AI and animal welfare: am I missing something?

Hi Siobhan! Great question and I'm genuinely glad you raised it. I'm the Exec Director of Sentient Futures which is trying to build out the AIxAnimals field and your confusion probably represents a failure on our part to properly communicate our ideas and the field's progress. I can mostly speak to why we as an org have decided not to work directly on AI applications or cultivated meat. 

A couple (non exhaustive) points:

1. There are two different concept you are talking about here: Applied AI vs Frontier AI
   1. The applied side: (i.e. accelerating cultivated, replacing animal testing, etc) takes a lot of industry expertise and has been tried and is still being tried outside of the EA space. There are a lot of capital, status quo, cultural, regulatory, etc barriers on this front that would take a lot of work and funding to overcome. With the constraints of talent and funding in the AIxAnimals space, I think this would be near impossible for this group of people to pivot there, especially since with the funding constraints of alt proteins, even far more experienced food scientists are finding themselves unemployed. We've tried to support these efforts by platforming them in our conferences and newsletters, but it would be spreading ourselves quite thin and we do not have the expertise in any of these areas.
      1. There are also other applied AI interventions that we've platformed such as inter-species communication, precision livestock farming, and AI to replace animal testing. We've known for a while that working on these interventions directly is not our comparative advantage.
   2. The frontier side: (i.e. getting AI models, developers, and regulations to consider animals more. I think this is more tractable and EAs are better positioned to do this at our current stage for a couple reasons:
      1. I don't think people are indifferent: AI folks at labs care more about animals than your average person. Many millions of dollars have already been donated to animal welfare efforts from AI lab employees (not public, but verified) with potentially hundreds of millions to come. Some are internal champions who really care, but just need a robust benchmark or red teaming case to pitch to their colleagues. Many are open to incorporating consideration for animals as long as it doesn't run up against some other tradeoff like capabilities, human safety, or substantial profit. The work of the AIxAnimals field is to find these win-win situations where possible and package them up in a way that the frontier labs are happy to make a change.
      2. Specifically the EU AI Act - a group of animal folks got together and participated in the working group consultations for the code of practice and were successful in getting non-human welfare added as a systemic risk consideration in the final language. It was a short window of time and if we had waited, then we would have missed this opportunity to set a policy precedent and build upon other things like the TFEU Article 13. This was just done with a small team of mostly volunteers or people taking time out of their normal jobs to draft arguments and attend meetings. Making sure that it is enforced is going to take a lot more work that we don't currently have capacity for. But the language precedent is there to build on and use as a foundation. There was essentially no opposition push back (unlike with cultivated or any other animal protection laws) because the animal industries are not paying attention to frontier AI regulation. So this is quite tractable, but I suspect it won't be for long.
2. Cultivated meat has a lot of enemies and things working against it - I do think fighting cultivated meat bans is one of the more important things that we could be doing for animals (see the lessons from this wargame we played) in the long term. But it seems like you can squash one and then another ban comes up, and it's just this perpetual game of whack-a-mole. There are many people who are skilled at policy who I respect and donate to that are taking care of this. I do think they could use more help, but I don't think it is as neglected or as underfunded. I think they also have more mainstream appeal and accessibility. For example, many governments and schools do give funding for alt protein R&D. Important, tractable, not as neglected. The AIxAnimals path goes to one that's less traveled and where there are not a bunch of institutions already trying to protect the status quo. I don't think it would be wise for everyone to go all in on cultivated because there is no guarantee that even if it was technically successful, it would get cultural adoption. We need to diversify our interventions.
3. Frontier and applied AI holds the promise of solving wild animal suffering. Most animal suffering is not man-made, and the current pace of research into reducing wild animal suffering is just a drop in the bucket compared to the scale of the problem. More advanced AI does hold the promise of being able to process large amounts of information and make complicated trade-off decisions and immediate actions. Right now, AI is awful at reasoning about wild animal welfare and defaults to ideals of conservation, biodiversity, and nonintervention. One concrete thing we've been working on is increasing the training data of animal perspectives and how AI intervention could be positive for animals. See this https://hyperstition.sentientfutures.ai/
4. If you want a concrete BOTEC about the value of aligning AI to animal welfare, check out this one by @MichaelDickens : https://mdickens.me/2026/03/24/alignment-to-animals_BOTEC/
5. For a broad overview of what AI means for animal advocacy (both applied and frontier), I'd highly recommend watching this interview with Lewis Bollard. 

