Thank you for writing this! I have been thinking about some ideas that could become mega projects, just throwing some of them out here (you have already listed some of them)
Pay to install electric stunners in “small fish slaughter machines” which is popular in China. The idea is to pay way higher than the cost of installing such stunners so that the whole industry that produces this machine is disrupted. I am doing research on this potential project. My tentative judgement is that installing such stunners might be cheap—it could be as simple as connecting electricity between the gear wheels that push the fish into the machine. According to the producers’ claims, these machines can kill 100-15000 fish per hour, depending on the species. I estimate that a $100 payment per machine is huge enough to incentivise the majority of these machines’ producers to agree to a deal.
Invent a new “cleansing machine” for crayfish (I estimate that Chinese people eat 4,000B of them each year). The special thing about crayfish is that they are so dirty for human consumption they require heavy “cleansing” before being cooked. Shallow research shows that the currently most common method for “cleansing” them is ultrasonic bath while they are still fully alive (for about 20 minutes), which seems extremely painful if they are sentient (in this video you can see the crayfish crawling out to escape the ultrasonicing water.) The idea is to invent a new method that is much less painful and sell the machine at a price that would crowd out all other methods. Without virtually any serious research, it seems like electrical bath before the cleaning starts is the way to go.
AI For Animals (This could be the name of a new charity)
AI to monitor the welfare situation (not according to factory farms’ definitions) in factory farms, transportation, and retail spots. This is especially important for fish, because there are many species farmed
AI to speed up PB/CM research. A lot of PB/CM startups now have 1-2 computer scientist/data scientist. Instead of each of them hiring expensive CS/DS/ML researchers and doing uncollaborated work, start a research hub to distribute the new science and technology
AI to decipher animal “language”. This is something that a few academic teams are already working on, but with small project scales, and on animtals that are not farmed. This is a moonshot—for maybe the truth is that high level nonhuman animal-human communication is impossible. But if we can directly communicate with them, or at least know what they want to express, it will help us understand a lot more about nonhuman animals, and probably help advocacy (imagine a knowing a fish’s complaint or pleading!)
AI for tackling wild animal suffering. This is a big idea with potentially many sub-applications:
AI (drones) to identify dying wild animals and euthanize them. This approach even avoids the problem/accusation that we might cause unpredictable changes and therefore possibly harms while we change what happens in ecosystems, for example, if a deer felt off a cliff and painfully waits for the death, whether the death is cause by time, hunger, a predator, or a drone (as far as the killing doesn’t leave stuff like bullets or toxic chemicals) likely won’t affect what will happen inside that ecosystem. The same might even be said in some cases where the animals are not yet dead—for example, if a forest fire is definitely going kill certain animals and the AI deemed it impossible to move those animals out of the fire’s affecting zone, the AI drones can kill them before the fire reach them. Again in this case whether the animals are dead before the fire reaches them won’t affect the ecosystem, the only difference is only the amount of suffering that happens (burning alive is often thought to be one of the most painful deaths).
AI to track/count wild animals. Quite a number of teams are already doing this, but none of them are interested in the suffering/welfare of individual animals. If a team is interested in tackling WAS the animals to train the AI to track/count would be very different. Being able to count wild animals, monitor their movements, and changes in population, are all hugely important for research that guide wild animal welfare interventions.
AI to identify wild animal welfare levels. This is the wild animal version of the farmed animal welfare identification AI, but probably much harder as wild animals are not in controlled environments and data gathering and data curation will be much harder.
Computer models to predict the effects of interventions. There are many people working on this, but as far as my knowledge goes those people are mostly, if not entirely, interested in “classical conservation” which is not interested in the welfare of individual animals. We need models that can predict the change of net welfarein a system when certain interventions are introduced to a system.
These are interesting ideas! I think that AI systems designed with animal welfare in mind would be more reliant on computer vision and sensory data than NLP, since animals don’t speak in human tongues. This blog post about using biologgers to measure animal welfare comes to mind.
I’m nervous about implementing AI solutions in the near-term, because, as you allude, what they are used to achieve is matter of who’s programming them :/
These are very interesting. The electric stunning can be both beneficial in the way that animals, if they are at least intuitively aware what they live for—to maybe be eaten or produce animal products and be eaten, then if this is they all their life just chill and then it is just a stun, then it’s quite ok. If they could they would probably contribute further, by some advancement, but since we currently only can use their contributions in this way, they may be quite ok just chilling taking care of their life. I read there at least were issues with the stunning machines in US slaughterhouses—simple technical issues—poor placement or inadequate current. Also, ritual killing is an issue. Stunning is more elegant and should be the new ritual.
Electrical bath for crayfish makes sense too. It can be just a simple electrode which prevents the issues of crawling (and thus loss of crayfish and capital). Of course, the alternative of eating rare tofus can be even better but for the time being—there should be manufacturers that would gladly produce this device.
