Posts
Comments
How about Melanie Mitchell’s Artificial Intelligence: A Guide for Thinking Humans?
Oh totally (and you probably know much more about this than me). I guess the key thing I'm challenging is the idea that there was something like a very fast transfer of power resulting just from upgraded computing power moving from chimp-ancestor brain -> human brain (a natural FOOM), which the discussion sometimes suggests. My understanding is that it's more like the new adaptations allowed for cumulative cultural change, which allowed for more power.
Psychology/anthropology:
The misleading human-chimp analogy: AI will stand in relation to us the same way we stand in relation to chimps. I think this analogy basically ignores how humans have actually developed knowledge and power--not by rapid individual brain changes, but by slow, cumulative cultural changes. In turn, the analogy may lead us to make incorrect predictions about AI scenarios.
In addition to (farmed and wild) animal organizations, OPIS is worth checking out.
Here's a list of organizations focusing on the quality of the long-term future (including the level of suffering), from this post:
If you are persuaded by the arguments that the expected value of human expansion is not highly positive or that we should prioritize the quality of the long-term future, promising approaches include research, field-building, and community-building, such as at the Center on Long-Term Risk, Center for Reducing Suffering, Future of Humanity Institute, Global Catastrophic Risk Institute, Legal Priorities Project, and Open Philanthropy, and Sentience Institute, as well as working at other AI safety and EA organizations with an eye towards ensuring that, if we survive, the universe is better for it. Some of this work has substantial room for more funding, and related jobs can be found at these organizations’ websites and on the 80,000 Hours job board.
I found this to be a comprehensive critique of some of the EA community's theoretical tendencies (over-reliance on formalisms, false precision, and excessive faith in aggregation). +1 to Michael Townsend's suggestions, especially adding a TLDR to this post.
Longtermism + EA might include organizations primarily focused on the quality of the long-term future rather than its existence and scope (e.g., CLR, CRS, Sentience Institute), although the notion of existential risk construed broadly is a bit murky and potentially includes these (depending on how much of the reduction in quality threatens “humanity’s potential”)
Cool diagram! I would suggest rephrasing the Longtermism description to say “We should focus directly on future generations.” As it is, it implies that people only work on animal welfare and global poverty because of moral positions, rather than concerns about tractability, etc.
Glad to have you here :D
I'm just going to plug some recommendations for suffering-focused stuff: You can connect with other negative utilitarians and suffering-focused people in this Facebook group, check out this career advice, and explore issues in ethics and cause prioritization here.
Julia Wise (who commented earlier) runs the EA Peer Support Facebook group, which could be good to join, and there are many other EA and negative utilitarian/suffering-focused community groups. Feel free to PM me!
Also spent hens are almost always sold to be slaughtered, where many are probably exposed to torture-level suffering. I remember looking into this a while back and only found one pasture farm where spent hens were not sold for slaughter. You can find details for many farms here: https://www.cornucopia.org/scorecard/eggs/
I think considerations like these are important to challenge the recent emphasis on grounding x-risk (really, extinction risk) in near-term rather than long-term concerns. That perspective seems to assume that the EV of human expansion is pretty much settled, so we don’t have to engage too deeply with more fundamental issues in prioritization, and we can instead just focus on marketing.
I’d like to see more written directly comparing the tractability and neglectedness of population risk reduction and quality risk reduction. I wonder if you’ve perhaps overstated things in claiming that a lower EV for human expansion suggests shifting resources to long-term quality risks rather than, say, factory farming. It seems like this claim requires a more detailed comparison between possible interventions.
Cool! Look forward to learning more about your work. Fyi your Discord invite link is broken.
On the other hand, this would exclude people whose main issue with longtermism is epistemic in nature. But maybe it’s too hard to come up with an acceptable catch-all term.
CRS has written about career advice for s-risks.
CLR does some coaching.
HLI seems to be working on career recommendations.
I might be misunderstanding but I don’t think the intuition you mentioned is really an argument for hedonism, since one can agree that there must be beings with conscious experiences for anything to matter without concluding that conscious experience itself is the only thing that matters.
I think this analysis should make more transparent its reliance on something like total utilitarianism and the presence of a symmetry between happiness and suffering. In the absence of these assumptions, the instances of extreme suffering and exploitation in factory farming more clearly entail "approximate veg*nism."