This is what I have for now. I am not sure if that is a satisfying answer for you because it is just such a nascent field and we are trying to figure it out as we go, but I really appreciate you raising it. 

"Are there such opportunities? So far, I can't break this down to anything tangible. 'If we don't do anything, the systems will become entrenched and determine animal outcomes for decades to come' - what systems? What outcomes? Who, where? Can someone give me a few clear examples of tractable situations? 

... Agricultural AI platforms from John Deere and Bayer are being deployed. But how is any of this tractable - what are you hoping orgs/grantees/EA people can do about those things?" 

... If that's so, then can someone tell me, in plain English, what that looks like? I.e '[lab] is currently planning [this development]. If we do [this action], we can change it to [this outcome], which will mean [x number] of animals experience [less suffering, presumably].' 

I just want to express that these are really sound questions and I hope you keep asking them. I believe they are, as of yet, unanswered. I'd love to see AIxAnimals field/thought leaders work out their theories of change more clearly and concretely. 

I think you correctly point out that there are gaps, and some of the framings I've seen seem to suggest that poultry farmers will just sit idly by as their animal-friendly AI COO's will make commercially suboptimal farm-management and procurement decisions. 

But, this is still a nacent field and I think questions like these will help it along!

Hi Siobhan. Thanks for the post. I broadly agree with the sentiment you express in it.

Sounds like this is in reaction to yesterday's launch of the Falcon Fund. I'm very excited this fund is happening, and I am personally donating to.

I appreciate you thinking out loud! As Constance said, I think this points to a lack of communication around the concrete ideas. I think Sentient Futures has done a phenomenal job in bringing attention to this whole area, and to me the whole point of the Falcon Fund is to start turning all the thinking into concrete projects.

My background is in ML in the alternative protein space. So I am also very excited about the prospect of AI helping the development of cultivated meat. However, at this stage, I think the AI-cultivated space suffers even moreso from the exact problem you're describing where there is no well-defined problem. I wrote about this here if you'd like to discuss it more. I don't think AI can model and run simulated experiments in this space. One day it might be able to, but that is much more likely going to be downstream of other general advances in the AI-sciences, and won't be developed in the cultivated industry itself. We don't have the money, talent, or data throughput to make that happen. I actually think what we should be aiming for in the short-medium term is something that looks like speeding up wet lab trial and error, I think there are a lot of gains to be had from designing experiments better and deploying something that looks like self-driving labs as soon as possible. If you have specific ideas for ideas that we can try testing now, I would love to talk.

In general I agree that we need to put more thought into the AI-cultivated plan.

On AIxAnimals I think there are some pretty clear things we need to do, which are also described on the Falcon Fund page.

Having an observatory/watchdog organization is really critical. We currently have no view on how AI is being used in industries that will impact animals. That will include when they run factory farms, do scientific research in areas like precision livestock farming, and help with ecological management. We need to see what's going on in order to make decisions.

Having benchmarks is similarly important to actually have visibility into how these systems behave that are going to be very important. Then yes, hill-climbing on these numbers would be great.

I am sympathetic to your skepticism, especially that things that limit profitability will be hard. But I think we can afford to try here. Human values are not the same things as what the markets provide. Humans stated preferences are to like animals. Every ballot measure presented in any state in the US, whether led by Republicans or Democrats, has passed. It is one of the few bipartisan issues. I think a lot of the issues of animal welfare are due to people not being informed. In which case AI can be a powerful tool in aligning people's behavior with their values, which is something we should encourage.

It is going to be very hard to be concrete about what effect we see in the real world until AI systems become industrially relevant. Until then the primary job is to observe as much as we can and make sure we're moving things in the right direction. We can do this by having a benchmark, and seeing if the change in model spec language improves the benchmark.

AI systems are being built to do exactly what they were designed to do, which is to faithfully execute human preferences. And those are, in aggregate, to eat cheap meat, conduct research on living organisms when it's convenient, and prioritise cost and efficiency in agricultural supply chains. AI is reflecting the values of the humans. I don't think you can sneak those values in, unless there are specific opportunities to tweak things here and there before they get cemented.

