I’ve been working in animal advocacy for two years and have a very amateur interest in AI. I don’t necessarily agree with all of the possible implications of AI for animals that I’ve set out below, or endorse all of the references that I’ve cited; they’re just a collection of feasible-sounding ideas that I’ve come across. I’m very open to feedback on any misinterpretations of the original arguments, errors, incorrect uses of terminology, or other indications of my lack of expertise.
Many thanks to Yip Fai Tse, Constance Li, Nicholas Kees Dupuis, Sam Tucker, and Vince Mak for taking the time to provide valuable feedback.
I’m writing this in a personal capacity, and am not representing the views of my employer.
Introduction
Humans make up a tiny percentage of life on Earth. There are 8 billion humans in the world. There are 40 billion land animals currently being farmed, over 79 billion insects, 125 billion fish, and 230 billion shrimps. There are a million billion birds, mammals, reptiles, amphibians, and fish living in the wild, and 10 billion billion arthropods (i.e., insects, crustaceans, arachnids, and myriapods, which is the group containing centipedes and millipedes). Of course, we don’t know how the ratio of humans to non-human animals might change in the long-term future (see Abraham Rowe’s post ‘Should Longtermists Mostly Think About Animals?’, including the comments, for some back-and-forth on this) but there’s a good chance that the number of animals will be astronomically large.
AI could fundamentally transform society, and these transformations would have major implications for non-human animals. Both directly and indirectly, AI could mitigate an immense amount of existing animal suffering, or cause an astronomical amount more of it. The rest of this post aims to give a very high-level overview of some of those implications.
As set out in a recent European Parliament report, AI systems could also lead to a clear divide between the ‘haves’ and the ‘have nots’, potentially eliminating smaller players in the farmed animal industry and benefitting larger, better-resourced corporations. This could result in even more animals being factory farmed. (It could also have unpredictable implications for corporate campaign work, with advocacy organizations targeting an increasingly small number of ever wealthier, ever more powerful companies.)
Reducing disease risks associated with increased stocking density
For animal agriculture businesses, AI advancements can mitigate the existing risks of poor animal treatment. For instance, AI-driven monitoring systems can detect early signs of infectious disease, such as avian influenza, potentially enabling producers to stock farmed animals even closer together without facing the costs currently associated with disease outbreaks.
Enabling genetic modifications that increase or mask farmed animal suffering
AI could facilitate advanced breeding techniques and genetic modifications aimed at maximizing farmed animals’ productivity, potentially compromising their well-being. Additionally, while efforts to breed animals that don’t suffer physically or psychologically could mitigate or eliminate a great amount of suffering, it is also possible in principle that such efforts could have terrible consequences. For example, these supposedly pain-insensitive animals might still be suffering, but unable to express that suffering, leading to worse overall welfare outcomes; they might reinforce speciesist beliefs that all animals are just mindless automatons; or they might be successfully implemented by only a very small number of producers, making the whole industry seem more palatable despite limited implementation. While such outcomes could also come about without the use of AI, AI could significantly contribute to accelerating progress in this area.
Increased AI automation in animal farming might diminish the already limited interactions between farmers and animals, further distancing humans from the realities of factory farming and reducing producers’ empathy for the animals in their care. (Though on the other hand, this reduced farmer-animal interaction could also reduce instances of abuse by human handlers.)
This reduced interaction could also require a fundamental restructuring of animal welfare legislation. Existing animal welfare laws typically place responsibility for animal welfare on a ‘person’ or ‘handler’, and hold those individuals responsible for instances of animal cruelty. In a factory farm entirely controlled by AI, whom would this refer to?
Boosting industries like fish and insect farming
AI systems could play a particularly large role in making emerging industries like fish and insect farming more economically viable. Currently, these industries are not nearly as optimized for ruthless efficiency as the farming of most other terrestrial animals, but the integration of AI could rapidly level the playing field in that regard.
For instance, AI could help solve many of the issues preventing offshore aquaculture from becoming economically scalable, such as by automating feeding and water quality monitoring in areas where it is costly and inconvenient to send human personnel. Better Origin, meanwhile, hails itself as the ‘world’s first AI-powered insect farm’. The company says of one of its ‘X1’ insect farms: ‘The X1 is completely automated. It takes care of seeding, feeding, growth, and harvest. You don’t need to know anything about insect farming because our AI does.’
Contributing to the use of farmed insects for biomaterials
Insects are already being farmed in order to produce biomaterials. For example, Insectta farms black soldier flies to produce biomaterials that can be used in semiconductors. AI could therefore not only facilitate insect farming (see point above) but also provide a burgeoning market for it.
Neglecting wild animal welfare
Failing to account for non-human animals in human infrastructure
Human infrastructure already has devastating consequences for non-human animals (such as the hundreds of millions of animals that are already killed as a result of road networks and windows each year in the United States alone). This failure to account for non-human animals is likely to be a problem when it comes to AI-assisted technology and infrastructure in the future. To take the example of self-driving cars: while these may be programmed to avoid large animals that could endanger the humans in the car, they may not be programmed to avoid animals that are too small to cause any damage (with the likely exception of cats and dogs, given the PR risks of self-driving cars killing companion animals).
Accelerating space colonization by humans and non-human animals
AI systems could accelerate space exploration and colonization; for example, a paper by the Future of Humanity Institute sees AI as being pivotal in allowing the level of automation needed for this. We might bring animals with us to these extraterrestrial environments (though this is far from a given). If we did, this could entail a huge level of suffering, depending on whether wild animals’ lives would be overall net-positive or net-negative in these new space environments.
Facilitating unethical means of population control
AI systems could facilitate the large-scale killing of animals deemed ‘invasive’. For example, a government-funded bird conservation project in New Zealand called the ‘Cacophony Project’ uses AI to ‘identify predators automatically using machine learning algorithms’ so that it can poison animals like rats, possums, and stoats en masse.
