LessWrong dev & admin as of July 5th, 2022.
RobertM
Karma: 923
Post-singularity worlds where people have the freedom to cause enormous animal suffering as a byproduct of legacy food production methods, despite having the option to not do so fully subsidized by third parties, seem like they probably overlap substantially with worlds where people have the freedom to spin up large quantities of digital entities capable of suffering and torture them forever. If you think such outcomes are likely, I claim that this is even more worthy of intervention. I personally don’t expect to have either option in most post-singularity worlds where we’re around, though I guess I would be slightly less surprised to have the option to torture animals than the option to torture ems (though I haven’t thought about it too hard yet).
But, as I said above, if you think it’s plausible that we’ll have the option to continue torturing animals post-singularity, this seems like a much more important outcome to try to avert than anything happening today.
I think you misunderstood my framing; I should have been more clear.
We can bracket the case where we all die to misaligned AI, since that leads to all animals dying as well.
If we achieve transformative AI and then don’t all die (because we solved alignment), then I don’t think the world will continue to have an “agricultural industry” in any meaningful sense (or, really, any other traditional industry; strong nanotech seems like it ought to let you solve for nearly everything else). Even if the economics and sociology work out such that some people will want to continue farming real animals instead of enjoying the much cheaper cultured meat of vastly superior quality, there will be approximately nobody interested in ensuring those animals are suffering, and the cost for ensuring that they don’t suffer will be trivial.
Of course, this assumes we solve alignment and also end up pointed in the right direction. For a variety of reasons it seems pretty unlikely to me that we manage to robustly solve alignment of superintelligent AIs while pointed in “wrong”[1] directions; that sort of philosophical unsophistication is why I’m pessimistic on our odds of success. But other people disagree, and if you think it’s at all plausible that we achieve TAI in a way that locks in reflectively-unendorsed values which lead to huge quantities of animal suffering, that seems like it ought to dominate effectively all other considerations in terms of interventions w.r.t. future animal welfare.
- ^
Like those that lead to enormous quantities of trivially preventable animal suffering for basically dumb contingent reasons, i.e. “the people doing the pointing weren’t really thinking about it at the time, and most people don’t actually care about animal suffering at all in most possible reflective equilibria”.
- ^
Is this post conditioning on AI hitting a wall ~tomorrow, for the next few decades? The analysis seems mostly reasonable, if so, but I think the interventions for ensuring that animal welfare is good after we hit transformative AI probably look very different from interventions in the pretty small slice of worlds where the world looks very boring in a few decades.
This is a bit of a sidenote, but while it’s true that “LWers” (on average) have a different threshold for how valuable criticism needs to be to justify its costs, it’s not true that “we” treat it as a deontic good. Observe, as evidence, the many hundreds of hours that various community members (including admins) have spent arguing with users like Said about whether their style of engagement and criticism was either effective at achieving its stated aims, or even worth the cost if it was[1]. “We” may have different thresholds, but “we” do not think that all criticism is necessarily good or worth the attentional cost.
The appropriate threshold is an empirical question whose answer will vary based on social context, people, optimization targets, etc.
Object-level, I probably agree that EA spends too much of its attention on bad criticism, but I also think it doesn’t allocate enough attention to good criticism, and this isn’t exactly the kind of thing that “nets out”. It’s more of a failure of taste/caring about the right things, which is hard to fix by adjusting the “quantity” dial.
- ^
He has even been subjected to moderation action more than once, so the earlier claim re: gadflies doesn’t stand up either.
- ^
No easily summarizable comment on the rest of it, but as a LessWrong dev I do think the addition of Quick Takes to the front page of LW was very good—my sense is that it’s counterfactually responsible for a pretty substantial amount of high quality discussion. (I haven’t done any checking of ground-truth metrics, this is just my gestalt impression as a user of the site.)
My claim is something closer to “experts in the field will correctly recognize them as obviously much smarter than +2 SD”, rather than “they have impressive credentials” (which is missing the critically important part where the person is actually much smarter than +2 SD).
