I basically agree with this critique of the results in the post, but want to add that I nonetheless think this is a very cool piece of research and I am excited to see more exploration along these lines!
One idea that I hadâmaybe someone has done something like this? -- is to ask people to watch a film or read a novel and rate the life satisfaction of the characters in the story. For instance, they might be asked to answer a question like âHow much does Jane Eyre feel satisfied by her life, on a scale of 1-10?â. (Note that we arenât asking how much the respondent empathizes with Jane or would enjoy being her, simply how much satisfaction they believe Jane gets from Janeâs life.) This might allow us to get a shared baseline for comparison. If peopleâs assessments of Janeâs life go up or down over time, (or differ between people) it seems unlikely that this is a result of a violation of âprediction invarianceâ, since Jane Eyre is an unchanging novel with fixed facts about how Jane feels. Instead, it seems like this would indicate a change in measurement: i.e. how people assign numerical scores to particular welfare states.
Zachary Brownđ¸
I think rescaling could make it steeper or flatter, depending on the particular rescaling. Consider that there is nothing that requires the rescaling to be a linear transformation of the original scale (like youâve written in your example). A rescaling that compresses the life satisfaction scores that were initially 0-5 into the range 0-3, while leaving the life satisfaction score of 8-10 unaffected will have a different effect on the slope than if we disproportionately compress the top end of life satisfaction scores.
Sorry if I expressed this poorlyâitâs quite late :)
To synthesize a few of the comments on this postâThis comment sounds like a general instance of the issue that @geoffrey points out in another comment: what @Charlie Harrison is describing as a violation of âprediction invarianceâ may just be a violation of âmeasurement invarianceâ; in particular because happiness (the real thing, not the measure) may have a different relationship with GMEOH events over time.
Thanks for the write-up! As a donor to the fund, itâs really nice to see these reports. I occasionally wonder if I could obtain more cost-effective results by donating independently without the overhead of a managed fund. These reports reassure me that I almost certainly could not. Really grateful to the team for finding these incredible funding opportunities.
This is a really great post!! I really appreciated the point industry consolidation point. I also appreciate how you describe advocacy for PLF as a âframingâ loss, since it implicitly concedes that we will be factory farming. This framing loss is an issue with a lot of welfarist interventions, and I donât think means we need to rule such interventions out, but I think it does make these sorts of interventions less attractive for public-facing campaigns. I think people sometimes underestimate the badness of framing loss, and I think this post makes the point really sharply; thanks.
I wrote a similar post arguing that animal advocates should oppose PLF, available here. A few ideas from that piece that I think are complementary to this one:
I think there are some narrative reasons why opposing the worst instances of PLF might make attractive campaign targets: the industry is still underdeveloped, automated farming is disturbing to the public, small farmers might be willing to support these campaigns (because of the concentration effects of PLF), and there are existing ties between animal advocates and AI firms (through EA). Some of these arguments are stronger than others of course.
I think PLF is likely to disproportionately increase the efficiency of farming small animals. This is because it allows farmers to deploy individual level monitoring where it was previously infeasible (because the labor costs of monitoring individual animals on e.g. a chicken farm with tens of thousands of animals is too high). This is another reason why the total number of animals farmed is likely to increase as a result of increased PLF adoption.
Another article that people might be interested in is this one, which proposes specific ethical restrictions/âguidelines for PLF.
Thanks for the comment. I was clearly too quick with that opening statement. Perhaps in part I let my epistemic guard down there out of general frustration at the neglectedness of the topic, and a desire to attract some attention with a bold opener. So much harm could accrue to nonhuman animals relative to humans, and I really want more discussion on this. PLF isâIâve argued, anywayâa highly visible threat to the welfare of zillions, but rarely mentioned. I hope youâll forgive an immodest but emotional claim.
Iâve edited the opener and the footnote to be more defensible, in response to this comment.
I actually donât believe, in the median scenario, that AIs are likely to both outnumber sentient animals and have a high likelihood of suffering, but I donât really want that to be the focus of this piece. And either way, I donât believe that with high certainty: in that respect, the statement was not reflective of my views.
Some of this discussion reminds me of Millâs in his super underrated essay âUtility of Religionâ. He proposes there a kind of yangy humanistic religion, against a backdrop of atheism and concern about the evils of nature. Worth a read.
Thanks for the comment!
I agree that thereâs a mixed case for political tractability. Iâm curious why you donât find the argument compelling about the particular people who have influence on AI policy being more amenable to animal-related concerns? (To put it bluntly, EAs care about animals and are influential in AI, and animal ag industry lobbying hasnât really touched this issue yet.)
I like the analogy to cage-free campaigns, although I think I would draw different lessons from the analogy. I donât really think that the support for cage-free campaigns comes from support for restrictions that help individual animals rather than support for restrictions that restrict the total number of farmed animals. Instead, I think it comes for support for traditional and ânaturalâ ways of farming (where the chickens are imagined to roam free) rather than industrialised, modern, and intensive farming methods. On this view, cage-free campaigns succeed because they target only the farming methods that the public disapproves of. This theory can also explain why people express disapproval of factory farming, but a strong approval of farming and farmers.
I think PLF is a politically tractable target for regulation because, like cage-free campaigns, it targets only the type of farming people already dislike. When I say âEnd AI-run factory farms!â, the slogan makes inherently salient the technological, non-natural, industrial nature of the farming method. Restrictions here might not be perceived as restrictions on farming, theyâll be perceived only as restrictions on a certain sinister form of unnatural industrialised farming. (The general public mostly doesnât realise that most farming is industrialised.) To put this another way: I think the most politically tractable pro-animal movements are the ones that explicitly restrict their focus to Big Evil Factory Farms, and leave Friendly Farmer Joe alone. I think PLF restrictions share this character with cage-free campaigns.