Backyard Brains already sells a device called the ‘RoboRoach’. In their own words: “We are excited to announce the world’s first commercially available cyborg! With our RoboRoach you can briefly wirelessly control the left/right movement of a cockroach by microstimulation of the antenna nerves. The RoboRoach is a great way to engage with neural microstimulation, learning, and electronics!” While this device is not controlled by AI, AI is likely to open up new possibilities for robotics, making these kinds of devices more advanced and making the production of such devices more widely accessible. This could facilitate trivial applications of AI on animals, entrenching the belief that animals are there for our own amusement, and encouraging companies to come up with new, potentially cruel ways for humans to act on that belief. (However, such technology could also have more worthy applications, such as using remote-controlled insects to find trapped humans in rescue missions.)
Entrenching speciesist thinking and reducing humans’ capacity for active compassion
Similarly, asking AI image generation tools (such as Midjourney or DallE) to generate an image of a ‘happy lobster’ (or crab, or shrimp) will typically produce a drawing of that animal colored in red; i.e., the color they are when cooked, rather than alive. Examples of AI-generated pictures showing cooked salmon ‘swimming’ in a river have gone viral. AI image generation tools are also more likely to e.g. present images of chickens in idyllic outdoor settings, rather than factory farms, potentially supporting the industry’s humane-washing efforts.
Prompting a de-prioritization of animal advocacy due to AI-related x-risks
Major risks associated with AI could redirect human attention and resources away from issues such as animal suffering. (For example, in The Precipice, Toby Ord notes that there are ‘serious concerns about AI entrenching social discrimination, producing mass unemployment, supporting oppressive surveillance, and violating the norms of war’). In a world of fear, societal division, and general chaos, it is likely that animal rights would become an even lesser concern than they are currently. This impact is even more pronounced if you believe, as some have argued, that future improvements to human welfare are likely to be the main driver of improvements to animal welfare.
Opening up the possibility of virtual companion animals
The creation of sentient virtual companion animals could be pretty terrible (as could the creation of digital minds more generally), especially given the possibility that simulated beings could feasibly experience levels of enjoyment or suffering of which flesh-and-blood beings are biologically incapable. There are also a host of other potentially troubling implications, such as the trivialization of animal care, and the possibility that we could create magnificent virtual companion animals that make all ‘regular’ flesh-and-blood animals look extremely dull by comparison, potentially diminishing our level of engagement with their rights and wellbeing. (Ted Chiang’s great novella The Lifecycle of Software Objectsexplores some of these implications in occasionally disturbing detail.)
Improving farmed animal welfare through Precision Livestock Farming
Precision Livestock Farming (PLF) uses advanced technologies to monitor and manage farmed animals. With the integration of AI, PLF can become increasingly efficient, quickly detecting health issues, stress, and other welfare concerns, and addressing these either directly or with human assistance. For example, Project Soundwel ‘aims to understand the encoding of emotion in pig vocalisations and use this knowledge to develop a tool that can assess welfare on-farm by determining the emotion state of pigs through their vocalisations’, while experts at Scotland’s Rural College have developed visual monitoring systems that can recognize individual pigs and detect when they are in distress.
Facilitating inspection of CCTV footage to detect animal welfare violations
AI tools could help increase transparency throughout the animal product supply chain. For example, IdentiGEN apparently uses technology that ‘combines each species’ unique DNA and data analytics to provide an evidence-based animal traceability solution, called DNA TraceBack, to accurately and precisely trace beef, seafood, pork, and poultry that is verifiable from farm-to-table’. This could help regulators ensure the legitimacy of supposedly ‘humane’ animal products, and facilitate the introduction of consumer transparency measures such as animal welfare labeling.
Providing alternatives to farmed animal identification methods
Drones are already being used to administer contraception in order to control certain animal populations. Powered by AI, drones could achieve this increasingly quickly and effectively. (For example, AI-powered drones are already being used to help identify wildfires with apparent success – which, as a side-note, could in itself be fantastic for animals, given that wildfires kill and injure far more non-human animals than humans.) This could ensure that fewer animals suffer and die due to resource scarcity, and also render less ethical forms of population control redundant.
Providing medical assistance to animals in the wild
AI-assisted drones could identify sick animals in the wild and provide medical assistance, or euthanize the animal if they are beyond medical help. (Thanks to Yip Fai Tse for this idea, which emerged during an ‘AI and Animals Idea Jam’ in June 2023.)
Enabling more ambitious and far-reaching interventions to improve wild animal welfare
AI could also support the ideation and execution of far more radical measures to fundamentally improve wild animal welfare. For example, David Pearce’s Abolitionist Project advocates for the use of genetic engineering and nanotechnology to abolish suffering throughout the living world.
Improving welfare of animals used for other purposes
Accelerating the transition to animal-free research
Organizations like Smarter Sorting are using AI to analyze existing toxicity data and extrapolate this to new products based on their ingredients, without the need for further animal testing, while other organizations like Verisim Life use AI to create digital simulations that can replace animal drug testing.
Providing virtual alternatives to animal-based entertainment
AI-driven virtual reality and augmented reality platforms can offer alternatives to traditional forms of entertainment that use animals. For instance, some modern circuses have already begun using holograms of animals rather than real ones, and these virtual animals are likely to become increasingly realistic and engaging with the assistance of advanced AI.
Supporting the ‘genetic disenhancement’ of animals exploited for human purposes
Accelerating the progress of plant-based animal product alternatives
AI could greatly accelerate the development of plant-based animal product alternatives. For example, AI could greatly improve the ‘extrusion process’ that helps create textures and structures that mimic those of traditional animal-based meats. This is the mission of the new non-profit GreenProtein AI: ‘to leverage advanced AI technology to optimize the extrusion process of plant-based meat production’, thereby enhancing ‘the quality, taste, and sustainability of plant-based meat products, all while reducing costs’.
Accelerating the mass production of targeted ingredients for use across different alternative protein industries
AI can enable companies to rapidly optimize the production of alternative protein ingredients by facilitating an understanding of the molecular structure of various proteins and how these can be used and combined with other ingredients, to provide e.g. different textures and nutritional profiles.
