TAI-driven clean meat won’t solve the problem but changes movement strategy anyway

[Bella]

This is a contribution to AGI & Animals Debate Week (March 23–29, 2026). Co-written with Claude and ChatGPT.

Summary: It seems reasonable to think that there’s a >20% chance^[1] that TAI will bring us a ‘technological roadmap’^[2] to clean meat within the next 15 years. A simple model: such a roadmap won’t solve the problem, but it will make ending factory farming much easier, i.e. the cost-effectiveness of factory farming interventions will increase – perhaps by a factor of 2-10x. The movement should therefore be investing instead of deploying, focusing on capacity-building, research, and literal investment of money. It may also be cost-effective to act now on some ways of removing barriers to the eventual adoption of clean meat (e.g. ensuring clean meat can be halal-certified at scale).

I will argue:

1. AI-driven R&D will likely deliver a ‘technological roadmap’ ^[1] to clean meat within the next 15 years or so.

2. However, this won’t end factory farming by itself – and some barriers can be worked on now.

3. More broadly, we should be anticipating a stark increase in cost-effectiveness, so we should move into an ‘investment mode’.

At the end, I’ll provide some specific ideas about useful work (e.g. working with cultural/​religious groups, fighting legal bans, and some community-building organisations that I am excited about).

1. AI-driven R&D will likely deliver a ‘technological roadmap’ to clean meat

It’s hard to predict exactly how TAI will affect the world, but one strong heuristic is that it will speed up R&D.^[3]

That doesn’t mean that all possible technologies will necessarily be developed. Instead, we’ll reach technologies that see rapid returns from intelligence.

Clean meat R&D seems like one such technology:

• Its key bottlenecks are scientific and engineering ones — especially media formulation, bioprocess/​bioreactor design, and scaffolding — where more and faster researcher-equivalents directly help.^[4]

• Its experimental feedback loops are relatively fast — relevant cells often divide on timescales of roughly 20–70 hours, so automated research could run many design-build-test cycles quickly.^[5]

• Many of its open problems fall into several distinct research tracks, making parallel work unusually plausible.^[6]

• Its main regulatory barriers are often to market access rather than to the underlying research itself.^[7]

The physical costs of production (energy, sterility, bioreactor construction) will remain, even after an intelligence explosion. However, these are likely to be resolved by an industrial explosion bringing cheap energy and automated manufacturing.

Based on this, it seems like the chance that TAI delivers a sufficient technological roadmap for price-, taste-, convenience-competitive clean meat within the next 15 years or so is greater than 20%.^[8]

Achieving price-, taste-, convenience-competitive clean meat also requires competitiveness: i.e. that traditional agriculture can’t be made even cheaper by the intelligence and industrial explosions. TAI will likely accelerate improvements to conventional animal agriculture – this is a key reason for my uncertainty.^[9]

2. However, this won’t end factory farming by itself – and some barriers can be worked on now

Two things that would not follow automatically from such a roadmap:

• Infrastructure. Building production capacity to replace a meaningful fraction of animal farming would be one of the biggest infrastructure projects in history — though TAI would likely help here too, particularly via the industrial explosion.

• Adoption. A roadmap tells you how to produce the meat; it doesn’t tell you whether people will eat it.

The key question therefore lies in adoption – others have written extensively about the adoption barriers.

I think there are things we can do now to reduce those barriers.

Religious certification. If clean meat isn’t halal, that’s potentially 2 billion people for whom the technology doesn’t solve the problem. There’s been real progress here: in May 2025, the International Islamic Fiqh Academy, one of the world’s leading authorities on Islamic law, operating under the Organisation of Islamic Cooperation, approved cultivated meat consumption at its 26th conference in Doha, subject to specific conditions. But there’s no universal consensus across Islamic schools of thought, and the actual production processes of existing cultivated meat companies don’t yet meet the stated criteria.^[10]

Legal bans. Seven US states have enacted bans or temporary prohibitions on the sale or manufacture of cultivated meat with several more considering similar legislation. Reversing or preventing the extension of these bans – and preventing new bans – could substantially increase the proportion of the population that is able to adopt clean meat.