Consider the fate of a broiler chicken being boiled alive. Many people think that such extreme suffering cannot be counterbalanced by other positive aspects of one's own life, so there is really no way to make the chicken's life "net positive." Moreover, many moral positions deny that the extreme suffering of one individual can be counterbalanced by positive aspects of others' lives. So even if most farmed animals did have lives better than non-existence, we might still find the whole practice objectionable because of how it affects the worst off. Looking beyond just pain and pleasure, many perspectives object to our instrumentalization and exploitation of other sentient beings, which is inherent to the practice of animal farming. And there are also those who subscribe to some sort of asymmetry in population ethics, in which case the creation of negative lives is more bad than the creation of new lives is good (which weighs strongly against a practice like factory farming, even supposing that the average life is net positive).
I'm not saying that these are all correct, but rather that we should do a better job of clarifying our background assumptions rather than just saying "EAs should..."
I think your discussion of concentration camps in the comments further highlights the need to look beyond one particular moral perspective. Even if you are right that most lives in concentration camps were better than non-existence, many people would probably find objectionable the idea of "sentience-maximizing concentration camps," i.e. supporting the creation of new lives in concentration camps while simultaneously working to make them slowly better, rather than altogether banning the practice (supposing that these are the only two options). Again, this could be motivated by the sort of positions I mentioned above.
On YouTube (from less to more detailed): Rob Miles, Two Minute Papers, Yannic Kilcher
Longtermism and animal advocacy are often presented as mutually exclusive focus areas. This is strange, as they are defined along different dimensions: longtermism is defined by the temporal scope of effects, while animal advocacy is defined by whose interests we focus on. Of course, one could argue that animal interests are negligible once we consider the very long-term future, but my main issue is that this argument is rarely made explicit.
This post does a great job of emphasizing ways in which animal advocacy should inform our efforts to improve the very long-term future, and ways in which a focus on the very long-term future should inform animal advocacy.
This is a key reading for anyone who wants to think more broadly about longtermism. We used this post as part of a fellowship at UCLA focused on effective animal advocacy, and our participants found it very thought-provoking.
I come back to this post quite frequently when considering whether to prioritize MCE (via animal advocacy) or AI safety. It seems that these two cause areas often attract quite different people with quite different objectives, so this post is unique in its attempt to compare the two based on the same long-term considerations.
I especially like the discussion of bias. Although some might find the whole discussion a bit ad hominem, I think people in EA should take seriously the worry that certain features common in the EA community (e.g., an attraction towards abstract puzzles) might bias us towards particular cause areas.
I recommend this post for anyone interested in thinking more broadly about longtermism.
Yeah I have been in touch with them. Thanks!
Yeah I’m not totally sure what it implies. For consequentialists, we could say that bringing the life into existence is itself morally neutral; but once the life exists, we have reason to end it (since the life is bad for that person, although we’d have to make further sense of that claim). Deontologists could just say that there is a constraint against bringing into existence tortured lives, but this isn’t because of the life’s contribution to some “total goodness” of the world. Presumably we’d want some further explanation for why this constraint should exist. Maybe such an action involves an impermissible attitude of callous disregard for life or something like that. It seems like there are many parameters we could vary but that might seem too ad hoc.
I mostly meant to say that someone who otherwise rejects totalism would agree to (*), so as to emphasize that these diverse values are really tied to our position on the value of good lives (whether good = virtuous or pleasurable or whatever).
Similarly, I think the transitivity issue has less to do with our theory of wellbeing (what counts as a good life) and more to do with our theory of population ethics. As to how we can resolve this apparent issue, there are several things we could say. We could (as I think Larry Temkin and others have done) agree with (b), maintaining that 'better than' or 'more valuable than' is not a transitive relation. Alternatively, we could adopt a sort of "tethered good approach" (following Christine Korsgaard), where we maintain that claims like "A is better/more valuable than B" are only meaningful insofar as they are reducible to claims like "A is better/more valuable than B for person P." In that case, we might deny that "a meh life is just as valuable as [or more/less valuable than] nonexistence " is meaningful, since there's no one for whom it is more valuable (assuming we reject comparativism, the view that things can be better or worse for merely possible persons). Michael St. Jules is probably aware of better ways this could be resolved. In general, I think that a lot of this stuff is tricky and our inability to find a solution right now to theoretical puzzles is not always a good reason to abandon a view.
Re: the dependence on future existence concerning the values of "freedom/autonomy, relationships (friendship/family/love), art/beauty/expression, truth/discovery, the continuation of tradition/ancestors' efforts, etc.," I think that most of these (freedom/autonomy, relationships, truth/discovery) are considered valuable primarily because of their role in "the good life," i.e. their contribution to individual wellbeing (as per "objective list" theories of wellbeing), so the contingency seems pretty clear here. Much less so for the others, unless we are convinced that people only value these instrumentally.