Helping solve problems in arable farming
AI technology can help optimize arable farming, such as by monitoring crops to detect disease before it spreads, distinguishing crops from weeds to allow for targeted herbicide use, and enabling self-driving tractors and combine harvesters. This could in principle make it cheaper to grow and purchase plant-based foods. (However, similar applications would of course apply to crops grown for animal feed, thereby increasing the profitability of animal farming.)
Fostering anti-speciesist thinking
Freeing humans up to care more about other animals
If AI were to accelerate economic growth and improve global health, this increased prosperity could better enable people to commit time and money to social justice issues, such as animal rights. People are also more likely to consider animal rights to be a legitimate concern if they are not themselves directly affected by poverty and poor health. (On the other hand, accelerated economic growth over the last few centuries has allowed us to scale up systems that entail huge amounts of animal suffering, such as intensive farming, and this may continue to be the case in the future.)
Fostering mutually reinforcing arguments about caring for animals and caring for sentient digital beings
Organizations such as Sentience Institute investigate considerations around the inclusion of digital minds and artificial sentience in humanity’s moral circle, including the potential of ‘moral spillover’ from one group, such as animals, to another, such as sentient digital minds. There could also feasibly be spillover in the opposite direction, from sentient digital minds to animals. Advocating for the rights and well-being of one group could reinforce the arguments for the other, accelerating a broader cultural shift towards ethical consideration for all sentient beings.
Allowing us to understand animals’ communications
Earth Species Project aims to ‘decode non-human communication’. Interspecies Internet aims to ‘accelerate our understanding of interspecies communication’. Project CETI aims to translate the communication of sperm whales. All of these projects use AI technology in some form or another. Success in this field could help us to understand animals’ needs and recognize them as morally relevant beings. (However, it could also provide humans with new means to manipulate animals. For example, biomimetic robots have successfully been used to communicate with bees in order to change their flight paths. In future, this kind of technology could make it easier for us to use animals for e.g. military purposes).
(Side note: The Animals In That Country is a great dystopian novel that explores what might happen if there were a pandemic that enabled people to understand animals.)
Opening up the possibility of virtual companion animals
Companion animals suffer a fair bit, and potentially much more than we would like to think. They also eat an incredible amount of animal products. Virtual companion animals could allow people to form bonds with animals without the current welfare concerns, both for the animals themselves and for the animals whose flesh and secretions they consume. It would also allow people to look after a companion animal who currently aren’t able to, with potential benefits to the spread of anti-speciesist thinking given the possible link between having a companion animal and harboring moral concern for animals more broadly.
Obviously the goalposts would shift radically if these virtual companion animals were sentient; indeed, this may be where the greatest demand emerges for genuinely sentient digital beings, and therefore the greatest commercial incentive to create them. There are also a whole host of other potentially troubling implications (see section above).
Increasing our recognition of our similarity to other animals, relative to AI systems
As AI systems become more obviously weird and inhuman, our difference from those systems could help to reinforce the extent and significance of our shared biological heritage with animals. This could potentially increase humans’ feeling of companionship with animals, and responsibility for them.
Empowering animal advocates
Increasing animal advocacy organizations’ research capabilities
Animal advocacy organizations could increasingly use AI-assisted tools to improve the quantity and quality of their research. Current examples include research search tools such as Elicit and Consensus, and animal advocacy-specific Large Language Models (LLMs) like VEG3. Tools such as GPT Researcher, which are in theory capable of more complex, multi-step research tasks, are also likely to become increasingly useful as their quality improves.
Facilitating grant-writing and grant-reading
General AI tools like ChatGPT, and specific grant-writing ones like Grantable, could increasingly assist in the grant-writing process. This could be particularly helpful for less well-resourced organizations and/or organizations where staff are not fluent in the language in which the grant application needs to be written. (This could also have a downside, in that grantmaking organizations may have a larger quantity of applications to get through as a result, and may be more suspicious about whether a proposal written using AI accurately reflects the plan of the organization.)
Supporting novel and effective campaigns and messaging
Vegan Hacktivists and Plant Based News recently ran a great webinar on how AI can already be used to create engaging animal advocacy campaigns. AI can help with both text-based campaigns (using models like ChatGPT, Bard, and Claude to generate content, summarize relevant studies and monitor online sources) and image-based (using tools like MidJourney to create visually arresting content that wouldn’t otherwise be possible, such as these pictures of old pigs, cows, and chickens that illustrate how young farmed animals typically are at the time of slaughter).
Tailoring outreach messaging to individuals
AI could also be used to tailor online advocacy to specific individuals; e.g., by analyzing publicly available user data on social media and delivering messaging that is likely to align with their interests and demographics. Tools such as Humantic can already predict personality traits based on social media posts. Combining such technology with a Large Language Model that can retrieve evidence on communication strategies could be used to accurately predict the type of messaging that is most likely to convince a specific individual to carry out a particular kind of behavior change.
Improving the accessibility of animal advocacy content
AI could increasingly make advocacy content more accessible. AI-powered text-to-speech (e.g., using Speechify), translation (e.g., using DeepL), and summarization (e.g., using ChatGPT, Bard, or Claude) could help advocacy messages and research reach a global and diverse audience.
Facilitating animal cruelty investigations
As AI becomes incorporated into an increasing number of devices, any AI system connected to a camera or speaker could be used to record instances of animal cruelty. While this raises much broader privacy concerns, it could also make on-farm investigations much easier. (However, there may be increasing skepticism of animal cruelty footage in a world of incredibly advanced image generation, reducing levels of public engagement and complicating any form of corporate litigation on the basis of animal welfare violations. Some animal advocates might actually end up using fake media, opening the movement up to being discredited even further.)
Increasing animal advocacy organizations’ general productivity
Like all organizations, animal advocacy organizations could increasingly use AI to enhance their productivity and efficiency (e.g., by automating administrative or other repetitive tasks). Tools such as AutoGPT and AgentGPT, which can in principle act more autonomously than e.g. ChatGPT by breaking complex tasks down into smaller sub-tasks, could be used as AI assistants for a huge variety of projects such as corporate campaigns, petitions, or social media outreach.
