Against Against Against Against Welfare Tech

and related excessive pedantry about how we use words.

Introduction

The animal advocacy movement has become obsessed with something that people refer to as “welfare tech,” or technologies that improve animal welfare.

There are new “venture studios“ (I’m not Silicon Valley enough to not add quotes, sorry) to launch businesses and philanthropy developing technologies. We are awash in screeds, anti-screeds, and anti???-screeds arguing over whether or not welfare tech is the first step on the path to animal liberation via consciousnessless chickens (and whether that’s a good thing at all). An animal welfare nonprofit runs a broiler chicken farm and people are unhappy about that. For some people, welfare tech fills an obvious gap in the ecosystem, while for others, it is an accelerant for factory farming.

But these proposals for technology often lack specifics, or vaguely point to in-ovo sexing and shrimp stunners as examples of welfare tech whose success we might replicate. The basic idea, in its most common form, is “start new companies that develop technologies that are good for animals, then get farms to deploy them.” Maybe they deploy them because of pressure from advocates, or maybe because they are commercially viable.

To lay my cards on the table:

• I’m a committed utilitarian, willing to swallow most bitter pills that come with that, and am fairly motivated by suffering-focused concerns — I think we should do the things that most reduce the suffering of animals most effectively.

• I don’t think factory farming will end, at least in the foreseeable future. I basically have zero hope in diet change work, and only slightly more than zero in alternative protein technologies making meaningful dents in animal farming in the near future.

• I wouldn’t call myself an optimist about helping animals — I’m quite pessimistic about our ability to help animals generally, and think that animal advocates too frequently ignore the largest sources of animal suffering, because they might point to inconvenient and complicated conclusions.

  • For example, I don’t believe, but I wouldn’t be surprised if the net effect of the animal advocacy movement so far is negative (not even from big-brain wild animal views, but due to decreasing beef production and increasing smaller animal production), despite what I view as the incredible progress of cage-free campaigns and shrimp stunning (which hopefully actually helps shrimps and doesn’t just electrocute them).

• I generally think technological progress is good for humans and bad for animals, but intentionally developing technology to help animals is one of the best things we can do for them.

I think all of the above should make me a shoo-in for loving welfare tech. And in the abstract, I do — I don’t think we’ll end factory farming, so I think we should use technology to make it better. I’m very excited about all these new projects to make new animal-helping technologies!

But I think the conversation about welfare tech is pretty confused and messy! Not all welfare tech is the same, and not all of it seems obviously better. This is my attempt to de-confuse the welfare tech conversation, and highlight what I think the best path forward for it is.

Welfare tech isn’t well-defined

My impression from the welfare tech debates is that the term “welfare tech” isn’t currently well-defined. I think this is a problem, because it is leading to confusion about the theory of change. Are welfare tech boosters “propping up factory farming?” It strongly depends on the specific intervention.

In my view, we should think of welfare tech as technology (which itself is not something easy to define) that, when applied on a farm, makes animal welfare better. Basically, technology that improves the welfare of the animals it is applied to. But under this definition, not all welfare tech will be equally important.

Spring Innovation Fund describes welfare tech as technology that improves animal welfare in ways that make commercial sense for producers to adopt. But we don’t have many examples of this, to my knowledge. The most frequently cited examples, in-ovo sexing for male chicks and shrimp stunners, are both more expensive than their alternatives, and their adoption was primarily driven by regulation and advocacy, not commercial benefit. Of course, advocacy pressure is a type of commercial pressure — “if we don’t do this protestors will show up and disrupt our business” is a business decision, after all. But I take “commercial sense” to mean something like “independent of any other factors, it is commercially sensible for producers to adopt this technology.”

But by citing the obvious technological wins of in-ovo sexing and shrimp stunners as examples of successful welfare tech, we smuggle in ideas about what welfare tech is, or the theoretical viability of commercial investment. In my view, in-ovo sexing and shrimp stunning are successes of advocacy, not of technology alone. We haven’t actually shown we can develop technologies that farms will adopt on their own, and will be good for welfare.

Additionally, welfare tech, as broadly used, refers to things that different parties might be more or less excited about.