Local vs. global optimization in career choice
Like many young people in the EA community, I often find myself paralyzed by career planning and am quick to second-guess my current path, developing an unhealthy obsession for keeping doors open in case I realize that I really should have done this other thing.
Many posts have been written recently about the pitfalls of planning your career as if you were some generic template to be molded by 80,000 Hours [reference Holden's aptitudes post, etc.]. I'm still trying to process these ideas and think that the distinction between local and global optimization may help me (and hopefully others) with career planning.
Global optimization involves finding the best among all possible solutions. By its nature, EA is focused on global optimization, identifying the world's most pressing problems and what we can do to solve them. This technique works well at the community level: we can simultaneously explore and exploit, transfer money between cause areas and strategies, and plan across long timescales. But global optimization is not as appropriate in career planning. Instead, perhaps it is better to think about career choice in terms of local optimization, finding the best solution in a limited set. Local optimization is more action-oriented, better at developing aptitudes, and less time-intensive.
The differences between global and local optimization are perhaps similar to the differences between sequence-based and cluster-based thinking [reference Holden's post]. Like sequence-based thinking, which asks and answers questions with linear, expected-value style reasoning, global optimization is too vulnerable to subtle changes in parameters. Perhaps I've enrolled in a public health program but find AI safety and animal suffering equally compelling cause areas. Suppose I'm too focused on global optimization. In that case, a single new report by the Open Philantrophy Project suggesting shorter timelines for transformative AI might lead me to drop out of my program and begin anew as a software engineer. But perhaps the next day, I find out that clean meat is not as inevitable as I once thought, so I leave my job and begin studying bioengineering.
Global optimization makes us more likely to vacillate between potential paths excessively. The problem, though, is that we need some stability in our goals to make progress and develop the aptitudes necessary for impact in any field. Adding onto this the psychological stress of constant skepticism about one's trajectory, it seems that global optimization can be a bad strategy for career planning. The alternative, local optimization, would have us look around our most immediate surroundings and do our best within that environment. Local optimization seems like a better strategy if we think that "good correlates with good," and aptitudes are likely to transfer if we later become convinced that no, really, I should have done this other thing.
I think the difficult thing for us is to find the right balance between these two optimization techniques. We don't want to fall into value traps or otherwise miss the forest for the trees, focusing too much on our most immediate options without considering more drastic changes. But too much global optimization can be similarly dangerous.
You say that care more about the preference of people than about total wellbeing, and that it'd change your mind if it turns out that people today prefer longtermist causes.What do you think about the preferences of future people? You seem to take the "rather make people happy than to make happy people" point of view on population ethics, but future preferences extend beyond their preference to exist. Since you also aren't interested in a world where trillions of people watch Netflix all day, I take it that you don't take their preferences as that important.
What do you mean by this?
OP said, "I also care about people’s wellbeing regardless of when it happens." Are you interpreting this concern about future people's wellbeing as not including concern about their preferences? I think the bit about a Netflix world is consistent with caring about future people's preferences contingent on future people existing. If we accept this kind of view in population ethics, we don't have welfare-related reasons to ensure a future for humanity. But still, we might have quasi-aesthetic desires to create the sort of future that we find appealing. I think OP might just be saying that they lack such quasi-aesthetic desires.
(As an aside, I suspect that quasi-aesthetic desires motivate at least some of the focus on x-risks. We would expect that people who find futurology interesting would want the world to continue, even if they were indifferent to welfare-related reasons. I think this is basically what motivates a lot of environmentalism. People have a quasi-aesthetic desire for nature, purity, etc., so they care about the environment even if they never ground this in the effects of the environment on conscious beings.)
Perhaps you are referring to the value of creating and satisfying these future people's preferences? If this is what you meant, a standard line for preference utilitarians is that preferences only matter once they are created. So the preferences of future people only matter contingent on the existence of these people (and their preferences).
There are several ways to motivate this, one of which is the following: would it be a good thing for me to create in you entirely new preferences just so I can satisfy them? We might think not.
This idea is captured in Singer's Practical Ethics (from back when he espoused preference utilitarianism):
The creation of preferences which we then satisfy gains us nothing. We can think of the creation of the unsatisfied preferences as putting a debit in the moral ledger which satisfying them merely cancels out... Preference Utilitarians have grounds for seeking to satisfy their wishes, but they cannot say that the universe would have been a worse place if we had never come into existence at all.