I don’t think reputation has anything to do with titotal’s original claim and wasn’t trying to make any arguments in that direction.
Also… putting that aside, that is one bullet point from my list, and everyone else except Qiaochu has a wikipedia entry, which is not a criteria I was tracking when I wrote the list but think decisively refutes the claim that the list includes many people who are not publicly-legible intellectual powerhouses. (And, sure, I could list Dan Hendryks. I could probably come up with another twenty such names, even though I think they’d be worse at supporting the point I was trying to make.)
This still feels wrong to me: if they’re so smart, where are the nobel laureates? The famous physicists?
I think expecting nobel laureates is a bit much, especially given the demographics (these people are relatively young). But if you’re looking for people who are publicly-legible intellectual powerhouses, I think you can find a reasonable number:
Wei Dai
Hal Finney (RIP)
Scott Aaronson
Qiaochu Yuan[1]
Many historical MIRI researchers (random examples: Scott Garrabrant, Abram Demski, Jessica Taylor)
Paul Christiano (also formerly did research at MIRI)
(Many more not listed, including non-central examples like Robin Hanson, Vitalik Buterin, Shane Legg, and Yoshua Bengio[2].)
And, like, idk, man. 130 is pretty smart but not “famous for their public intellectual output” level smart. There are a bunch of STEM PhDs, a bunch of software engineers, some successful entrepreneurs, and about the number of “really very smart” people you’d expect in a community of this size.
- ^
He might disclaim any current affiliation, but for this purpose I think he obviously counts.
- ^
Who sure is working on AI x-risk and collaborating with much more central rats/EAs, but only came into it relatively recently, which is both evidence in favor of one of the core claims of the post but also evidence against what I read as the broader vibes.
first-hand accounts of people experiencing/overhearing racist exchanges
Sorry, I still can’t seem to find any of these, can you link me to such an account? I have seen one report that might be a second-hand account, though it could have been a non-racial slur.
(I’m generally not a fan of this much meta, but I consider the fact that this was strong downvoted by someone to be egregious. Most of the comment is reasonable speculation that turned out to be right, and the last sentence is a totally normal opinion to have, which might justify a disagree vote at worst.)
And I think this is related to a general skepticism I have about some of the most intense calls for the highest decoupling norms I sometimes see from some rationalists.
I think this is kind of funny because I (directionally) agree with a lot of your list, at least within the observed range of human cognitive ability, but think that strong decoupling norms are mostly agnostic to questions like trusting AI researchers who supported Lysenkoism when it was popular. Of course it’s informative that they did so, but can be substantially screened off by examining the quality of their current research (and, if you must, its relationship to whatever the dominant paradigms in the current field are).
People who’d prefer to not have them platformed at an event somewhat connected to EA don’t seem to think this is a trade off.
Optimizing for X means optimizing against not-X. (Well, at the pareto frontier, which we aren’t at, but it’s usually true for humans, anyways.) You will generate two different lists of people for two different values of X. Ergo, there is a trade off.
Anecdotally, a major reason I created this post was because the amount of very edgy people was significantly higher than the baseline for non-EA large events. I can’t think of another event that I have attended where people would’ve felt comfortable saying the stuff that was being said.
Note that these two sentences are saying very different things. The first one is about the percentage of attendees that have certain views, and I am pretty confident that it is false (except in a trivial sense, where people at non-EA events might have different “edgy” views). If you think that percentage of the general population that holds views at least as backwards as “typical racism” is less than whatever it was at Manifest (where I would bet very large amounts of money the median attendee was much more egalitarian than average for their reference class)...
The second one is about what was said at the event, and so far I haven’t seen anyone describe an explicit instance of racism or bigotry by an attendee (invited speaker or not). There were no sessions about “race science”, so I am left at something of a loss to explain how that is a subject that could continue to come up, unless someone happened to accidentally wander into multiple ongoing conversations about the subject. Absent affirmative confirmation of such an event, my current belief is that much more innocous things are being lumped in under a much more disparaging label.