I’d guess there is other work to do here – I haven’t thought about this much. I’d guess this kind of work is among the most important for the animal movement right now.

Importantly, this work looks quite different from most current alt protein spending in the EA community. Most current alt protein work is focused on speeding up technological progress on clean meat, which I expect to be negligible compared to AI-driven progress.

3. The movement should move into an ‘investment mode’

There’s a very long and wonderful EA literature about whether to give now or save and give later.

Lots of this comes down to whether you can really beat a 4-7% real rate of return from investment, with all sorts of fascinating and nuanced arguments in both directions.

But if TAI delivers price-, taste-, convenience-competitive clean meat, a huge barrier to ending factory farming will have dropped. So (depending on how much more money it would take to end factory farming at that point) the cost-effectiveness of the marginal animal welfare intervention plausibly increases by a factor of something like 2–5x in 5-15 years.^[11] That’s an annualised return of something more like 4-40% (with >20% probability) – and if we don’t get clean meat, you still get some extra money to deploy (likely even more if we see AI-driven growth).

Think of it like expecting house prices to drop: if there’s a reasonable chance prices are about to fall substantially, you probably shouldn’t buy a house.^[12]

This has a few practical implications.

The most literal version of ‘investment’ is just saving money. Funders could invest financial resources at market returns and deploy them when the strategic landscape shifts.

But that isn’t the only way to invest. When clean meat arrives (if it does), the movement will need skilled campaigners, policy expertise, organisational infrastructure, relationships with policymakers, experienced leadership, and research to understand this whole TAI situation. All of this takes years to build.

I’d guess there should still be welfare-focused spending (e.g. on corporate campaigns), but the right level, and the specific short-term actions to take probably look different if you’re optimising for capacity-building rather than near-term cost-effectiveness.

4. Some specific ideas

I haven’t looked deeply into most of these, and I’d like to see researchers and funders thinking about the clean meat frame and working out what follows from it more carefully than I have. But here are some starting points.

Adoption. As mentioned above, working with religious authorities to ensure clean meat production processes are halal; fighting cultivated meat bans; and mapping adoption barriers we haven’t identified yet. And, as mentioned above, this probably doesn’t look like conventional alt protein work, which will cause only negligible speedup compared to AI-assisted R&D.

Movement building. It’s unclear to me whether we should be building a broad movement now. My instinct is that broadness becomes much easier once clean meat exists, which might favour the targeted approach for now. Specific orgs:

• Animal Advocacy Careers is specifically focused on building a small, talented base – which is the kind of thing the movement will need once we have clean meat.

• Farmkind is more broad, but focuses specifically on a constituency of people who are sympathetic to farmed animals, who wouldn’t go vegan now, but who would plausibly switch to clean meat when it arrives – that seems like an important group to capture.

Financial investment. I think it’s likely that new funding from new donors will enter the animal welfare space in the near future, which would reduce the neglectedness of the area and therefore the marginal returns to financial investment. This points overall toward spending on capacity building — which is harder for new entrants to replicate quickly – rather than pure financial saving.

Research. I expect there are better ideas than mine, and I’d like to see people working on factory farming think seriously about what the clean meat frame means for their work. Also, while I think clean meat is likely to be one of the key ways TAI affects animals (because it follows from one of the strongest TAI heuristics – that it accelerates R&D), there will be other important effects we haven’t yet identified. I’d like to see much more research on this question.^[13]

1. ^

   My actual estimate is something like 40%, but the argument in this article likely holds for values between 20% and 100%.

2. ^

   By ‘technological roadmap’ I mean a document detailing how to produce price, taste and convenience-competitive clean meat without the need for further R&D. This would not include: the infrastructure to produce the clean meat, any cultural or societal requirements for adoption, etc.