AI monitoring welfare—I would not implement it, maybe in a few years when institutions become more interested in monitoring—it’s a moonshot plus there may be other tech solutions with higher marginal cost-effectiveness. For example, actually, if you focus on cricket farms—I think that if they miss simple nutrients, such as salt, they eat each others. This can extensively slump the atmosphere there for large numbers of individuals. So, some maybe salinity/humidity/etc monitoring device that even a worker can go around with and just poke around and depending on the values nutrients are automatically dispersed. Of course, insect welfare research should perhaps be prioritized because what if crickets just love the thrill of eating others and being eaten since they live to the fullest or suffer in any case so optimal salinity makes very little difference.
I think plant-based is more promising with the cost right now than cultured meat. Still, the equipment is suboptimal since it is made for meat. Probably, these are relatively simple engineering solutions.
Animal language can be deciphered based on evolutionary empathy. For example, when one really eats some vegetable, they can feel like a bug in the same situation. When they are unsure about an unfamiliar object and looking at it, like a bird in that situation. You probably do not need to ‘require’ animals to communicate, since it is quite clear what they want—or, it is similar to what one can perceive in some globally poor/disempowered contexts: individuals do not have their own objectives since no one has asked them. I wonder if this should be different with animals: if developing their own interests would be a challenge/detrimental to their subjective wellbeing since they are used to/capable of only very simple lives.
You need research on wild animal welfare before you can be reasonably certain that interventions will be welcome. For example, animals can negotiate/make agreements/cooperate by the exhibit of power. Since they cannot imagine treatment and specialize in order to increase efficiencies by trade little (are independent of others outside of their family/tribe, more competing for scarce resources or benefiting from others’ death), they can just have very different attitudes to concepts such as longevity, disability-adjusted life year. For them, it can be ‘are what they should and protect those who they should’ or not. It is a good life—righteousness transcends pain, which is accepted by its inevitability. Maybe. Engagement of similar humans can help elucidate non-humans’ way of thinking.
Yes, you would need to know the quality of the lives of the number of animals to have valuable data. Even the number, and other metrics, can be beneficial, if maybe in the future the quality is associated with these metrics and historical developments inform optimal solutions as well as present welfare states.
Hm, yes, computer models that track the developments of populations, e. g. based on predation rates are ok but the welfare is missing.
Thank you for writing this! I have been thinking about some ideas that could become mega projects, just throwing some of them out here (you have already listed some of them)
Pay to install electric stunners in “small fish slaughter machines” which is popular in China. The idea is to pay way higher than the cost of installing such stunners so that the whole industry that produces this machine is disrupted. I am doing research on this potential project. My tentative judgement is that installing such stunners might be cheap—it could be as simple as connecting electricity between the gear wheels that push the fish into the machine. According to the producers’ claims, these machines can kill 100-15000 fish per hour, depending on the species. I estimate that a $100 payment per machine is huge enough to incentivise the majority of these machines’ producers to agree to a deal.
Invent a new “cleansing machine” for crayfish (I estimate that Chinese people eat 4,000B of them each year). The special thing about crayfish is that they are so dirty for human consumption they require heavy “cleansing” before being cooked. Shallow research shows that the currently most common method for “cleansing” them is ultrasonic bath while they are still fully alive (for about 20 minutes), which seems extremely painful if they are sentient (in this video you can see the crayfish crawling out to escape the ultrasonicing water.) The idea is to invent a new method that is much less painful and sell the machine at a price that would crowd out all other methods. Without virtually any serious research, it seems like electrical bath before the cleaning starts is the way to go.
AI For Animals (This could be the name of a new charity)
AI to monitor the welfare situation (not according to factory farms’ definitions) in factory farms, transportation, and retail spots. This is especially important for fish, because there are many species farmed
AI to speed up PB/CM research. A lot of PB/CM startups now have 1-2 computer scientist/data scientist. Instead of each of them hiring expensive CS/DS/ML researchers and doing uncollaborated work, start a research hub to distribute the new science and technology
AI to decipher animal “language”. This is something that a few academic teams are already working on, but with small project scales, and on animtals that are not farmed. This is a moonshot—for maybe the truth is that high level nonhuman animal-human communication is impossible. But if we can directly communicate with them, or at least know what they want to express, it will help us understand a lot more about nonhuman animals, and probably help advocacy (imagine a knowing a fish’s complaint or pleading!)