And we know from cage-free campaigns that people are sometimes willing to tolerate restrictions of this sort even if they are personally costly.
ďAnÂiÂmal adÂvoÂcates should camÂpaign to reÂstrict AI preÂciÂsion liveÂstock farming
I basically fail to imagine a scenario where publishing the Trust Agreement is very costly to Anthropicâespecially just sharing certain details (like sharing percentages rather than saying âa supermajorityâ)âexcept that the details are weak and would make Anthropic look bad.
Anthropic might be worried that the details are strong, and would make Anthropic look vulnerable to similar governance chaos to what happened at OpenAI during the board turnover saga. A large public conversation on this could be bad for Anthropicâs reputation among its investors, team, or other stakeholders, who have concerns other than longterm safety, or might think that Anthropicâs non profit-motivated governance is opaque or bad for whatever other reason. To put this another way: Anthropic is probably reputation-managing, but it might not be their safety reputation that they are trying to manage. It might be their reputationâto potential investors, sayâas a reliable actor with predictable decision-making that wonât be upturned at the whims of the trust.
I would expect, though, that Anthropicâs major investors know the details of the governance structure and mechanics.
Iâm in the early stages of corporate campaign work similar to whatâs discussed in this post. Iâm trying to mobilise investor pressure to advocate for safety practices at AI labs and chipmakers. Iâd love to meet with others working on similar projects (or anyone interested in funding this work!). Iâd be eager for feedback.
You can see a write-up of the project here.
Frankenstein (Mary Shelley): moral circle expansion to a human created AI, kinda.
Elizabeth Costello (J M Coetzee): novel about a professor who gives animal rights lectures. The chapter thatâs most profoundly about animal ethics was published as âThe Lives of Animalsâ which was printed with commentary from Peter Singer (in narrative form!).
Darkness at Noon (Arthur Koestler): Novel with reflections from an imprisoned old Bolshevik, reflecting on his past revolutionary activity. Interesting reflections on ends vs. means reasoning, and on weighing considerations of moral scale /â the numbers affected vs personal emotional connection in moral tradeoff scenarios.
ď[Linkpost] Eric SchÂwitzgebel: AI sysÂtems must not conÂfuse users about their senÂtience or moral status
Thanks for putting this together! Super helpful.
I really appreciated this post and itâs sequel (and await the third in the sequence)! The âsecond mistakeâ was totally new to me, and I hadnât grasped the significance of the âfirst mistakeâ. The post did persuade me that the case for existential risk reduction is less robust than I had previously thought.
One tiny thing. I think this should read âfrom 20% to 10% riskâ:
More rarely, we talk about absolute reductions, which subtract an absolute amount from the current level of risk. It is in this sense that a 10% reduction in risk takes us from 80% to 70% risk, from 20% to 18% risk, or from 10% to 0% risk. (Formally, relative risk reduction by f takes us from risk r to risk r â f).
Thanks for writing this! Hoping to respond more fully later.
In the meantime: I really like the example of what a ânear-term AI-Governance factor collection could look likeâ.
So the question is âwhat governance hurdles decrease risk but donât constitute a total barrier to entry?â
I agree. There are probably some kinds of democratic checks that honest UHNW individuals donât mind, but have relatively big improvements for epistemics and community risk. Perhaps there are ways to add incentives for agreeing to audits or democratic checks? It seems like SBFâs reputation as a businessman benefited somewhat from his association with EA (I am not too confident in this claim). Perhaps offering some kind of âSuper Effective Philanthropistâ title/âprize/âtrophy to particular UHNW donors that agree to subject their donations to democratic checks or financial audits might be an incentive? (Iâm pretty skeptical, but unsure.) Iâd like to do some more creative thinking here.
I wonder if submitting capital to your proposal seems a bit too much like the latter.
Probably.
I think this is a great post, efficiently summarizing some of the most important takeaways from recent events.
I think this claim is especially important:
âItâs also vital to avoid a very small number of decision-makers having too much influence (even if they donât want that level of influence in the first place). If we have more sources of funding and more decision-makers, it is likely to improve the overall quality of funding decisions and, critically, reduce the consequences for grantees if they are rejected by just one or two major funders.â
Hereâs a sketchy idea in that vein for further consideration. One additional way to avoid extremely wealthy donors having too much influence is to try to insist that UHNW donors subject their giving to democratic checks on their decision-making from other EAs. For instance, what if taking a Giving What We Can pledge entitled you to a vote of some kind on certain fund disbursements or other decisions? What if Giving What We Can pledgers could put forward âshareholder proposalsâ on strategic decisions (subject to getting fifty signatures, say) at EA orgs, which other pledgers could then vote on? (Not necessarily just at GWWC) Obviously there are issues:
voters may not be the epistemic peers of grantmaking experts /â EA organization employees
voters may not be the epistemic peers of the UHNW donors themselves who have more reputational stake in ensuring their donations go well
UHNW donors have a lot of bargaining power when dealing with EA institutions and few incentives to open themselves up to democratic checks on their decision-making
determining who gets to vote is hard
some decisions need to be made quickly
sometimes there are infohazards
But there are advantages too, and I expect that often they outweigh the disadvantages:
wisdom of crowds
diversified incentives
democracy is a great look
This comment seems to be generating substantial disagreement. Iâd be curious to hear from those who disagree: which parts of this comment do you disagree with, and why?
Oh, great, thanks so much! Iâll check this out.