Further considerations
AI systems might cause a great deal of harm to animals without our realizing
Humans rarely take the interests of animals into account when developing new technologies. For example, when humans invented explosives and drills, it’s unlikely that anyone tried to predict the potential impacts on rodents and underground-dwelling insects who would be killed and have their habitats destroyed when humans used these new inventions to carry out construction work. However, even if humans did care enough to factor in animals’ interests when developing AI, it can be very hard to understand what’s going on beneath the surface of AI systems. For example, in this 80,000 Hours podcast, Richard Ngo says:
‘[...] we really have no idea what’s going on inside the systems that we’re training. So you can get a system that will write you dozens of lines of code, that implements a certain function, that leads to certain outcomes on the screen, and we have no way of knowing what it’s doing internally that leads it to produce that output.’
For many end users, if they provide AI systems with a set of instructions and get a successful outcome, they may be unlikely to question the system’s methods in achieving that outcome. As systems become increasingly powerful, it is feasible that the system’s chosen method could involve significant harm to non-human animals. For example, an AI system tasked with optimizing transportation routes for a delivery company might identify a shortcut through less-traveled rural areas. Beneath the apparent success, this new route might cut through wildlife habitats, resulting in a surge of roadkill incidents. The delivery company would feasibly celebrate the reduced fuel costs and faster delivery times while failing to notice or address the negative consequences for wild animals.
It’s extremely unpredictable what forms AI systems will take and what implications these might have
Seemingly harmless or beneficial technologies can facilitate very harmful outcomes. As noted above, cars and windows kill hundreds of millions of animals each year in the United States alone. Artificial light exacerbates the rate of window collisions, while also disrupting animals’ natural movements, migration patterns, and reproductive cycles. Plastic kills around a hundred million marine animals each year just due to ingestion or entanglement. Take something as potentially powerful and pervasive as AI, and there are countless feasible ways it could be used or abused, intentionally or unintentionally, that we couldn’t hope to predict today.
This is exacerbated by the problem of value extrapolation. As AI systems become more capable, they might face decisions in contexts that were not explicitly covered during their training or programming, and they might extrapolate their values in ways that are not in line with programmers’ original intentions. This just makes things even more unpredictable.
AI could bring about a ‘lock-in’ scenario
AI development could lead to scenarios where certain AI-driven processes become difficult or impossible to reverse. If this occurs, the AI systems’ values at the point of lock-in, and the beings that these are extended to, will heavily dictate how good or bad the future is for all human and non-human animals.
One example of this is Technological-Legal Lock-In. When you turn legal rules and decision-making processes into algorithms, they become more permanent. Altering an established algorithm can be technically challenging and resource-intensive, and might require the approval of many stakeholders. This can make it much less likely that legislation will evolve once it has been incorporated into AI systems.
Conclusion
AI could be incredibly bad for animals. It could precipitate a future where hyper-productive animals suffer horribly in automated mega-facilities that pump out cut-price meat for consumption by disengaged consumers whose only interactions with live animals come from battling sentient Neopets in a simulated fighting pit.
AI could also be incredibly good for animals. It could precipitate a future where advanced technology has rendered all animal exploitation economically redundant and humanity’s newfound prosperity has granted us such levels of wisdom, wellbeing, and financial freedom that our only concern is the invention of ever more ingenious technological means to benefit all sentient life.
It probably won’t be either of those scenarios. But given the astronomical amount of wellbeing at stake, doing as much as possible to shift the dial towards the latter scenario seems like a worthwhile use of resources.
I’m planning on writing a follow-up post setting out ideas for specific actions that animal advocates could take to help contribute to making AI a positive force for animals. If you have any ideas about that, please leave a comment below or send me a private message. Thanks!
What AI could mean for animals
I’ve been working in animal advocacy for two years and have a very amateur interest in AI. I don’t necessarily agree with all of the possible implications of AI for animals that I’ve set out below, or endorse all of the references that I’ve cited; they’re just a collection of feasible-sounding ideas that I’ve come across. I’m very open to feedback on any misinterpretations of the original arguments, errors, incorrect uses of terminology, or other indications of my lack of expertise.
Many thanks to Yip Fai Tse, Constance Li, Nicholas Kees Dupuis, Sam Tucker, and Vince Mak for taking the time to provide valuable feedback.
I’m writing this in a personal capacity, and am not representing the views of my employer.
Introduction
Humans make up a tiny percentage of life on Earth. There are 8 billion humans in the world. There are 40 billion land animals currently being farmed, over 79 billion insects, 125 billion fish, and 230 billion shrimps. There are a million billion birds, mammals, reptiles, amphibians, and fish living in the wild, and 10 billion billion arthropods (i.e., insects, crustaceans, arachnids, and myriapods, which is the group containing centipedes and millipedes). Of course, we don’t know how the ratio of humans to non-human animals might change in the long-term future (see Abraham Rowe’s post ‘Should Longtermists Mostly Think About Animals?’, including the comments, for some back-and-forth on this) but there’s a good chance that the number of animals will be astronomically large.
Non-human animals can experience suffering and wellbeing. This is being increasingly acknowledged by governments (e.g., by the UK Government in their 2022 Animal Welfare (Sentience) Act), scientists (e.g., in the 2012 Cambridge Declaration on Consciousness), and the public (e.g., a 2021 survey by the Sentience Institute, which indicated that 91.8% of people think animals have roughly the same capacity to experience pain as humans).
There is mounting evidence that this applies to animal groups that many would find surprising. For example, the UK’s 2022 Sentience Act mentioned above explicitly includes decapod crustaceans, drawing on Jonathan Birch’s Review of the Evidence of Sentience in Cephalopod Molluscs and Decapod Crustaceans. There has been a great deal of rigorous discussion recently around the capacity for sentience and welfare in insects. And Rethink Priorities’ recent Moral Weight Project tentatively concluded that vertebrate animals’ capacity for pleasure and pain are probably within an order of magnitude of humans’, while invertebrate animals’ capacity for pleasure and pain are probably within two orders of magnitude of vertebrate animals’.