Your comment seems to be pretty straightforwardly advocating for optimizing for very traditional political considerations (appearance of respectability, relationships with particular interest groups, etc) by very traditional political means (disassociating with unfavorables). The more central this is to how “EA” operates, the more fair it is to call it a political project.
I agree that many rationalists have been alienated by wokeness/etc. I disagree that much of what’s being discussed today is well-explained by a reactionary leaning-in to edginess, and think that the explanation offered—that various people were invited on the basis of their engagement with concepts central to Manifest, or for specific panels not related to their less popular views—is sufficient to explain their presence.
With that said, I think Austin is not enormously representative of the rationalist community, and it’s pretty off-target to chalk this up as an epistemic win for the EA cultural scene over the rationalist cultural scene. Observe that it is here, on the EA forum, that a substantial fraction of commenters are calling for conference organizers to avoid inviting people for reasons that explicitly trade off against truth-seeking considerations. Notably, there are people who I wouldn’t have invited, if I were running this kind of event, specifically because I think they either have very bad epistemics or are habitual liars, such that it would be an epistemic disservice to other attendees to give those people any additional prominence.
I think that if relevant swathes of the population avoid engaging with e.g. prediction markets on the basis of the people invited to Manifest, this will be substantially an own-goal, where people with 2nd-order concerns (such as anticipated reputational risk) signal boost this and cause the very problem they’re worried about. (This is a contingent, empirical prediction, though unfortunately one that’s hard to test.) Separately, if someone avoided attending Manifest because they anticipated unpleasantness stemming from the presence of these attendees, they either had wildly miscalibrated expectations about what Manifest would be like, or (frankly) they might benefit from asking themselves what is different about attending Manifest vs. attending any other similarly large social event (nearly all of which have invited people with similarly unpalatable views), and whether they endorse letting the mere physical presence of people they could choose to pretend don’t exist stop them from going.
Huh, it’s a bit surprising to me that people disagree so strongly with this comment, which seems to be (uncharitably but not totally inaccurately) paraphrasing the parent, which has much more agreement.
(Maybe most people are taking it literally, rather than interpreting it as a snipe?)
Perhaps it’s missing from the summary, but there is trivially a much stronger argument that doesn’t seem addressed here.
Humans must be pretty close to the stupidest possible things that could design things smarter than them.
This is especially true when it comes to the domain of scientific R&D, where we only have even our minimal level of capabilities because it turns out that intelligence generalizes from e.g. basic tool-use and social modeling to other things.
We know that we can pretty reliably create systems that are superhuman in various domains when we figure out a proper training regime for those domains. e.g. AlphaZero is vastly superhuman in chess/go/etc, GPT-3 is superhuman at next token prediction (to say nothing of GPT-4 or subsequent systems), etc.
The nature of intelligent search processes is to route around bottlenecks. The argument re: bottlenecks proves too much, and doesn’t even seem to stand up historically. Why did bottlenecks not fail to stymie superhuman capabilities in the domains where we’re achieved them?
Humanity, today, could[1] embark on a moderately expensive project to enable wide-scale genomic selection for intelligence, which within a single generation would probably produce a substantial number of humans smarter than any who’ve ever lived. Humans are not exactly advantaged in their ability to iterate here, compared to AI.
The general shape of Thorstad’s argument doesn’t really make it clear what sort of counterargument he would admit as valid. Like, yes, humans have not (yet) kicked off any process of obvious, rapid, recusive self-improvement. That is indeed evidence that it might take humans a few decades after they invent computing technology to do so. What evidence, short of us stumbling into the situation under discussion, would be convincing?
- ^
(Social and political bottlenecks do exist, but the technology is pretty straightforward.)
I’ve spent some time thinking about the same question and I’m glad that there’s some multiple discovery; the AI Control agenda seems relevant here.
