MichaelStJules

7699Joined May 2016

Comments
1827

Topic Contributions
12

It seems like you're overselling Shapley values here, then, unless I've misunderstood. They won't help to decide which interventions to fund, except for indirect reasons (e.g. assigning credit and funding ex post, judging track record).

You wrote "Then we walk away saying (hyperopically,) we saved a life for $5,000, ignoring every other part of the complex system enabling our donation to be effective. And that is not to say it's not an effective use of money! In fact, it's incredibly effective, even in Shapley-value terms. But we're over-allocating credit to ourselves."

But if $5000 per life saved is the wrong number to use to compare interventions, Shapley values won't help (for the right reasons, anyway). The solution here is to just model counterfactuals better. If you're maximizing the sum of Shapley values, you're acknowledging we have to model counterfactuals better anyway, and the sum is just expected utility, so you don’t need the Shapley values in the first place. Either Shapley value cost-effectiveness is the same as the usual cost-effectiveness (my interpretation 1) and redundant, or it's a predictably suboptimal theoretical target (e.g. maximizing your own Shapley value only, as in Nuno's proposal, or as another option, my interpretation 2, which requires unrealistic counterfactual assumptions).

The solution to the non-EA money problem is also to just model counterfactuals better. For example, Charity Entrepreneurship has used estimates of the counterfactual cost-effectiveness of non-EA money raised by their incubated charities if the incubated charity doesn't raise it.

Do you mean maximizing the sum of Shapley values or just your own Shapley value? I had the latter in mind. I might be mistaken about the specific perverse examples even under that interpretation, since I'm not sure how Shapley values are meant to be used. Maximizing your own Shapley value seems to bring in a bunch of counterfactuals (i.e. your counterfactual contribution to each possible coalition) and weigh them ignoring propensities to cooperate/coordinate, though.

On the other hand, the sum of Shapley values is just the value (your utility?) of the "grand" coalition, i.e. everyone together. If you're just maximizing this, you don’t need to calculate any of the Shapley values first (and in general, you need to calculate the value of the grand coalition for each Shapley value). I think the point of Shapley values would just be for assigning credit (and anything downstream of that), not deciding on the acts for which credit will need to be distributed.

If you're maximizing this sum, what are the options you're maximizing over?

  1. On one interpretation, if you're maximizing this only over your own actions and their consequences, including on others' responses (and possibly acausal influence), it's just maximizing expected utility.

  2. On another interpretation, if you're maximizing it over everyone's actions or assuming everyone else is maximizing it (and so probably that everyone is sufficiently aligned), then that would be ideal (game-theoretically optimal?), but such an assumption is often unrealistic and making it can lead to worse outcomes. For example, our contributions to a charity primarily supported (funding, volunteering, work) by non-EAs with little room for more support might displace that non-EA support towards far less cost-effective uses or even harmful uses. And in that case, it can be better to look elsewhere more neglected by non-EAs. The assumption may be fine in some scenarios (e.g. enough alignment and competence), but it can also be accommodated in 1.

So, I think the ideal target is just doing 1 carefully, including accounting for your influence over other agents and possibly acausal influence in particular.

Are you suggesting we maximize Shapley values instead of expected utility with counterfactuals? That's going to violate standard rationality axioms, and so (in idealized scenarios with "correctly identified" counterfactual distributions) is likely to lead to worse outcomes overall. It could recommend, both in practice and in theory, doing fully redundant work for just for the credit and no extra value. In the paramedic example, depending on how the numbers work out, couldn't it recommend perversely pushing the paramedics out of the way to do CPR yourself and injuring the individual's spine, even knowing ahead of time you will injure them?  There's a coalition - just you - where your marginal/counterfactual impact is to save the person's life, and giving that positive weight could make up for the injury you actually cause.

I think we should only think of Shapley values as a way to assign credit. The ideal is still to maximize expected utility (or avoid stochastic dominance or whatever). Maybe we need to model counterfactuals with other agents better, but I don't find the examples you gave very persuasive.

The most important thing about your decision theory is that it shouldn't predictably and in expectation leave you worse off than if you had used a different approach. My claim in the post is that we're using such an approach, and it leaves us predictably worse off in certain specific cases.

 

This isn't a problem with expected utility maximization (with counterfactuals), though, right? I think the use of counterfactuals is theoretically sound, but we may be incorrectly modelling counterfactuals.

This is a technically accurate definition, but I still had trouble intuiting this as equivalent to a daily experience of disabling physical pain equivalent to having your leg sliced open with a hot, sharp live wire.

Nest deprivation could be in the bottom half of the disabling pain intensity range. Ren put their tattoo experiences described as "**** me, make it stop. Like someone slicing into my leg with a hot, sharp live wire." near the high end of disabling. Also, the latter just sounds excruciating to me personally, not merely disabling, but we discussed that here.

