I totally agree on the consequentialist part. Regarding value change, I find Oscar Horta's line of reasoning quite compelling. Value changes might take some time to materialize into actual improvements, but when they do, the effects tend to be long-lasting. For example, we still struggle with certain ideas of Descartes more than 400 years after his death.

On the positive side, the only reason why cage-free campaigns can be successful is that people genuinely care. When it comes to AI, my personal opinion is that values are incredibly important. If these systems become even more powerful than they already are, I would certainly prefer them to think that bullfights are appalling, rather than deciding it is a good idea to book tickets for them without even being asked.

I guess this comes back to the old net-negative-lives debate. If you apply the logic of the logger, having more bulls around is good. Otherwise it's likely bad, even if a particular bull has a net positive live.

Maybe it's worth citing the entire declaration here:

Which animals have the capacity for conscious experience? While much uncertainty remains, some points of wide agreement have emerged.

First, there is strong scientific support for attributions of conscious experience to other mammals and to birds.

Second, the empirical evidence indicates at least a realistic possibility of conscious experience in all vertebrates (including reptiles, amphibians, and fishes) and many invertebrates (including, at minimum, cephalopod mollusks, decapod crustaceans, and insects).

Third, when there is a realistic possibility of conscious experience in an animal, it is irresponsible to ignore that possibility in decisions affecting that animal. We should consider welfare risks and use the evidence to inform our responses to these risks.

It is hard to imagine how one could be any more measured in making a claim, or in what sense it could possibly be considered premature.

As always, a truly insightful post. I would argue, however, that the psychological dimension deserves consideration as well. There is well-established evidence that nature has a profoundly positive impact on mental health. A growing number of people requiring psychiatric care, combined with a shrinking workforce due to mental health conditions, carries significant economic consequences. Furthermore, people are consistently willing to pay a premium for housing near thriving, healthy ecosystems. Once those ecosystems are lost, so too is that value.

... then offsetting with animal welfare interventions means more feed is required, resulting in fewer wild invertebrates 

There are two main effects. Higher welfare standards generally mean fewer animals are raised, since they and their products become more expensive. But those that are raised require more feed. As far as I know there is no consensus on which effect dominates.

However, if the person thinks that wild invertebrates lives are net negative, they would prefer the animal welfare interventions offset, because not only would that help the farmed animals, but it would also reduce the bad utilities of wild invertebrates lives.

They may prefer this kind of intervention and consequently donate to the relevant charities, but the amount they are supposed to pay to offset should stay the same. Since offsetting is generally framed around conservative estimates, it makes no sense to pay less just because you believe in something.

Though this may seem contradictory, I think there is a large variation in difficulty of going vegan (taste preferences, opportunity cost of time, impact on health, etc), so it is most effective if the people for whom it is easier to go vegan are exposed to the arguments.

I totally agree. However, I'm not aware of a single charity that turns people who had no inclination of becoming vegan into vegans. That would be an almost impossible achievement, unless you paid them to do so and set up control mechanisms to make sure they stood true to their word. 

Charities like Veganuary are targeted specifically at people who are already motivated to go vegan in the first place. So if someone turns vegan during the challenge it's impossible to know whether the donor was counterfactual. For all we know the person could have become vegan two months earlier but chose to wait for the challenge. So we may want to use Shapley values. There is also a broader principle at play here: there is general consensus that the same outcome cannot be claimed multiple times for offsetting purposes. This is precisely how certificate trading works, whether for carbon or anything else. So, if we assign some weight to the person who invented the challenge, some weight to the staff who work for below market rates, some weight to every organisation who had previously influenced the person who eventually went vegan, and of course the main weight to that person themselves, little value is left for the donor.

Without access to the relevant data it is hard to say anything definitive, but it seems worth asking whether the cost per outcome would look quite different if conservative estimates were applied more rigorously.

For example, even if we get AI agents to not book tickets to bullfights, I don't see it as a win from a consequentialist perspective (that's not too controversial a take?).

Why would this be? Ending bullfights may be a small victory in the short term as there are relatively few animals are involved. But still a victory.

However I’d argue  it would be a bigger win in value shifts. As long as people enjoy watching a bull tortured in an arena it’s hard to imagine that they will be very sympathetic to animal welfare ideas. Likewise if a system thinks it’s ok to book tickets for bullfights/ propose them without being asked it’s hard to imagine that they deeply care about animal welfare. I’m not an expert in AI by any measure so I’m very interested in pushback from you on this thesis.

Lastly many resources at the moment go into fighting bullfight. If systems start to cancel them  and bullfights get eventually forbidden those resources are freed to work in other areas of interest.

Of course not. At this point it simply isn't possible.Which is exactly what I argued:

Unless we possess solid data on wild animal welfare ranges and how feed production affects them, we have no way of knowing how to offset a non-vegan diet's impact on wild populations, or what the true cost would be. 

I would argue, however, that vegans have less impact on wild animals which is precisely why framing welfare campaigns as an "offset" for a non-vegan diet strikes me as misleading. I have broader reservations about the concept of offsetting animal suffering altogether, but if such offsetting is to be meaningful in any real sense, it should focus on projects that reduce the consumption of animal products in order to replicate these effects.

That's exactly what I said. Uncertainty is a big issue here. But that doesn't mean one is justified in not offsetting the impact one's dietary choices have on wild animals, simply because a handful of people have put forward the hypothesis of net negative lives.

I still feel there must be a way to flip this somehow and find a way to assign and recognise impact and contributions that feels motivating and uplifiting to everybody

I feel it would be interesting, but not necessarily uplifting, to estimate how much the EA community spends on any given intervention. Let's return to your Atlantic Shepherd example. Sea Shepherd is one of those classic charities that handles everything in-house from fundraising and awareness-raising to carrying out the actual anti-poaching interventions. The charities typically recommended by EAs, on the other hand, externalise a great deal of their costs.

Imagine we set up "Atlantic Shepherd" as an EA charity. Before it ever received funding, an organisation like Rethink Priorities would likely have helped make the case for ocean interventions. Grantmakers would have supported them in getting started. Animal Charity Evaluators would assess them. Several other charities would recommend them. The 80,000 Hours podcast would feature them. And then there are all the charities working on building the EA community itself, as well as those raising funds on Atlantic Shepherd's behalf. All of these organisations have spent real resources advancing the cause, yet none of those costs ever appear when analysing Atlantic Shepherd's cost-effectiveness. It would be entirely possible for the community to be spending twice as much per fish saved as Sea Shepherd does, while the cost-effectiveness analysis still shows Atlantic Shepherd coming out ahead by a wide margin.

But I think a lot of the offsetting is corporate campaigns to get higher welfare standards. 

I agree that these campaigns are promising, and I am a donor to Sinergia Animal myself. However, mass animal production creates issues beyond farmed welfare, such as climate change and resource diversion from starving populations. While those can be offset, my main concern is wild animal suffering caused by feed production.

Every farmed animal impacts multiple wild animals through agriculture. Unless we possess solid data on wild animal welfare ranges and how feed production affects them, we have no way of knowing how to offset a non-vegan diet's impact on wild populations, or what the true cost would be. Consequently, I believe, claiming we can offset a non-vegan diet by sponsoring welfare campaigns is highly misleading at this stage.