This is true in a way. A deeper problem is that we don't know what values AI is reflecting. If you talk to an LLM, it will express some values, but it gives inconsistent answers depending on what questions you ask. We have no way of knowing whether its expressed values reflect its "true" values, if it has any. And we don't know how things will change as AI becomes increasingly powerful.

Mo Putera @ 2026-04-26T12:12 (+2)

What do you think of efforts like Saffron Huang et al 2025? It's from a year ago as of this week so I'd guess Anthropic to have developed this line of work further since and integrated it into other workstreams and such.

AI assistants can impart value judgments that shape people's decisions and worldviews, yet little is known empirically about what values these systems rely on in practice. To address this, we develop a bottom-up, privacy-preserving method to extract the values (normative considerations stated or demonstrated in model responses) that Claude 3 and 3.5 models exhibit in hundreds of thousands of real-world interactions. We empirically discover and taxonomize 3,307 AI values and study how they vary by context. We find that Claude expresses many practical and epistemic values, and typically supports prosocial human values while resisting values like "moral nihilism". While some values appear consistently across contexts (e.g. "transparency"), many are more specialized and context-dependent, reflecting the diversity of human interlocutors and their varied contexts. For example, "harm prevention" emerges when Claude resists users, "historical accuracy" when responding to queries about controversial events, "healthy boundaries" when asked for relationship advice, and "human agency" in technology ethics discussions. By providing the first large-scale empirical mapping of AI values in deployment, our work creates a foundation for more grounded evaluation and design of values in AI systems.

MichaelDickens @ 2026-04-26T19:58 (+2)

1. An LLM's expressed values are not the same thing as its actual values, insofar as it has any.
2. This paper doesn't really tell us anything about how ASI values will work. This paper is relevant to the immediate problem of making Claude commercially useful and non-destructive, but it's not relevant to ASI.

Mo Putera @ 2026-04-27T03:32 (+2)

What kind of empirical evidence would update you positively?

MichaelDickens @ 2026-04-27T03:50 (+4)

I'm not sure if there's any. My concerns are more theoretical than empirical, so it would take theoretical work to significantly change my mind.

Empirical work can provide a small amount of information, e.g. the fact that Claude expresses concern for ethics is a slight positive update relative to the world where Claude doesn't care about ethics, and I would feel slightly better about a Claude-based ASI than a ChatGPT-based ASI. But only slightly, because I don't think empirically observable behavior is that relevant to determining whether an AI is aligned. At least not using any empirical methods that we've devised so far.

For more on this, see e.g. A central AI alignment problem: capabilities generalization, and the sharp left turn, especially the part starting from 'How is the "capabilities generalize further than alignment" problem upstream of these problems?'

(ETA: On how various plans miss the hard bits of the alignment challenge is also kind of about this...I was looking around for some writings on why current empirical work isn't that relevant but it's hard to find anything that directly makes the argument)

From my POV we are deeply confused about what it would even mean to align ASI. If I could even describe what sort of theoretical work would be good evidence of progress on alignment, we'd be in a better place than we are currently.

A significant reason for my high P(doom) is that most safety researchers at AI companies are ignoring theoretical issues and pretending that alignment is purely an engineering problem. I don't think they are institutionally capable of solving alignment.

Mo Putera @ 2026-04-27T05:40 (+3)

By empirical evidence I meant anything empirical at all, including things like emergent misalignment and what might come out of Jacob Steinhardt's interpretability program and what Ryan Greenblatt says here and whatever the right value-analogue of Anthropic's functional emotions paper is (below) and so on, not just observable behavior. Maybe I'm conflating things or overloading "empirical", in which case my apologies.

Regarding the sharp left turn, Byrnes' opinionated review is the best argument for worrying about this that I'm aware of, but he isn't talking about today's LLMs and their descendants, which rules out your last paragraph's pointer to current work. Roger Dearnaley's intuition pump behind his take that the sharp left turn might not be as hopeless as it seems is resonant with me, but his description seems vibes-based so I can't tell if he's misunderstanding the sharp left turn. I do think Dearnaley's personal "full-stack" attempt at assessing alignment progress is the sort of answer I'd want to your question re: what sort of work would be good evidence, although my impression is you disagree for high-level generator reasons that would be ~intractable to resolve within the margins of EA forum comments...