I think that neither of those are selective uses of analogies. They do point to similarities between things we have access to and future ASI that you might not think are valid similarities, but that is one thing that makes analogies useful—they can make locating disagreements in people’s models very fast, since they’re structurally meant to transmit information in a highly compressed fashion.
There is no button you can press on demand to publish an article in either a peer-reviewed journal or a mainstream media outlet.
Publishing pieces in the media (with minimal 3rd-party editing) is at least tractable on the scale of weeks, if you have a friendly journalist. The academic game is one to two orders of magnitude slower than that. If you want to communicate your views in real-time, you need to stick to platforms which allow that.
I do think media comms is a complementary strategy to direct comms (which MIRI has been using, to some degree). But it’s difficult to escape the fact that information posted on LW, the EA forum, or Twitter (by certain accounts) makes its way down the grapevine to relevant decision-makers surprisingly often, given how little overhead is involved.
ETA: feel free to ignore the below, given your caveat, though you may find it helpful if you choose to write an expanded form of any of the arguments later to have some early objections.
Correct me if I’m wrong, but it seems like most of these reasons boil down to not expecting AI to be superhuman in any relevant sense (since if it is, effectively all of them break down as reasons for optimism)? To wit:
Resource allocation is relatively equal (and relatively free of violence) among humans because even humans that don’t very much value the well-being of others don’t have the power to actually expropriate everyone else’s resources by force. (We have evidence of what happens when those conditions break down to any meaningful degree; it isn’t super pretty.)
I do not think GPT-4 is meaningful evidence about the difficulty of value alignment. In particular, the claim that “GPT-4 seems to be honest, kind, and helpful after relatively little effort” seems to be treating GPT-4′s behavior as meaningfully reflecting its internal preferences or motivations, which I think is “not even wrong”. I think it’s extremely unlikely that GPT-4 has preferences over world states in a way that most humans would consider meaningful, and in the very unlikely event that it does, those preferences almost certainly aren’t centrally pointed at being honest, kind, and helpful.
re: endogenous reponse to AI—I don’t see how this is relevant once you have ASI. To the extent that it might be relevant, it’s basically conceding the argument: that the reason we’ll be safe is that we’ll manage to avoid killing ourselves by moving too quickly. (Note that we are currently moving at pretty close to max speed, so this is a prediction that the future will be different from the past. One that some people are actively optimising for, but also one that other people are optimizing against.)
re: perfectionism—I would not be surprised if many current humans, given superhuman intelligence and power, created a pretty terrible future. Current power differentials do not meaningfully let individual players flip every single other player the bird at the same time. Assuming that this will continue to be true is again assuming the conclusion (that AI will not be superhuman in any relevant sense). I also feel like there’s an implicit argument here about how value isn’t fragile that I disagree with, but I might be reading into it.
I’m not totally sure what analogy you’re trying to rebut, but I think that human treatment of animal species, as a piece of evidence for how we might be treated by future AI systems that are analogously more powerful than we are, is extremely negative, not positive. Human efforts to preserve animal species are a drop in the bucket compared to the casual disregard with which we optimize over them and their environments for our benefit. I’m sure animals sometimes attempt to defend their territory against human encroachment. Has the human response to this been to shrug and back off? Of course, there are some humans who do care about animals having fulfilled lives by their own values. But even most of those humans do not spend their lives tirelessly optimizing for their best understanding of the values of animals.
I think the modal no-Anthropic counterfactual does not have an alignment-agnostic AI company that’s remotely competitive with OpenAI, which means there’s no external target for this Amazon investment. It’s not an accident that Anthropic was founded by former OpenAI staff who were substantially responsible for OpenAI’s earlier GPT scaling successes.
Fraud is bad.
In any case, people already don’t have enough worthwhile targets for donating money to, even under short timelines, so it’s not clear what good taking out loans would do. If it’s a question of putting one’s money where one’s mouth is, I personally took a 6-figure paycut in 2022 to work on reducing AI x-risk, and also increased my consumption/spending.