3. ^

   This is in contrast to Ben West and Lizka’s shallow review, which argues that the post-transformation world is hard to predict. I agree that we should be sceptical of most speculative TAI-for-animals interventions — but I think “TAI will accelerate R&D” is a robust enough prediction to build on. See Amodei, Machines of Loving Grace, October 2024; Forethought, Preparing for the Intelligence Explosion, March 2025; Forethought, Will AI R&D Automation Cause a Software Intelligence Explosion?, March 2025; Forethought, The Industrial Explosion, May 2025.

4. ^

   The basic picture in the public literature is that the main remaining bottlenecks are technical ones. GFI writes: “Scaling will also require solving an array of complex challenges that will influence the cost of production. These challenges span five key areas: cell lines, cell culture media, bioprocess design, scaffolding, and end product design and characterization.” Their more detailed pages are in the same vein: “The bioprocess design holds the key to unlocking large-scale production of cultivated meat,” while competitive media will require “significant reductions in costs, innovations for serum removal, and optimization across a diverse set of species and cell types.” On media formulation specifically, Todhunter et al. write: “Designing culture media is fundamentally a high-dimensional search problem.” Taken together, this is some evidence that the main bottlenecks are scientific and engineering ones — especially media, bioprocessing, and related design problems — rather than a single regulatory permission or one missing conceptual insight. See GFI, “The science of cultivated meat”; GFI, “Cultivated meat bioprocess design”; GFI, “Cultivated meat cell culture media”; and Michael E. Todhunter et al., “Artificial intelligence and machine learning applications for cultured meat,” Frontiers in Artificial Intelligence (2024).

5. ^

   Hauser et al. treat proliferation speed as a central practical lever for cultivated-meat production. They write: “Because cell proliferation is an inherently exponential process, even modest improvements to cell proliferation rates can yield significant cost reductions for the process of CM production.” They then give concrete doubling times: “For reference, typical doubling times of bESCs and bMSCs are 24–70 h, chicken fibroblasts can be optimized to divide every 20 h and the fastest doubling times for mammalian cells are 5–10 h …” So the relevant experimental cadence here is often measured in days rather than months. See Mor Hauser et al., “Challenges and opportunities in cell expansion for cultivated meat,” Frontiers in Nutrition (2024).

6. ^

   The strongest direct support here is that cultivated-meat R&D is regularly broken into several distinct technical tracks. GFI writes that the field’s challenges “span five key areas: cell lines, cell culture media, bioprocess design, scaffolding, and end product design and characterization.” Todhunter et al. organise their review around “four major areas of cultured meat research and development: establishing cell lines, cell culture media design, microscopy and image analysis, and bioprocessing and food processing optimization.” This does not show that the subproblems are fully independent. But it does suggest that cultivated-meat R&D is not a single monolithic bottleneck, which makes parallel work more plausible than in some other technologies. See GFI, “The science of cultivated meat” and Michael E. Todhunter et al., “Artificial intelligence and machine learning applications for cultured meat,” Frontiers in Artificial Intelligence (2024).

7. ^

   Across several important jurisdictions, the formal hurdle is mainly to selling the product rather than to doing the underlying research. In Great Britain, the Food Standards Agency writes: “Cell-cultivated products must be authorised before they can be placed on the market in GB.” But the same page also says: “Tasting trials of unauthorised novel foods may be permitted if the intention behind them is to conduct research to develop the novel food.” Singapore uses similar language: the Singapore Food Agency says that cultivated meat “must be assessed for safety before they can be allowed to be used in food for sale,” and its 2025 novel-food guidance says: “No approval from SFA is required for research on novel food and novel food ingredients. Novel food and novel food ingredients that are under research and have not received pre-market approval from SFA should not be made available for sale.” In the United States, the federal language is also framed around market entry: the FDA says its pre-market consultation process evaluates safety “before it enters the market,” and USDA-FSIS says establishments that intend to harvest or process cell-cultured meat and poultry “must apply for and obtain a USDA grant of inspection.” This is not universal: Florida now makes it “unlawful for any person to manufacture for sale, sell, hold or offer for sale, or distribute cultivated meat in this state.” See Food Standards Agency, “Cell-cultivated products”; Singapore Food Agency, “Safety of Alternative Protein”; Singapore Food Agency, Requirements for the Safety Assessment of Novel Foods and Novel Food Ingredients (PDF); FDA, “Human Food Made with Cultured Animal Cells”; USDA-FSIS, “Human Food Made with Cultured Animal Cells”; and Florida Statutes § 500.452.