AI could fundamentally transform society, and these transformations would have major implications for non-human animals. Both directly and indirectly, AI could mitigate an immense amount of existing animal suffering, or cause an astronomical amount more of it. The rest of this post aims to give a very high-level overview of some of those implications.
For much more expert views on this topic, please check out papers like Peter Singer and Yip Fai Tse’s ‘AI ethics: the case for including animals’ and Simon Coghlan and Christine Parker’s ‘Harm to Nonhuman Animals from AI: a Systematic Account and Framework’. There’s also a conference on this topic (at Princeton University, including a webinar option) on October 6th-7th 2023 entitled ‘Artificial Intelligence, Conscious Machines, and Animals: Broadening AI Ethics’.
Ways that AI could be bad for animals
Perpetuating factory farming
Increasing the profitability of factory farming through efficiency gains
AI-assisted Precision Livestock Farming (PLF) could make factory farming more profitable by e.g. enabling increased stocking densities, reducing rates of animal disease, and streamlining slaughter and processing. This could help factory farming to remain a financially viable industry, and could further reduce the price of conventional meat for producers and consumers, potentially making it harder for alternative proteins to out-compete conventional meat (as mentioned e.g. by Yip Fai Tse in this video interview).
As set out in a recent European Parliament report, AI systems could also lead to a clear divide between the ‘haves’ and the ‘have nots’, potentially eliminating smaller players in the farmed animal industry and benefitting larger, better-resourced corporations. This could result in even more animals being factory farmed. (It could also have unpredictable implications for corporate campaign work, with advocacy organizations targeting an increasingly small number of ever wealthier, ever more powerful companies.)
Reducing disease risks associated with increased stocking density
For animal agriculture businesses, AI advancements can mitigate the existing risks of poor animal treatment. For instance, AI-driven monitoring systems can detect early signs of infectious disease, such as avian influenza, potentially enabling producers to stock farmed animals even closer together without facing the costs currently associated with disease outbreaks.
Enabling genetic modifications that increase or mask farmed animal suffering
AI could facilitate advanced breeding techniques and genetic modifications aimed at maximizing farmed animals’ productivity, potentially compromising their well-being. Additionally, while efforts to breed animals that don’t suffer physically or psychologically could mitigate or eliminate a great amount of suffering, it is also possible in principle that such efforts could have terrible consequences. For example, these supposedly pain-insensitive animals might still be suffering, but unable to express that suffering, leading to worse overall welfare outcomes; they might reinforce speciesist beliefs that all animals are just mindless automatons; or they might be successfully implemented by only a very small number of producers, making the whole industry seem more palatable despite limited implementation. While such outcomes could also come about without the use of AI, AI could significantly contribute to accelerating progress in this area.
Making factory farming seem more palatable
By (supposedly) addressing certain welfare issues, AI systems might provide factory farming with an ethical facade, potentially reducing the perceived urgency for more radical changes in livestock farming practices. This could be reinforced by any contributions by AI systems, meaningful or not, to make factory farming operations more environmentally friendly and decrease the risk of zoonotic disease outbreaks.
Further reducing farmer-animal interaction
Increased AI automation in animal farming might diminish the already limited interactions between farmers and animals, further distancing humans from the realities of factory farming and reducing producers’ empathy for the animals in their care. (Though on the other hand, this reduced farmer-animal interaction could also reduce instances of abuse by human handlers.)
This reduced interaction could also require a fundamental restructuring of animal welfare legislation. Existing animal welfare laws typically place responsibility for animal welfare on a ‘person’ or ‘handler’, and hold those individuals responsible for instances of animal cruelty. In a factory farm entirely controlled by AI, whom would this refer to?
Boosting industries like fish and insect farming
AI systems could play a particularly large role in making emerging industries like fish and insect farming more economically viable. Currently, these industries are not nearly as optimized for ruthless efficiency as the farming of most other terrestrial animals, but the integration of AI could rapidly level the playing field in that regard.
For instance, AI could help solve many of the issues preventing offshore aquaculture from becoming economically scalable, such as by automating feeding and water quality monitoring in areas where it is costly and inconvenient to send human personnel. Better Origin, meanwhile, hails itself as the ‘world’s first AI-powered insect farm’. The company says of one of its ‘X1’ insect farms: ‘The X1 is completely automated. It takes care of seeding, feeding, growth, and harvest. You don’t need to know anything about insect farming because our AI does.’
Contributing to the use of farmed insects for biomaterials
Insects are already being farmed in order to produce biomaterials. For example, Insectta farms black soldier flies to produce biomaterials that can be used in semiconductors. AI could therefore not only facilitate insect farming (see point above) but also provide a burgeoning market for it.
Neglecting wild animal welfare
Failing to account for non-human animals in human infrastructure
Human infrastructure already has devastating consequences for non-human animals (such as the hundreds of millions of animals that are already killed as a result of road networks and windows each year in the United States alone). This failure to account for non-human animals is likely to be a problem when it comes to AI-assisted technology and infrastructure in the future. To take the example of self-driving cars: while these may be programmed to avoid large animals that could endanger the humans in the car, they may not be programmed to avoid animals that are too small to cause any damage (with the likely exception of cats and dogs, given the PR risks of self-driving cars killing companion animals).
Accelerating space colonization by humans and non-human animals
AI systems could accelerate space exploration and colonization; for example, a paper by the Future of Humanity Institute sees AI as being pivotal in allowing the level of automation needed for this. We might bring animals with us to these extraterrestrial environments (though this is far from a given). If we did, this could entail a huge level of suffering, depending on whether wild animals’ lives would be overall net-positive or net-negative in these new space environments.
Facilitating unethical means of population control
AI systems could facilitate the large-scale killing of animals deemed ‘invasive’. For example, a government-funded bird conservation project in New Zealand called the ‘Cacophony Project’ uses AI to ‘identify predators automatically using machine learning algorithms’ so that it can poison animals like rats, possums, and stoats en masse.
Worsening the welfare of other animal groups
Justifying animal testing
A lot of AI research depends on animal testing, such as research on brain-computer interfaces (which allow direct communication between the brain and an external device, such as a robotic limb). Neuralink is working on a brain-computer interface implant and has killed over 1,500 animals since 2018 in the name of research, and apparently caused significant suffering through negligence and malpractice.