• Animal liberationists should be excited by in-ovo sexing — it not only reduces the number of animals on farms, it makes eggs marginally more expensive. It doesn’t make sense for hatcheries to adopt on their own absent advocacy pressure.

• Backfire-sensitive welfarists should be excited by technologies like shrimp stunning — it improves animal welfare, but likely doesn’t increase the number of animals farmed.

• Non-backfire-sensitive welfarists should be excited by breeding for welfare — it probably improves welfare for individual animals, but might allow farmers to push farm efficiency higher, so there is some risk of a tradeoff.

Basically, calling all of these “welfare tech” implies that they are one kind of thing, when they are actually a broad suite of things, and under different values, might be more or less beneficial.

I’m trying to provide a breakdown of the types of welfare tech that have been proposed to be concrete about where these disagreements might lie, and what successful “welfare tech” work might look like.

So what is welfare tech?

There are lots of ways that technology intersects with animal welfare, and we might advance “welfare tech” via different routes:

• Developing and deploying new welfare technologies

• Figuring out what existing technologies are better for animals then deploying them

• Getting rid of existing technology that is worse for animals

Sometimes technology is good for animals and sometimes it is bad for animals. So we should develop and promote the good stuff and get rid of the bad stuff. I don’t really see these as that different fundamentally, given that our track record with welfare tech mainly runs through advocacy and outreach, though usually only the first is referred to as welfare tech.

I don’t think we should refer to technologies that solely reduce the number of animals farmed as “welfare tech,” because while they might improve animal welfare in the aggregate, there is no individual animal who has a better life due to them. For example, I don’t think developing and selling alternative proteins is “welfare tech” if it leads to decreased animal farming. For this reason, to be excessively pedantic, I also don’t think we should refer to in-ovo sexing as welfare tech — it is on-farm technology, yes, but it doesn’t improve any animal’s welfare — it reduces the number of animals needed to produce a food output.

What are the different types of “welfare tech”?

On-farm technology improvements for welfare

The most obvious versions of “welfare tech” are “we develop a new technology and deploy it on a farm, helping animals” or “we take an existing technology and apply it to farms to improve welfare.” As examples:

• Immunocastration:

  • Since 1998, we’ve had the ability to castrate pigs via a vaccine, instead of via unanesthetized surgery.

  • Immunocastration is also more profitable for farmers! It could increase profitability by $5-$15 per pig for US farmers.

  • Most pigs in Brazil are immunocastrated, and around 30% of Canadian pigs are. Despite this, few US, EU, or Chinese pigs are (despite it being legal in the US since 2011).

• Ozempic for chickens

  • Modern broiler chickens used for breeding are, for various reasons, often very hungry, and face a variety of stress-related welfare harms due to it. Robert Yaman has suggested giving these breeders GLP-1 drugs to curb their hunger.

  • We aren’t making anything new here — we’re just taking an existing drug, and giving it to animals.

  • Actually deploying this, as Yaman points out, might involve further drug development or deployment technologies.

• Controlled atmosphere stunning

  • The Better Chicken Commitment recommends controlled atmosphere stunning, where basically chickens are rendered unconscious via a sealed, low-oxygen gas chamber before being slaughtered.

  • Typical practice, live-shackle slaughter, involves taking live chickens, shackling them upside down, and dragging them through an electrified water bath to stun them prior to slaughter. The stunners sometimes miss, and presumably the entire thing is very stressful.

  • The basic idea of controlled atmosphere stunning has been around for something like 150 years, and has been fully developed for chickens for 30 years. Advocates pushed for the adoption of a non-novel technology.

• Non-shrimp crustacean stunners

  • Some welfare tech work has focused on developing crustacean stunners.

  • We’ve got shrimp stunners already, so maybe we can just go and stun all the crustaceans?

  • This seems hopefully good for animals, but we aren’t really doing anything novel — we’re taking an existing technology, and applying it to new animals.

