Animal welfare guy tuning in. My own take is that the majority of the world actually is almost entirely indifferent about animal suffering, so if AI tries to reflect global values (not just the values of the progressive, elite silicon valley bubble) there is a real risk that it will be indifferent to animal suffering. Consider how Foie Gras is still legal in most countries, or bullfighting, both of which are totally unnecessary. And those are just examples from western countries.
I think it’s very likely that TAI will lock in only a very mild concern for animal welfare. Or perhaps, concern for animal welfare in certain contexts (e.g. pets), and none in others (e.g. chicken). Maybe that will lead to a future without factory farming, but it will lead to a future with unnecessary animal suffering nonetheless.
What I’m not sure about is: how do we ensure that TAI locks in a strong valuation of animal welfare? One route is to try to change how much society cares about animal welfare, and hope that TAI then reflects that. I guess this is the hope of many animal advocates. But I admit that seems too slow to work at this stage, so I agree that animal advocates should probably prioritize trying to influence those developing AI right now.
Tristan Katz
Agree. The factor on which we are weighting animals—their ability to feel—is objective, even if our assessment of it is uncertain (although it’s becoming increasingly certain).
One can disagree that this is the factor that we should be focusing on, but I’m yet to see such an argument which isn’t also speciesist.
I am familiar with Kymlicka’s work, and honestly I think this has little to do with effective altruism because the evidence component is lacking. The idea that animals live in “communities” has been strongly critiqued (see: Humanitarian intervention in nature: crucial questions and probable answers by Adriano Mannino) - a better analogy is that they live in a state of civil war. We cannot expect that they will be able to reason about their conflicts and collective action problems. Furthermore, Kymlicka’s claims might work for highly intelligent animals, but most wild animals are very small and the hope of doing politics ‘with them’ is a pipe dream. Unfortunately, Kymlicka has not responded to these criticisms.
It’s been clarified in the comments that with the statement “I think the reality is that we might never get there” what was meant was that we might not end factory farming within our lifetime. Well, if your focus is on reducing suffering, why does it matter less if that suffering is reduced in the next generation, or the one after that?
Animals will continue to suffer terribly as long as factory farming exists. Changing the size of cages, or the exact type of feed given, only changes that suffering by degrees. But others have pointed out that we are making progress towards the long-term goal of ending factory farming altogether. Doing so will reduce animal suffering much more dramatically, and that seems like a long-term goal worth pursuing to me.
You need to consider the counterfactual in its entirety. Ending factory farming increases net welfare if being inhabit the space of the factory farms have more positive welfare than the beings on the factory farms.
I find it incredibly hard to believe that any alternative to factory farming will have an even lower level of welfare, regardless of whether factory farmed animals have lives worth living or not.
Hi, sorry I somehow missed this until just now—far too late! Still I feel I need to say that I think you’ve missed the point. With windmills it is not necessarily the case that the average bird will live a shorter life. If some birds are killed by windmills, then others will have more to eat or will have more habitat (whatever factor it is that otherwise limits this bird population) and will then succumb to that death later. Or perhaps with windmills their main predator will just scavenge on the dead birds, rather than predating them. The point is, there is no reason to assume that the average life will be shorter—either way, some factors are limiting the population, and without directly comparing them we just cannot make this claim.
Establishing a norm of caring about wild animals is good, but it seems to me that you are only encouraging people to care about human-caused harms, rather than the welfare of wild animals generally. There is a risk that people only care about the dirtiness of their own hands, rather than the animals themselves.
I hope that makes sense.
I think my argument will be even clearer if I talk about mitigating AI risk. Imagine if all AI safety orgs operated independently, even competing with each other. It would be (or arguably already is) a mess! There would be no ‘open letter’, just different people shouting separately. And surely AI safety could be advanced further if the already existing orgs worked together better.
So yes, choice is good, but to some degree we are and should be working towards common goals.
I like the analogy, and I found myself agreeing a lot with your suggestions, but I think there is a danger in this analogy:
It portrays EAs as functioning very individualistically, and this could ultimately be ineffective.
Imagine a market where many of the buyers want to make a cake (say the cake represents improving global health—other people may go to the market for other reasons, but many go for the same reason: in this case, cake). Each only has a little bit of money though, not enough to buy all the ingredients for a cake, so they buy one hoping that others will buy the others, and that they will be working off the same recipe. Inevitably, though, they have different ideas of the cake they want to make and what you end up with is a mess of different ingredients. The buyers DO get to eat something, but it isn’t cake. What this represents in the real world is that EAs have targeted several individual health problems, but they haven’t actually worked towards reducing global health/poverty as a whole.
Now imagine an alternative market where the buyers coordinate. They know they all want cake, so they discuss together what the best recipe would be. Then, when they have one, they organise who should buy what—and then together they are able to make a damn good cake. In the real world, this would mean tackling global poverty in a coordinated way, which involves addressing systemic change.
What I’m pointing out is that if the buyers were to coordinate they might be able to do far more good. I think this is currently a big problem in EA, that we don’t coordinate or think strategically enough. We focus on what individual donors can do, and thereby miss out on tackling the bigger problems. So, I like your analogy, but I don’t want people to think that what we ought to have is a classic free market with self-interested actors. Rather, we should have a free market with (at least some) actors working for (at least some) common good(s).
Thanks for the detailed and informative post. My response is a bit critical, but I mean it constructively:
After reading the title of the post, I was hopeful to hear about the potential ways that AI could be used to reduce wild animal suffering—and so a bit disappointed to see this only as a minor consideration in section 8.3. Give that it is wild animals that we are concerned about here (not just removing our guilt), surely we should be looking at their lives as a whole, not just how we affect them directly. So when we look at something like wind turbines, we shouldn’t just be interested in how many birds are killed by them—we should be interested in how this changes the lives of such wild birds in general (including those that don’t get killed, and now have greater access to food or habitat).
My own opinion is that your consideration in 8.3 “HWC deaths may be less painful than counterfactual deaths” could very well be correct. There seems to be no good reason for thinking that it is incorrect. Therefore, to advocate for these solutions to HWC seems to be a very non-ideal way at improving WAW. Given the uncertainty here, it seems that we should focus not on reducing the deaths that we cause, but rather on reducing the most painful kinds of deaths. These are unlikely to be from things such as wind turbines—and much more likely to be from disease, starvation etc. I think this needs to be the central focus of the discussion.
Thanks for this post—it was desperately needed—but it’s striking to me how many questions there are for which we don’t have good answers. I would go so far as to say we’re largely clueless as to what effects AGI will have on animals.
The recommendations that we try to direct the movement toward considering the role of AI in its future, and try to influence AI decision makers rather than the general public, seem reasonable. Maybe that’s the best we can do?