AI for tackling wild animal suffering. This is a big idea with potentially many sub-applications:
AI (drones) to identify dying wild animals and euthanize them. This approach even avoids the problem/accusation that we might cause unpredictable changes and therefore possibly harms while we change what happens in ecosystems, for example, if a deer felt off a cliff and painfully waits for the death, whether the death is cause by time, hunger, a predator, or a drone (as far as the killing doesn’t leave stuff like bullets or toxic chemicals) likely won’t affect what will happen inside that ecosystem. The same might even be said in some cases where the animals are not yet dead—for example, if a forest fire is definitely going kill certain animals and the AI deemed it impossible to move those animals out of the fire’s affecting zone, the AI drones can kill them before the fire reach them. Again in this case whether the animals are dead before the fire reaches them won’t affect the ecosystem, the only difference is only the amount of suffering that happens (burning alive is often thought to be one of the most painful deaths).
AI to track/count wild animals. Quite a number of teams are already doing this, but none of them are interested in the suffering/welfare of individual animals. If a team is interested in tackling WAS the animals to train the AI to track/count would be very different. Being able to count wild animals, monitor their movements, and changes in population, are all hugely important for research that guide wild animal welfare interventions.
AI to identify wild animal welfare levels. This is the wild animal version of the farmed animal welfare identification AI, but probably much harder as wild animals are not in controlled environments and data gathering and data curation will be much harder.
Computer models to predict the effects of interventions. There are many people working on this, but as far as my knowledge goes those people are mostly, if not entirely, interested in “classical conservation” which is not interested in the welfare of individual animals. We need models that can predict the change of net welfare in a system when certain interventions are introduced to a system.
These are interesting ideas! I think that AI systems designed with animal welfare in mind would be more reliant on computer vision and sensory data than NLP, since animals don’t speak in human tongues. This blog post about using biologgers to measure animal welfare comes to mind.
I’m nervous about implementing AI solutions in the near-term, because, as you allude, what they are used to achieve is matter of who’s programming them :/
These are very interesting. The electric stunning can be both beneficial in the way that animals, if they are at least intuitively aware what they live for—to maybe be eaten or produce animal products and be eaten, then if this is they all their life just chill and then it is just a stun, then it’s quite ok. If they could they would probably contribute further, by some advancement, but since we currently only can use their contributions in this way, they may be quite ok just chilling taking care of their life. I read there at least were issues with the stunning machines in US slaughterhouses—simple technical issues—poor placement or inadequate current. Also, ritual killing is an issue. Stunning is more elegant and should be the new ritual.
Electrical bath for crayfish makes sense too. It can be just a simple electrode which prevents the issues of crawling (and thus loss of crayfish and capital). Of course, the alternative of eating rare tofus can be even better but for the time being—there should be manufacturers that would gladly produce this device.
AI monitoring welfare—I would not implement it, maybe in a few years when institutions become more interested in monitoring—it’s a moonshot plus there may be other tech solutions with higher marginal cost-effectiveness. For example, actually, if you focus on cricket farms—I think that if they miss simple nutrients, such as salt, they eat each others. This can extensively slump the atmosphere there for large numbers of individuals. So, some maybe salinity/humidity/etc monitoring device that even a worker can go around with and just poke around and depending on the values nutrients are automatically dispersed. Of course, insect welfare research should perhaps be prioritized because what if crickets just love the thrill of eating others and being eaten since they live to the fullest or suffer in any case so optimal salinity makes very little difference.
I think plant-based is more promising with the cost right now than cultured meat. Still, the equipment is suboptimal since it is made for meat. Probably, these are relatively simple engineering solutions.
Animal language can be deciphered based on evolutionary empathy. For example, when one really eats some vegetable, they can feel like a bug in the same situation. When they are unsure about an unfamiliar object and looking at it, like a bird in that situation. You probably do not need to ‘require’ animals to communicate, since it is quite clear what they want—or, it is similar to what one can perceive in some globally poor/disempowered contexts: individuals do not have their own objectives since no one has asked them. I wonder if this should be different with animals: if developing their own interests would be a challenge/detrimental to their subjective wellbeing since they are used to/capable of only very simple lives.
You need research on wild animal welfare before you can be reasonably certain that interventions will be welcome. For example, animals can negotiate/make agreements/cooperate by the exhibit of power. Since they cannot imagine treatment and specialize in order to increase efficiencies by trade little (are independent of others outside of their family/tribe, more competing for scarce resources or benefiting from others’ death), they can just have very different attitudes to concepts such as longevity, disability-adjusted life year. For them, it can be ‘are what they should and protect those who they should’ or not. It is a good life—righteousness transcends pain, which is accepted by its inevitability. Maybe. Engagement of similar humans can help elucidate non-humans’ way of thinking.
Yes, you would need to know the quality of the lives of the number of animals to have valuable data. Even the number, and other metrics, can be beneficial, if maybe in the future the quality is associated with these metrics and historical developments inform optimal solutions as well as present welfare states.
Hm, yes, computer models that track the developments of populations, e. g. based on predation rates are ok but the welfare is missing.