 

Besides the evidence you mention, they also mention vocalizations (gakel-calls), which seem generally indicative of frustration across contexts (dustbathing deprivation, food/water deprivation, nesting deprivation), and hens made more of them when nest deprived than when deprived of food, water or dustbathing in Zimmerman et al., 2000, although in that study, the authors discuss the possibility that nest deprivation gakel-calls are more specific and not necessarily indicative of frustration:

In the period Frustration, the number of gakel-calls was higher in treatment Nest than in the other treatments. This might mean that in this treatment the level of frustration was higher. However, this is not supported by higher levels of other behaviours indicative of frustration in treatment Nest compared to the other treatments. An alternative explanation for the higher number of gakel-calls in treatment Nest is suggested by the occurrence of the gakel-call under natural circumstances. The gakel-call is given before oviposition and probably has evolved as a signal towards the rooster McBride et al., 1969; Thornhill, 1988 . According to Meijsser and Hughes 1989 , the performance of the gakel-call is related to finding a suitable nest site, also under husbandry conditions. Another explanation is offered by the motivational model proposed by Wiepkema 1987 . It implies that the gakel-call under these circumstances is an emotional expression of the detection of a prolonged mismatch between actual ‘‘no nest site found’’ state and desired state ‘‘find a suitable nest site’’ and is an indication of frustration. Both oviposition and the detection of a prolonged mismatch could at the same time contribute to the occurrence of gakel-calls. The surplus of gakel-calls in treatment Nest compared to the other treatments might be the gakel-calls specifically related to oviposition.

This latter finding might account for the difference in temporal characteristics of gakel-calls between treatment Nest and the treatments Water and Dust. Gakel-calls in treatment Nest lasted longer and consisted of more notes than in the treatments Water and Dust. Schenk et al. 1983 found that the mean duration of a single gakel-call was longer when dustbathing was thwarted stronger by longer deprivation. However, from the present study, nothing decisive can be concluded about the relation between the number of gakel-calls and their temporal characteristics on the one hand, and the intensity of thwarting in the different treatments on the other

I agree with you that that is definitely conceivable. But I think that, as Carl argued in his post (and elaborated on further in the comment thread with gwern), our default assumption should be that efficiency (and probably also intensity) of pleasure vs pain is symmetric.

I think identical distributions for efficiency is a reasonable ignorance prior, ignoring direct intuitions and evidence one way or the other, but we aren't so ignorant that we can't make any claims one way or the other. The kinds of claims Shulman made are only meant to defeat specific kinds of arguments for negative skew over symmetry, like direct intuition, not to argue for positive skew. Given the possibility that direct intuition in this case could still be useful (and indeed skews towards negative being more efficient, which seems likely), contra Shulman, then without arguments for positive skew (that don't apply equally in favour of negative skew), we should indeed expect the negative to be more efficient.

Furthermore, based on the arguments other than direct intuition I made above, and, as far as I know, no arguments for pleasure being more efficient than pain that don't apply equally in reverse, we have more reason to believe efficiencies should skew negative.

Also similar to gwern's comment, if positive value on non-hedonistic views does depend on things like reliable perception of the outside world or interaction with other conscious beings (e.g. compared to the experience machine or just disembodied pleasure) but bads don't (e.g. suffering won't really be any less bad in an experience machine or if disembodied), then I'd expect negative value to be more efficient than positive value, possibly far more efficient, because perception and interaction require overhead and may slow down experiences.

However, similar efficiency for positive value could still be likely enough that the expected efficiencies are still similar enough and other considerations like their frequency dominate.

Ah, welfare range estimates may already be supposed to capture the probability that an animal can experience intense suffering, like excruciating pain.

Thanks for writing this!

You might be able to make some informed guesses or do some informative sensitivity analysis about net welfare in wild animals, given your pain intensity ratios. I think it's reasonable to assume that animals don't experience any goods as intensely good (as valuable per moment) as excruciating pain is intensely bad. Pleasures as intense as disabling pain may also be rare, but that could be an assumption to vary.

Based on your ratios and total utilitarian assumption, 1 second of excruciating pain outweighs 11.5 days of annoying pain or 1.15 days of hurtful pain, or 11.5 days of goods as intense as annoying pain or 1.15 days of goods as intense as hurtful pain, on average.

Just quickly Googling for the most populous groups I'm aware of, mites, springtails and nematodes live a few weeks at most and copepods up to around a year. There might be other similarly populous groups of aquatic arthropods I'm missing that you should include, but I think mites and springtails capture terrestrial arthropods by moral weight. I think those animals will dominate your calculations, the way you're doing them. And their deaths could involve intense pain and perhaps only a very small share live more than a week. However, it's not obvious these animals can experience very intense suffering at all, even conditional on their sentience, but this probability could be another sensitivity analysis parameter.

(FWIW, I'd be inclined to exclude nematodes, though. Including them feels like a mugging to me and possibly dominated by panpsychism.)

Ants may live up to a few years and are very populous, and I could imagine have relatively good lives on symmetric ethical views, as eusocial insects investing heavily in their young. But they're orders of magnitude less populous than mites and springtails.

Although this group seems likely to be outweighed in expectation, for wild vertebrates (or at least birds and mammals?), sepsis seems to be one of the worst natural ways to die, with 2 hours of excruciating pain and further time at lower intensities in farmed chickens (https://welfarefootprint.org/research-projects/cumulative-pain-and-wild-animal-welfare-assessments/ ). With your ratios, this is the equivalent of more than 200 years of annoying pain or 20 years of hurtful pain, much longer than the vast majority of wild vertebrates (by population and peehaps species) live. I don't know how common sepsis is, though. Finding out how common sepsis is in the most populous groups of vertebrates could have high value of information for wild vertebrate welfare.

I'm not aware of high status individuals in the community justifying prioritizing humans on the mere basis of species membership. Usually I see claims about differences in capacities and interests. Are there public examples you can share?

Thanks for sharing your experiences. There's also an article here with some useful info on and others' experiences with inadequate pain relief for childbirth in the UK: https://www.vice.com/en/article/8x7mm4/childbirth-pain-relief-denied

Load More