Enabling trivial and exploitative experiments
Backyard Brains already sells a device called the ‘RoboRoach’. In their own words: “We are excited to announce the world’s first commercially available cyborg! With our RoboRoach you can briefly wirelessly control the left/right movement of a cockroach by microstimulation of the antenna nerves. The RoboRoach is a great way to engage with neural microstimulation, learning, and electronics!” While this device is not controlled by AI, AI is likely to open up new possibilities for robotics, making these kinds of devices more advanced and making the production of such devices more widely accessible. This could facilitate trivial applications of AI on animals, entrenching the belief that animals are there for our own amusement, and encouraging companies to come up with new, potentially cruel ways for humans to act on that belief. (However, such technology could also have more worthy applications, such as using remote-controlled insects to find trapped humans in rescue missions.)
Entrenching speciesist thinking and reducing humans’ capacity for active compassion
Normalizing speciesist biases
AI could perpetuate and amplify speciesist tendencies present in the data it is trained on. For instance, ChatGPT’s current responses reflect the human belief that it is ethical to eat pig meat and the livers of force-fed geese (though not dog meat or cat liver).
Similarly, asking AI image generation tools (such as Midjourney or DallE) to generate an image of a ‘happy lobster’ (or crab, or shrimp) will typically produce a drawing of that animal colored in red; i.e., the color they are when cooked, rather than alive. Examples of AI-generated pictures showing cooked salmon ‘swimming’ in a river have gone viral. AI image generation tools are also more likely to e.g. present images of chickens in idyllic outdoor settings, rather than factory farms, potentially supporting the industry’s humane-washing efforts.
Prompting a de-prioritization of animal advocacy due to AI-related x-risks
Major risks associated with AI could redirect human attention and resources away from issues such as animal suffering. (For example, in The Precipice, Toby Ord notes that there are ‘serious concerns about AI entrenching social discrimination, producing mass unemployment, supporting oppressive surveillance, and violating the norms of war’). In a world of fear, societal division, and general chaos, it is likely that animal rights would become an even lesser concern than they are currently. This impact is even more pronounced if you believe, as some have argued, that future improvements to human welfare are likely to be the main driver of improvements to animal welfare.
Opening up the possibility of virtual companion animals
The creation of sentient virtual companion animals could be pretty terrible (as could the creation of digital minds more generally), especially given the possibility that simulated beings could feasibly experience levels of enjoyment or suffering of which flesh-and-blood beings are biologically incapable. There are also a host of other potentially troubling implications, such as the trivialization of animal care, and the possibility that we could create magnificent virtual companion animals that make all ‘regular’ flesh-and-blood animals look extremely dull by comparison, potentially diminishing our level of engagement with their rights and wellbeing. (Ted Chiang’s great novella The Lifecycle of Software Objects explores some of these implications in occasionally disturbing detail.)
There are also various other ways in which humans might use advanced AI to simulate sentient animals in virtual environments, such as to simulate how life might evolve on different planets, which raise similar ethical concerns.
Ways that AI could be good for animals
Improving farmed animal welfare
Improving farmed animal welfare through Precision Livestock Farming
Precision Livestock Farming (PLF) uses advanced technologies to monitor and manage farmed animals. With the integration of AI, PLF can become increasingly efficient, quickly detecting health issues, stress, and other welfare concerns, and addressing these either directly or with human assistance. For example, Project Soundwel ‘aims to understand the encoding of emotion in pig vocalisations and use this knowledge to develop a tool that can assess welfare on-farm by determining the emotion state of pigs through their vocalisations’, while experts at Scotland’s Rural College have developed visual monitoring systems that can recognize individual pigs and detect when they are in distress.
Facilitating inspection of CCTV footage to detect animal welfare violations
AI can help humans analyze vast amounts of CCTV footage in real-time from slaughterhouses. This means quicker detection of any animal welfare violations, helping ensure that these violations are promptly addressed – particularly if used by regulators, rather than by employees of the slaughterhouses themselves.
Promoting transparency in the food industry
AI tools could help increase transparency throughout the animal product supply chain. For example, IdentiGEN apparently uses technology that ‘combines each species’ unique DNA and data analytics to provide an evidence-based animal traceability solution, called DNA TraceBack, to accurately and precisely trace beef, seafood, pork, and poultry that is verifiable from farm-to-table’. This could help regulators ensure the legitimacy of supposedly ‘humane’ animal products, and facilitate the introduction of consumer transparency measures such as animal welfare labeling.
Providing alternatives to farmed animal identification methods
Farmed animals are often mutilated in order to assist with identification, such as with ear tags or branding. AI tools could identify animals from their natural physical features, making such mutilations redundant.
Improving wild animal welfare
Enabling more ethical forms of population control
Drones are already being used to administer contraception in order to control certain animal populations. Powered by AI, drones could achieve this increasingly quickly and effectively. (For example, AI-powered drones are already being used to help identify wildfires with apparent success – which, as a side-note, could in itself be fantastic for animals, given that wildfires kill and injure far more non-human animals than humans.) This could ensure that fewer animals suffer and die due to resource scarcity, and also render less ethical forms of population control redundant.
Providing medical assistance to animals in the wild
AI-assisted drones could identify sick animals in the wild and provide medical assistance, or euthanize the animal if they are beyond medical help. (Thanks to Yip Fai Tse for this idea, which emerged during an ‘AI and Animals Idea Jam’ in June 2023.)
Making human infrastructure safer for animals
AI systems can track the movements of animals to prevent potential hazards. For example, AI systems are already helping to guide the construction of road and train underpasses based on the concentration of certain species, and they could also reduce collisions between bats and wind turbines. Self-driving cars could also be programmed to identify and avoid animals on the road much more effectively than human drivers are currently able to (though see the section above for concerns about self-driving cars having speciesist biases encoded).