Technologies like immunocastration are the cleanest example of “welfare tech” as described by proponents. While immunocastration isn’t new, its development was genuinely novel, though notably it was developed to reduce “boar taint,” an unpleasant odor and taste found in some meat from uncastrated pigs, not to improve animal welfare. And, as Spring highlights, it might make commercial sense for producers to adopt it.

New animal genetics

Many welfare tech advocates are also interested in “breeding for welfare,” or genetically modifying (via selective breeding or other techniques) breeds to be higher welfare. The idea is that by making animals who are less sensitive to pain or stress, or more resilient in the face of life on the factory farm, we could see higher welfare on farms.

• Breeding for welfare

  • Advocates have discussed breeding chickens with reduced hunger or stress.

  • Then, when they are on farms where they face various stressors, these chickens would be less hungry and stressed!

  • Maybe in some cases, this could reduce mortality, making farms more profitable, so industry would want to adopt these breeds.

• More speculative things

  • Some people have discussed more speculative things, like breeding perpetually happy, or consciousnessless chickens. If we did this, we’d see a large improvement in welfare, presumably.

  • We don’t really know how to do these things right now — making clean meat at scale is really really hard, but I suspect breeding a consciousnessless or perpetually happy chicken is even harder.

De-teching

Sometimes technology is bad and we should just get rid of it!

• Cage-free laying systems

  • In the 1930s, some people had the idea to put laying hens in small cages. This turned out poorly for the hens.

  • In the mid-2010s, some animal advocates got the idea to maybe stop doing that, and got a lot of restaurants, grocery stores, etc. to stop buying these eggs. This turned out to probably be the best thing for farmed animals we’ve ever done, and in my view, is the crowning accomplishment of the animal welfare movement.

  • So, sometimes, for welfare, we want to de-tech.

Not welfare tech at all, just tech

Some supposed welfare tech is just technology that reduces animal farming or suffering. That’s it. They aren’t welfare tech (or if they are, everything that helps animals is welfare tech). They just cause there to be less animal farming, or make existing technology cheaper (but don’t improve welfare more per animal). That’s cool, and helps animals, so we should do it! But I wouldn’t call it welfare tech.

• In-ovo sexing

  • In-ovo sexing lets us identify the sex of a chick before an egg hatches. Then, instead of culling male chicks whose female siblings are destined to become laying hens, we destroy the egg, theoretically avoiding a death shortly after the egg hatches.

  • I don’t think we should call this welfare tech. The theory of change is the same as advocating for reducing animal product consumption via blending mushrooms in with beef — you now need fewer animals to produce some animal product. No individual animal has a better life — the intervention causes fewer animals to be used in farming. I think of animal welfare as improving the welfare of specific animals or reducing the suffering of specific animals, not reducing the number born into the farming system.

  • In-ovo sexing is amazing! I’m complaining about terminology, but it is incredibly impressive that since I started working professionally to help animals, we’ve just kind of cracked how to... reduce the number of chickens hatched into the laying hen supply chain by roughly 50%.

• Clean meat, theoretically

  • Assuming it is ever actually a substitute for animal products, clean meat is a technological advancement that reduces animal farming via people purchasing fewer animal products. Like in-ovo sexing, it would cause fewer animals to be born then killed.

What should we think about welfare tech?

While I think welfare tech is used to refer to many things, I think that many proposals mainly fit into the on-farm improvements categories I list above. So, I’m going to focus on those when describing how I think about welfare tech.

I am both excited and nervous about welfare tech. I’m excited because I think advocates need more tools in their toolkit that robustly improve animal welfare, and new technology is one of the best ways to get there. I’m nervous because I think welfare tech is perilous, and we have to be careful how we deploy. So, I’m going to raise questions and concerns I have about the technology, then outline the case I see for beneficial welfare tech.

What has driven welfare tech adoption?

A core claim of welfare tech advocates is that we can make technologies that improve animal welfare and are commercially sensible for companies to adopt. But so far, I don’t think we’ve seen much evidence of being able to find these. Let’s look at the three big wins often claimed for welfare tech:

In-Ovo Sexing

In-ovo sexing has been widely adopted in Europe (around 40% of layers born are processed via it). The US is far behind, with only 4% committed to do so in the future.

