Also, in regards to Bob's book, it seems like it would make sense, rather than adding the model's estimates together to instead act on each model separately, i.e. assigning 5% of resources to help animals according to the equality model, 25% according to the neurophysiological model, and 70% according to the proxies model. Then, within the 70% for the proxies model, we could assign 33% of resources to applying a high exponent to the welfare ranges, 33% to applying no exponent to the welfare ranges, and 33% to applying a low exponent to the welfare ranges.

Something like this is proposed in the article "Moral Uncertainty, Proportionality and Bargaining".

Only physical consciousness is falsifiable? I think so. So I believe it is better to focus on physicalist theories of consciousness. Relatedly, you may be interested in this discussion between me and Wladimir Alonso from the Welfare Footprint Institute (WFI) about whether sentience is binary.

Hi Gregory. Great post.

I mentioned before that IRT is basically an Elo for the game of responding correctly to benchmark questions. For Elo, here’s the win probability of player A vs. player B, given respective strengths Sa and Sb:

P(A wins) = 1 / [1 + 10^((Sb - Sa)/400)]

Nitpick. I think the above is the probability of winning plus half of the probability of drawing.

I broadly agree with the points you made in that comment.

You may be interested in my estimates of the total welfare of various organism populations assuming welfare per fully-happy-animal-year is a power law of the individual number of neurons, or of the basal metabolic rate (BMR) at 25 ºC. Here is some context about why a power lay may be appropriate.

Hi Aidan and Aaron. Great news. The program sounds very promising.

Meanwhile, the animal movement needs more new interventions and organizations in the pipeline so that efforts to fix factory farming can scale in the coming years.

I like that you say "fix" instead of "end" factory farming. Fixing it is more aligned with increasing animal welfare, as there is large uncertainty about which animals have positive or negative lives (worth living or not from their own perspective).

Thanks, Anthony.

I think it would help a lot if you spelled out the premises more, because they're quite opaque to me as written. E.g. I don't know what "the norm reveals a choice function" means. (I think this kind of use of jargon without giving context is a common failure mode of current LLM summaries.)

I asked Claude to update the post to address your comment. There is now a section with the setup of the theorems, and clearer premises. Are they sufficiently understandable now?

Also, if I understand correctly, "behaviour maximizes a family of preference orderings..." means that the result only shows that we can represent an agent's behavior as satisfying completeness.

I only briefly skimmed the article, but I agree with your interpretation. Claude agrees too.

The completeness here is a property of a ranking reconstructed from choices, and it comes almost entirely from premise 0: because the rule always names something acceptable, every pair gets settled. [This is now clarified in the section of the linkpost with the setup.] But the rule may name both options as acceptable, which the theorem records as the two being equally good — and that is also exactly how an agent who found them incomparable, and picked arbitrarily, would behave. The result therefore cannot distinguish "equally good" from "not comparable", and so does not show that an agent deliberating about what to do must arrive at a complete ranking. It shows their behaviour is representable as if they had one.

Hi James.

This question gets to me a lot. It seems totally unclear how much the valence of an experience correlates to behavior. Like, it seems plausible that extraordinarily little pain could motivate behavior in simple organisms, but it's also hard to imagine, since, as a complex organism, I need pretty intense pain to be motivated to significantly change my behavior. And, it doesn't seem like there's any way we could reduce uncertainty about this.

Makes sense. Relatedly, you may be interested in the post "Do primitive sentient organisms feel extreme pain? disentangling intensity range and resolution" by Wladimir Alonso and Cynthia Schuck.

Also, in regards to Bob's book, it seems like it would make sense, rather than adding the model's estimates together to instead act on each model separately, i.e. assigning 5% of resources to help animals according to the equality model, 25% according to the neurophysiological model, and 70% according to the proxies model. Then, within the 70% for the proxies model, we could assign 33% of resources to applying a high exponent to the welfare ranges, 33% to applying no exponent to the welfare ranges, and 33% to applying a low exponent to the welfare ranges.

I think the above goes against maximising expected value in principle, even if it may be a good approach in practive, and I believe maximising expected value follows from very intuitive premises.

The problem of what percentages you should use and what models to use would, of course, still remain though.

Right.

I think consistently applying your arguments suggest you should assign 0 probability in these cases, and so your models blow up in complexity and you become too credulous, contrary to Occam's razor.

There is also Hitchens's razor.

What can be asserted without evidence can also be dismissed without evidence.

Readers of this post may be interested in this summary of the setup, premises, and conclusions of the theorems presented in Consequentialist Foundations for Expected Utility by Peter J. Hammond.

I just linkposted a summary of the setup, premises, and conclusions of the theorems presented in Consequentialist Foundations for Expected Utility by Peter J. Hammond.