, it will most likely become possible in steps 3 and 4. 

The reasons cultivated meat will become possible are that the entire manufacturing process will be run by robots, equipment, material and facilities upstream in the supply chain will become cheap to produce, and AI will give us a deeper understanding and ability to control biology. 

These are all things that will affect the price of animal protein as well. During the industrial explosion, traditional animal husbandry will also go through an AI transformation that will lower its costs. Husbandry will become more automated and the cost of facilities and equipment will go down. Just as AI helps us understand the cell better for cell culture, it will also help animal breeders understand livestock genetics and health better.

In the limit of technological capability, I expect the price of animal-based meat and cultivated meat to be fairly similar to each other, but both extremely cheap relative to societal wealth. The cost of both will primarily be driven by the cost of the primary feedstock (like corn and soy). Fixed costs of equipment and facilities will fall, and be amortized over huge volumes. 

It’s hard to forecast which production paradigm will be cheaper. I suspect that there will be variation based on product category and meat type. For example, the theoretical argument for why cultivated meat might someday be cheaper than animal meat is because animals waste energy doing things that we don’t care about, like growing bones or thinking. However, this inefficiency can be measured through the feed conversion ratio, which is lower for smaller animals. For example, shrimp have a feed conversion rate of under 2, which means their total metabolic waste is fairly small. Cultivated shrimp may therefore never be cost competitive with conventional shrimp. But it may be competitive for something like beef, especially more expensive cuts. 

People will choose whichever they want more, and we have no evidence people will want cultivated meat.

That said, relative differences in cost might not matter that much, since both will be extremely cheap relative to consumer purchasing power. If I were a millionaire and could eat animal-based meat for a year for $10, or cultivated meat for a year for $10.25, the cost would not be decision-relevant to me at all. I would choose whichever one I want more.

I can think of two reasons why people might want to eat cultivated meat over animal-based meat:

1. They’re persuaded by moral arguments
2. Cultivated meat has some additional value proposition over animal-based meat. For example, it’s healthier or tastes better. (In a previous post, I talked about AI-pilled cultivated meat development focusing on increasing the consumer value proposition, and this is what I meant).

On the moral arguments - It’s nice to think about AI making the world more ethical, and there are plausible mechanisms by which this would happen, e.g. AI having superhuman persuasion and animal-friendly ethics. But I don’t think we have much evidence that people actually have fundamental values that would say animal farming is wrong. As Bentham’s Bulldog points out, only 13% of current consumers would prefer cultivated meat if it existed today, and only a very narrow subset of people think eating animals is fundamentally wrong. We shouldn’t assume that once cultivated meat is cost-competitive that people will switch because it’s the “right thing to do.”

In fact, we seem to have substantial evidence that people won’t want to eat cultivated meat. Current humans have a strong naturalistic preference when it comes to food, particularly animal products (as demonstrated by the rise of the MAHA movement). It seems equally plausible that AI could empower and expand these kinds of preferences, making it more difficult for a weird unnatural technocratic solution like cultivated meat to succeed.

Objections

Matt Reardon offers one defense of the view that factory farming will be made obsolete by AI: 

Addressing each claim in turn:

• “Factory farming might spread across the stars” makes narrower assumptions than the view that it won’t.
  • I’m not sure why this would be true. In order to claim that people will stop eating meat by default, you have to assume that future humans will be different from current humans.
• Things like intelligence explosions and the ability to settle space imply the ability to generate way, way higher (selfish) returns to the use of matter and energy than terraforming worlds for meat-humans we’ll never meet.
  • Returns according to what value system? It seems that if we succeed at aligning AI to human values, then the value system we’re talking about is fairly likely to be one in which humans want to eat meat. In fact, you could make a similar argument about today’s economy – why waste so much energy growing animals when we could all just eat soybeans directly?
• The factory farming path requires a pathological obsession with maintaining traditions in the face of a vastly wider and more efficient options for how to live.
  • People demonstrate many obsessions with tradition that seem to defy “efficiency” and “reason.” Matt may think these are pathological, but that’s relative to his own value system that I, but relatively few others, share. What Matt calls efficient many would call “weird” and hence undesirable. I’m not sure we have a reason to think AI will mitigate this rather than amplify it.

I don’t mean to pick on Matt, he’s just the only person I’ve seen to try to defend the view publicly. I would love for others who hold this view to articulate their reasoning so we can get to the bottom of this important question!

AI could make factory farming significantly more humane

To not be totally pessimistic, I think it’s also possible that AI could make factory farming significantly more humane, perhaps even to the level where animals have net positive lives. 

The fundamental reason that animals are treated poorly on factory farms is because to treat them well costs money, and someone has to pay for it. As wealth increases, the relative value of this cost will decrease. The industrial explosion and the ensuing material abundance will give us flexibility to afford many things which are now considered luxuries, and animal welfare could be one of them. 

This is a silver lining in this generally pessimistic forecast, although it’s certainly not inevitable. There are lots of things we could already do for animal welfare that are insignificant costs that we don’t do, and Bentham’s Bulldog’s vision of torture chambers spreading throughout the galaxy is certainly a realistic possibility that’s sobering to consider.

But as the cost of animal welfare goes towards epsilon, it could in theory be possible to structure society in a way that this externality is always mitigated (e.g. welfare regulations, or robust AI monitoring and certification schemes). Engel's law is an empirical principle stating that as wealth increases, food expenditures as a share of total expenditures go down, but total food expenditures go up. Given that we do currently have lots of evidence that people value animal welfare to some extent, this presents a theory of change for animals that doesn’t rely on future humans having different preferences or better morals from current humans. 

A major uncertainty I have is how economic incentives will change in a world of massive material abundance. Will we still see “race to the bottom” dynamics for things that generate negative externalities, even if the cost of eliminating those externalities falls to epsilon? Or is there any mechanism by which they’d be solved by default once the costs get sufficiently low? I’d love someone to try to investigate this question more deeply.