Enabling more ambitious and far-reaching interventions to improve wild animal welfare
AI could also support the ideation and execution of far more radical measures to fundamentally improve wild animal welfare. For example, David Pearce’s Abolitionist Project advocates for the use of genetic engineering and nanotechnology to abolish suffering throughout the living world.
Improving welfare of animals used for other purposes
Accelerating the transition to animal-free research
Organizations like Smarter Sorting are using AI to analyze existing toxicity data and extrapolate this to new products based on their ingredients, without the need for further animal testing, while other organizations like Verisim Life use AI to create digital simulations that can replace animal drug testing.
Providing virtual alternatives to animal-based entertainment
AI-driven virtual reality and augmented reality platforms can offer alternatives to traditional forms of entertainment that use animals. For instance, some modern circuses have already begun using holograms of animals rather than real ones, and these virtual animals are likely to become increasingly realistic and engaging with the assistance of advanced AI.
Supporting the ‘genetic disenhancement’ of animals exploited for human purposes
New gene-editing technologies make it theoretically possible to genetically manipulate animals to reduce or eliminate their capacity for suffering, allowing humans to continue using them for research, farming, and other purposes, (arguably) without the same degree of ethical concerns. (However, this technology could also have terrible implications – see section above.)
Developing alternatives to animal products
Accelerating the progress of cellular agriculture
AI could greatly accelerate the development of cellular agriculture. For example, Machine Learning can help predict the best ingredients and conditions for the media used in cellular agriculture, making the process more efficient and cost-effective. The Alberta Machine Intelligence Institute (Amii) and New Harvest recently launched a one-year research project specifically focused on the applications of Machine Learning in cellular agriculture.
Accelerating the progress of plant-based animal product alternatives
AI could greatly accelerate the development of plant-based animal product alternatives. For example, AI could greatly improve the ‘extrusion process’ that helps create textures and structures that mimic those of traditional animal-based meats. This is the mission of the new non-profit GreenProtein AI: ‘to leverage advanced AI technology to optimize the extrusion process of plant-based meat production’, thereby enhancing ‘the quality, taste, and sustainability of plant-based meat products, all while reducing costs’.
Accelerating the mass production of targeted ingredients for use across different alternative protein industries
AI can enable companies to rapidly optimize the production of alternative protein ingredients by facilitating an understanding of the molecular structure of various proteins and how these can be used and combined with other ingredients, to provide e.g. different textures and nutritional profiles.
Helping solve problems in arable farming
AI technology can help optimize arable farming, such as by monitoring crops to detect disease before it spreads, distinguishing crops from weeds to allow for targeted herbicide use, and enabling self-driving tractors and combine harvesters. This could in principle make it cheaper to grow and purchase plant-based foods. (However, similar applications would of course apply to crops grown for animal feed, thereby increasing the profitability of animal farming.)
Fostering anti-speciesist thinking
Freeing humans up to care more about other animals
If AI were to accelerate economic growth and improve global health, this increased prosperity could better enable people to commit time and money to social justice issues, such as animal rights. People are also more likely to consider animal rights to be a legitimate concern if they are not themselves directly affected by poverty and poor health. (On the other hand, accelerated economic growth over the last few centuries has allowed us to scale up systems that entail huge amounts of animal suffering, such as intensive farming, and this may continue to be the case in the future.)
Fostering mutually reinforcing arguments about caring for animals and caring for sentient digital beings
Organizations such as Sentience Institute investigate considerations around the inclusion of digital minds and artificial sentience in humanity’s moral circle, including the potential of ‘moral spillover’ from one group, such as animals, to another, such as sentient digital minds. There could also feasibly be spillover in the opposite direction, from sentient digital minds to animals. Advocating for the rights and well-being of one group could reinforce the arguments for the other, accelerating a broader cultural shift towards ethical consideration for all sentient beings.
Allowing us to understand animals’ communications
Earth Species Project aims to ‘decode non-human communication’. Interspecies Internet aims to ‘accelerate our understanding of interspecies communication’. Project CETI aims to translate the communication of sperm whales. All of these projects use AI technology in some form or another. Success in this field could help us to understand animals’ needs and recognize them as morally relevant beings. (However, it could also provide humans with new means to manipulate animals. For example, biomimetic robots have successfully been used to communicate with bees in order to change their flight paths. In future, this kind of technology could make it easier for us to use animals for e.g. military purposes).
(Side note: The Animals In That Country is a great dystopian novel that explores what might happen if there were a pandemic that enabled people to understand animals.)
Opening up the possibility of virtual companion animals
Companion animals suffer a fair bit, and potentially much more than we would like to think. They also eat an incredible amount of animal products. Virtual companion animals could allow people to form bonds with animals without the current welfare concerns, both for the animals themselves and for the animals whose flesh and secretions they consume. It would also allow people to look after a companion animal who currently aren’t able to, with potential benefits to the spread of anti-speciesist thinking given the possible link between having a companion animal and harboring moral concern for animals more broadly.
Obviously the goalposts would shift radically if these virtual companion animals were sentient; indeed, this may be where the greatest demand emerges for genuinely sentient digital beings, and therefore the greatest commercial incentive to create them. There are also a whole host of other potentially troubling implications (see section above).
Increasing our recognition of our similarity to other animals, relative to AI systems
As AI systems become more obviously weird and inhuman, our difference from those systems could help to reinforce the extent and significance of our shared biological heritage with animals. This could potentially increase humans’ feeling of companionship with animals, and responsibility for them.
Empowering animal advocates
Increasing animal advocacy organizations’ research capabilities
Animal advocacy organizations could increasingly use AI-assisted tools to improve the quantity and quality of their research. Current examples include research search tools such as Elicit and Consensus, and animal advocacy-specific Large Language Models (LLMs) like VEG3. Tools such as GPT Researcher, which are in theory capable of more complex, multi-step research tasks, are also likely to become increasingly useful as their quality improves.
Facilitating grant-writing and grant-reading
General AI tools like ChatGPT, and specific grant-writing ones like Grantable, could increasingly assist in the grant-writing process. This could be particularly helpful for less well-resourced organizations and/or organizations where staff are not fluent in the language in which the grant application needs to be written. (This could also have a downside, in that grantmaking organizations may have a larger quantity of applications to get through as a result, and may be more suspicious about whether a proposal written using AI accurately reflects the plan of the organization.)