8. ^

   Will the inference cost required to produce this roadmap be paid? My guess is that the existing alt protein investment market (plus future profitability from plausibly capturing a meaningful share of the meat market) provides sufficient incentive to pay that cost, so we should expect the market to provide here.

9. ^

   I’d guess that conventional agriculture has fewer efficiency gains to be found, and is more bottlenecked by the rate of generational turnover (~1 year for chickens, ~5 years for cattle). With further technological progress this bottleneck (and others) could be overcome with tools like CRISPR-like gene editing, which is a key reason for my uncertainty. That said, fundamentally, growing nociceptors seems like a removable energy cost, which suggests that clean meat will win out (although maybe not within 15 years).

10. ^

    Conditions for halal cultivated meat from Saudi scholars include: the cell line must come from an animal permissible to eat under Islamic law; that animal must have been slaughtered according to Islamic law; the growth medium must contain no prohibited substances (e.g. blood, alcohol, pork-derived materials); and the product must be confirmed safe by a relevant regulatory agency. These conditions mean, for instance, that cultivated pork is still not halal, and that meat produced using fetal bovine serum (FBS) — which is derived from the blood of unborn calves — would not qualify. GOOD Meat’s chicken, which has FDA and Singapore approval, does not currently meet these criteria. IIFA’s ruling likely included similar criteria. See Green Queen, June 2025; GOOD Meat, September 2023.

11. ^

    How much does PTC-competitive clean meat increase cost-effectiveness? I’m genuinely uncertain. A rough Fermi: there are something like 2 trillion SAD-years of suffering in factory farms annually (per AIM). Over a decade of transition, that’s ~20 trillion SADs averted. A global campaign to end factory farming might cost on the order of $1 trillion in advocacy alone — though this estimate is very rough (for comparison, the most expensive single campaign I could find was California’s 2022 sports-betting ballot, costing something like $500 million, and that was two propositions in one state). At $1 trillion that gives ~20 SADs/​$, compared to current top estimates of ~30 SADs/​$ (roughly AIM’s ideal). If the real advocacy cost is $100 billion, you get ~200 SADs/​$, which is closer to a 7–10x improvement. At $10 trillion (i.e. if the political fight is far harder than I expect), it might be less than 1x and the whole argument falls apart. I’d like to see more careful analysis of this parameter.

12. ^

    This argument has been circulating in the EA animal welfare community for some time. See “Animal advocates should respond to transformative AI maybe arriving soon” (EA Forum, August 2025), which notes: “maybe we can predict that animal advocacy interventions will become more cost-effective in the future due to increased wealth or more affordable higher alternative protein products, and therefore focus for now on capacity building and recruitment over immediate impact.” See also “Strategic Considerations from AI and Alternative Proteins” (EA Forum, October 2025), which arrives at similar conclusions about front-loading institutional investment.

13. ^

    Another approach people have considered is shaping AI values to be more animal-friendly (e.g. persuading AI companies to build concern for animal welfare into their models). I’m sceptical this is a major lever. If AI systems are aligned, and the market is competitive, then an AI that refuses to help someone cook chicken simply loses that user to a competitor. For a broader discussion of AI-values-for-animals and related approaches (e.g. precision livestock farming, lock-in risks), see Ben West and Lizka’s shallow review, which I broadly agree with.

[MatthewDahlhausen]

Can you define TAI in the summary of your post?