So why did Europe adopt in-ovo sexing, and why hasn’t the US?

• Europe

  • Germany banned culling due to a lawsuit under its Animal Welfare Act, which required culling to be phased out once there was a suitable alternative. Once in-ovo sexing matured as a technology, there was such an alternative.

  • In France, animal advocacy group L214 campaigned on banning chick culling, which led to political pressure that resulted in requiring sexing equipment to be used.

  • After this, several other countries passed regulations in response to animal advocacy pressure.

  • Some countries, like Norway and Switzerland, adopted it voluntarily, in response to advocacy and to be competitive in the European market (though this was notably easy because supply chain was highly concentrated — Norway needed just two machines to convert their entire national egg supply).

  • So effectively, European adoption was driven by advocates getting regulations passed, forcing companies to change practices. It also helped that for European layers, in-ovo sexing can generally be cheap — white in-ovo sexed eggs likely now cost less than $0.01 more per egg, and for brown layers, it is even cheaper (because the technology is cheapest for brown layers). Europe has lots of brown layers, and thus it was cheap for these practices to change.

• US

  • US chickens are overwhelmingly white feathered. The cheapest machines sex eggs by reading feather color through shells, but this doesn’t work on white chickens. US hatcheries have to use more expensive MRI or biomarker systems.

    • This suggests that while the cost differences are slight, they might still matter! If there was consumer demand for in-ovo sexing, it might have led to greater uptake in the US.

  • The US also has some issues passing animal welfare laws generally, so there hasn’t been regulatory pressure.

So, the story of in-ovo sexing is roughly that advocacy forced adoption in Europe, which has gone well, but in the US, we haven’t seen progress, despite the low cost. This doesn’t seem like a clean victory solely on the basis of the tech itself, but a victory due to a combination of advocacy, policy, and technology.

Shrimp Stunning

Shrimp stunning is entirely unnecessary for commercial shrimp production, and does not change the cost of shrimp meaningfully. The shrimp are stunned then put on ice slurries. Before stunners, they were just put on ice slurries.

Shrimp stunners were first adopted because Shrimp Welfare Project literally bought them for farmers. Then, adoption increased because corporate campaigners pressured retailers in the UK and elsewhere to adopt animal welfare policies that required shrimp stunning.

Shrimp stunning might be an impressive deployment of technology, but it provides no benefit at all to farmers — there aren’t any particularly strong reasons, for example, to think that farmers will pay out-of-pocket to replace their stunners when the current ones break down, etc.

Immunocastration

Immunocastration theoretically should save farmers $5-$15 per pig. Animal advocates advocate for it on these grounds! Unlike shrimp stunning and in-ovo sexing, it actually might make commercial sense for farmers to adopt it. Despite this, the US and Europe have seen very low adoption, less than 5% in both regions. But this is our perfect welfare tech — why hasn’t it worked (yet)?

In the US, immunocastration vaccines operate under various verification protocols that require technician training, quality assurance, etc. These might make it hard for it to diffuse across the supply chain.

In Europe, in 2010, the EU targeted phasing out surgical castration by 2018, but basically hasn’t met this goal. Some EU countries have banned surgical castration without anesthetic, and other alternatives to immunocastration are also used (such as killing pigs before puberty, selective castration to reduce the risk of boar taint, etc). Despite there theoretically being regulatory pressure, farmers seem to sometimes use other alternatives that meet regulatory requirements. There is also some fear of “vaccinated” pork among producers.

Overall, immunocastration offers a strange counterexample to the core ideas of welfare tech promotion. It should be more widely used just due to its economics, but isn’t, because it isn’t the only alternative that farmers have (when forced by policy), and there are some PR concerns.

Who should pay for welfare tech?

Lots of donor money may go to welfare tech initiatives. As someone who is unambiguously pro-research (and who thinks the majority of animal welfare funding should be spent on research), I don’t see this as an issue in-and-of-itself. But the logic underlying animal donors investing in for-profit welfare tech companies is slightly confusing:

• If these products are going to be commercially viable, why won’t non-aligned venture capital pay for their development (especially since this would be vastly more effective than aligned donors paying)?