Supporting novel and effective campaigns and messaging
Vegan Hacktivists and Plant Based News recently ran a great webinar on how AI can already be used to create engaging animal advocacy campaigns. AI can help with both text-based campaigns (using models like ChatGPT, Bard, and Claude to generate content, summarize relevant studies and monitor online sources) and image-based (using tools like MidJourney to create visually arresting content that wouldn’t otherwise be possible, such as these pictures of old pigs, cows, and chickens that illustrate how young farmed animals typically are at the time of slaughter).
Tailoring outreach messaging to individuals
AI could also be used to tailor online advocacy to specific individuals; e.g., by analyzing publicly available user data on social media and delivering messaging that is likely to align with their interests and demographics. Tools such as Humantic can already predict personality traits based on social media posts. Combining such technology with a Large Language Model that can retrieve evidence on communication strategies could be used to accurately predict the type of messaging that is most likely to convince a specific individual to carry out a particular kind of behavior change.
Improving the accessibility of animal advocacy content
AI could increasingly make advocacy content more accessible. AI-powered text-to-speech (e.g., using Speechify), translation (e.g., using DeepL), and summarization (e.g., using ChatGPT, Bard, or Claude) could help advocacy messages and research reach a global and diverse audience.
Facilitating animal cruelty investigations
As AI becomes incorporated into an increasing number of devices, any AI system connected to a camera or speaker could be used to record instances of animal cruelty. While this raises much broader privacy concerns, it could also make on-farm investigations much easier. (However, there may be increasing skepticism of animal cruelty footage in a world of incredibly advanced image generation, reducing levels of public engagement and complicating any form of corporate litigation on the basis of animal welfare violations. Some animal advocates might actually end up using fake media, opening the movement up to being discredited even further.)
Increasing animal advocacy organizations’ general productivity
Like all organizations, animal advocacy organizations could increasingly use AI to enhance their productivity and efficiency (e.g., by automating administrative or other repetitive tasks). Tools such as AutoGPT and AgentGPT, which can in principle act more autonomously than e.g. ChatGPT by breaking complex tasks down into smaller sub-tasks, could be used as AI assistants for a huge variety of projects such as corporate campaigns, petitions, or social media outreach.
Further considerations
AI systems might cause a great deal of harm to animals without our realizing
Humans rarely take the interests of animals into account when developing new technologies. For example, when humans invented explosives and drills, it’s unlikely that anyone tried to predict the potential impacts on rodents and underground-dwelling insects who would be killed and have their habitats destroyed when humans used these new inventions to carry out construction work. However, even if humans did care enough to factor in animals’ interests when developing AI, it can be very hard to understand what’s going on beneath the surface of AI systems. For example, in this 80,000 Hours podcast, Richard Ngo says:
‘[...] we really have no idea what’s going on inside the systems that we’re training. So you can get a system that will write you dozens of lines of code, that implements a certain function, that leads to certain outcomes on the screen, and we have no way of knowing what it’s doing internally that leads it to produce that output.’
For many end users, if they provide AI systems with a set of instructions and get a successful outcome, they may be unlikely to question the system’s methods in achieving that outcome. As systems become increasingly powerful, it is feasible that the system’s chosen method could involve significant harm to non-human animals. For example, an AI system tasked with optimizing transportation routes for a delivery company might identify a shortcut through less-traveled rural areas. Beneath the apparent success, this new route might cut through wildlife habitats, resulting in a surge of roadkill incidents. The delivery company would feasibly celebrate the reduced fuel costs and faster delivery times while failing to notice or address the negative consequences for wild animals.
It’s extremely unpredictable what forms AI systems will take and what implications these might have
Seemingly harmless or beneficial technologies can facilitate very harmful outcomes. As noted above, cars and windows kill hundreds of millions of animals each year in the United States alone. Artificial light exacerbates the rate of window collisions, while also disrupting animals’ natural movements, migration patterns, and reproductive cycles. Plastic kills around a hundred million marine animals each year just due to ingestion or entanglement. Take something as potentially powerful and pervasive as AI, and there are countless feasible ways it could be used or abused, intentionally or unintentionally, that we couldn’t hope to predict today.
This is exacerbated by the problem of value extrapolation. As AI systems become more capable, they might face decisions in contexts that were not explicitly covered during their training or programming, and they might extrapolate their values in ways that are not in line with programmers’ original intentions. This just makes things even more unpredictable.
AI could bring about a ‘lock-in’ scenario
AI development could lead to scenarios where certain AI-driven processes become difficult or impossible to reverse. If this occurs, the AI systems’ values at the point of lock-in, and the beings that these are extended to, will heavily dictate how good or bad the future is for all human and non-human animals.
One example of this is Technological-Legal Lock-In. When you turn legal rules and decision-making processes into algorithms, they become more permanent. Altering an established algorithm can be technically challenging and resource-intensive, and might require the approval of many stakeholders. This can make it much less likely that legislation will evolve once it has been incorporated into AI systems.
Conclusion
AI could be incredibly bad for animals. It could precipitate a future where hyper-productive animals suffer horribly in automated mega-facilities that pump out cut-price meat for consumption by disengaged consumers whose only interactions with live animals come from battling sentient Neopets in a simulated fighting pit.
AI could also be incredibly good for animals. It could precipitate a future where advanced technology has rendered all animal exploitation economically redundant and humanity’s newfound prosperity has granted us such levels of wisdom, wellbeing, and financial freedom that our only concern is the invention of ever more ingenious technological means to benefit all sentient life.
It probably won’t be either of those scenarios. But given the astronomical amount of wellbeing at stake, doing as much as possible to shift the dial towards the latter scenario seems like a worthwhile use of resources.
I’m planning on writing a follow-up post setting out ideas for specific actions that animal advocates could take to help contribute to making AI a positive force for animals. If you have any ideas about that, please leave a comment below or send me a private message. Thanks!