• If these products aren’t commercially viable, how will they be adopted? We’ll need to also pay for advocacy and policy to implement them.

One answer might be that many of these products live in the “valley of death,” a phenomenon where commercially viable products don’t take off because no one wants to pay the high costs for their development. This is a good reason for animal advocates to pay for these technologies — non-aligned venture capital might not see the reward as being great enough to take a bet on something with a 5% chance of working and a relatively small addressable market, but animal advocates care less about this.

Animal advocates have been here before, and should tread carefully. In the late 2010s, as alternative protein investment skyrocketed, many donors also invested in these companies, hoping to make more money to spend on animal welfare. Plenty of unaligned capital was flowing into these companies, but animal donors doubled down. I hold that this was a bad idea, given stagnant growth in the market for alternative proteins in subsequent years, and the losses some donors incurred from these investments. Donor investments today, when outside investment appetite is lower, might be much more effective than they were at the peak of the boom.

Overall, I’m incredibly excited about a model of organizations developing new, carefully considered welfare tech, and working in coordination with advocates to deploy it, as the models that we’ve seen for shrimp stunning and in-ovo sexing have demonstrated can work. So what is carefully considered welfare tech?

What welfare tech might be beneficial?

In my view, not all welfare tech is created equal. We need to identify robust animal welfare improvements that can be made with welfare tech. By robust, I mean improvements that we are confident:

• Are actually good for animal welfare.

• Aren’t increasing the amount of animal farming to a degree that offsets the welfare benefit.

Unfortunately, I don’t think all welfare tech shares these features.

Why might an animal welfare improvement be worse for animals?

The Jevons paradox

The Jevons paradox is an economic phenomenon in which when technological improvements lead to an efficiency increase in the production or use of a resource, there is a rise in use of that resource. The canonical example was coal use in the 19th century — observers expected increasing efficiency of coal consumption to decrease coal use, because you needed less to produce the same amount of power. Instead, demand rose, because you could do more activities more cheaply.

Similarly, Jevons concerns might arise for welfare tech. For example, if you reduce the rate of disease on a farm, you might decrease mortality. This might mean that farmers can sell their products more cheaply, because loss from disease is lower, leading to more animals being farmed overall.

I worry that many welfare tech products will have Jevons issues. In particular, if our theory is that it makes commercial sense for producers to adopt these technologies, then it is almost explicitly the case that we expect them to improve efficiency — why else would producers adopt them on economics alone? So, we’ll need to be very careful to ensure our practices aren’t suffering from Jevons issues severe enough to offset the welfare gains.

Of course, you might not expect all Jevons issues to be a problem — if you farm 100 animals, at -5 welfare (-500 net), and improve it to -4 while causing an increase to 120 animals (-480 net), you’ve still improved overall welfare. But in practice, these economic effects are hard to predict and while in my toy model I can pretend to know the animal welfare benefit, in reality measuring animal welfare is not easy. Treading carefully seems warranted. We likely can’t reason in advance about these effects accurately, and will need empirical data from seeing how producers actually adopt these technologies and change prices.

Robert Yaman has argued that the Jevons paradox doesn’t apply to chicken farming. Like some of the commenters, I think this is an over-simplification of a very complex set of dynamics, but the points are well made. Basically, my view is “Jevons issues are real, tread carefully when increasing efficiency, and do reversible things because predicting economic effects is hard.”

Responsive practices

One of the most common mistakes I see the animal welfare space making is assuming that they are acting in a vacuum, where other actors will not be responsive to their actions. If we spend more money, those who disagree with us will also spend more. When deploying welfare tech, we also need to consider how the industry will respond. I think this is one of the bigger risks for “breeding for welfare”. If we create animals with increased stress tolerance, for example, what stops industry from just increasing intensification to the point that animals reach their new economically-viable stress limit? Or, if the highest welfare tech we invent is expensive, but a less-good-but-cheaper alternative can be created, what stops farmers from instead switching to that cheaper version? We need to carefully consider how others will respond to the deployment of our technology.

Doing good welfare tech

All caveats aside, I think there are specific pathways for welfare tech that are quite exciting!

Slaughter technology

Animal slaughter is often painful. We can make it less painful. Usually, we don’t need to mess with efficiency to do this. When we can, we should find technologies like shrimp stunners, which render animals unconscious prior to slaughter. This seems less likely to face Jevons issues (though in some cases it still might, if it speeds up slaughter rates), and seems like a clean welfare improvement.

Replacing a specific cruel practice

Like pig castration, many farming practices are cruel. If we can find ways to replace them with less painful alternatives that don’t change the profitability or efficiency of farms, we should!

Pairing with policy and advocacy

The history of welfare tech is not a story of welfare tech succeeding — it’s a story of advocacy advancing the deployment of welfare tech. We need to think carefully about the strategy for deployment of welfare tech, and likely have a high degree of coordination with advocacy groups to advance welfare on farms. The best technologies for welfare seem almost certainly to be economically inefficient for farms to adopt, so we should likely expect that there are promising avenues for impact through this pairing.

Reversibility

Shrimp stunners are not well-evidenced as an animal welfare improvement, but they seem, under the right circumstances, to work to render shrimp unconscious. But they have another feature that is highly desirable for uncertain interventions — they are reversible. If we stop advocating for their use, farmers likely won’t pay for them (because they don’t increase profitability), and as the current ones break down, they won’t be replaced by farmers without advocates also continuing to push (though the more adoption is driven by corporate commitments, the less true this may be). Animal welfare science and economics are hard. If we aren’t confident in the benefit of an intervention, but think it might be good, we should do it reversibly.

We need animal welfare science more than we need technological improvement

Fundamentally, all welfare tech is bottlenecked on the most basic problem of animal welfare: we have little idea how to help animals, or how to measure what their lives are like.

As an example, the most well-funded animal welfare funder has spent millions over several years funding science to determine if cage-free systems are higher welfare than battery cages. The result: under reasonable assumptions we have decent-but-not-amazing evidence in favor of cage-free, but also face some tricky philosophical tradeoffs some might disagree with. This is hardly reassuring about making easy progress on studying animal welfare. Of course, there has been some progress — for example, we are fairly confident that effective stunning at slaughter is better than live slaughter for mammals. But the point is, animal welfare science is slow, uncertain work. While one challenge for welfare tech is identifying technical solutions to animal welfare issues on farms, a much harder problem is identifying what those animal welfare challenges actually are. This is a problem not easily solved by commercial investment — often it might need years of research with no obvious payoff.

When I look at the history of cage-free, in some sense, I think the welfarists got lucky. While there were some reasons for thinking it might be better than caged systems in 2015, when the campaign scaled up, the evidence really wasn’t robust. Then, we built the evidence base, and the decision, in retrospect, seems like a good one. For shrimp stunning, similarly, the effort was launched without the level of confidence in animal welfare science we’d want. I think that shrimp stunning will ultimately prevail as a benefit to animals, but we genuinely do not yet know, and by advancing it we are taking a risk. If we make another 10 such risky bets, some will probably fail, and end up hurting animals. If we make these risky bets as independent, self-perpetuating for-profit companies, we might not be able to reverse that failure.

As anyone who has studied animals will tell you, we can’t really just look at an animal and tell if something is good for it. We have to infer from behavioral and physiological indicators. Doing this is expensive, takes a long time, and is fraught and uncertain work. Animal welfare science is in its infancy. Before we massively invest in new technologies, we should make sure we are confident they are actually helping animals. The animal movement that wants to make factory farms way better than they are today needs a massive scale up of welfare science far more than it needs technological progress, but that’s much harder to fundraise for because it offers no promise of a self-adopting flywheel and frankly, isn’t as cool-sounding.

I think that the biggest bottleneck in animal welfare, both farmed and wild, is actual empirical information on how to make the lives of animals better and what downstream effects those changes will have on other animals. A type of welfare tech that is under-discussed is tools to monitor animals and collect better data on their welfare. Building a robust and massive field of animal welfare science, ecological modeling, and animal welfare economics will provide massive returns to the entire movement, and make welfare tech far easier to deploy confidently in the future.