Yeah my mistake, I should have been clearer about the link for the proposed changes. I think we’re mostly in agreement. My proposed list is probably overcorrecting, and I definitely agree that more criticisms of both approaches are needed. Perhaps a compromise would be just including the reading entitled “Common Ground for Longtermists,” or something similar.
I think you’re right that many definitions of x-risk are broad enough to include (most) s-risks, but I’m mostly concerned about the term “x-risk” losing this broader meaning and instead just referring to extinction risks. It’s probably too nuanced for an intro syllabus, but MichaelA’s post (https://forum.effectivealtruism.org/posts/AJbZ2hHR4bmeZKznG/venn-diagrams-of-existential-global-and-suffering) could help people to better understand the space of possible problems.
Hey Mauricio, thanks for your reply. I’ll reply later with some more remarks, but I’ll list some quick thoughts here:
-
I agree that s-risks can seem more “out there,” but I think some of the readings I’ve listed do a good job of emphasizing the more general worry that the future involves a great deal of suffering. It seems to me that the asymmetry in content about extinction risks vs. s-risks is less about the particular examples and more about the general framework. Taking this into account, perhaps we could write up something to be a gentler introduction to s-risks. The goal is to prevent people from identifying “longtermism” as just extinction risk reduction.
-
Yeah this is definitely true, but completely omitting such a distinctively EA concept like s-risks seems to suggest that something needs to be changed.
-
I think the reading I listed entitled “Common Ground for Longtermists” should address this worry, but perhaps we could add more. I tend to think that the potential for antagonism is outweighed by the value of broader thinking, but your worry is worth addressing.
Hi Aaron, thanks for your reply. I’ve listed some suggestions in one of the hyperlinks above, but I’ll put it here too: https://docs.google.com/document/d/1niRwbh3eejByFQwoiZ0NiaSZDUawn206PUmHs7aKL0A/edit?usp=sharing
I have not put much time into this, so I’d love to hear your thoughts on the proposed changes.
Some criticism of the EA Virtual Programs introductory fellowship syllabus:
I was recently looking through the EA Virtual Programs introductory fellowship syllabus. I was disappointed to see zero mention of s-risks or the possible relevance of animal advocacy to longtermism in the sections on longtermism and existential risk.
I understand that mainstream EA is largely classical utilitarian in practice (even if it recognizes moral uncertainty in principle), but it seems irresponsible not to expose people to these ideas even by the lights of classical utilitarianism.
What explains this omission? A few possibilities:
- The people who created the fellowship syllabus aren't very familiar with s-risks and the possible relevance of animal advocacy to longtermism.
- This seems plausible to me. I think founder effects heavily influence EA, and the big figures in mainstream EA don't seem to discuss these ideas very much.
- These topics seem too weird for an introductory fellowship.
- It's true that a lot of s-risk scenarios are weird. But there's always some trade-off that we have to make between mainstream palatability and potential for impact. The inclusion of x-risks shows that we are willing to make this trade-off when the ideas discussed are important. To justify the exclusion of s-risks, the weirdness-to-impact ratio would have to be much larger. This might be true of particular s-risk scenarios, but even so, general discussions of future suffering need not reference these weirder scenarios. It could also make sense to include discussion of s-risks as optional reading (so as to avoid turning off people who are less open-minded).
- The possible relevance of animal advocacy to longtermism does not strike me as any weirder than the discussion of factory farming, and the omission of this material makes longtermism seem very anthropocentric. (I think we could also improve on this by referring to the long-term future using terms like "The Future of Life" rather than "The Future of Humanity.")
More generally, I think that the EA community could do a much better job of communicating the core premise of longtermism without committing itself too much to particular ethical views (e.g., classical utilitarianism) or empirical views (e.g., that animals won't exist in large numbers and thus are irrelevant to longtermism). I see many of my peers just defer to the values supported by organizations like 80,000 Hours without reflecting much on their own positions, which strikes me as quite problematic. The failure to include a broader range of ideas and topics in introductory fellowships only exacerbates this problem of groupthink.
[Note: it's quite possible that the syllabus is not completely finished at this point, so perhaps these issues will be addressed. But I think these complaints apply more generally, so I felt like posting this.]
Yeah I'm not really sure why we use the term x-risk anymore. There seems to be so much disagreement and confusion about where extinction, suffering, loss of potential, global catastrophic risks, etc. fit into the picture. More granularity seems desirable.
https://forum.effectivealtruism.org/posts/AJbZ2hHR4bmeZKznG/venn-diagrams-of-existential-global-and-suffering is helpful.
Just adding onto this, for those interested in learning how a Kantian meta-ethical approach might be compatible with a consequentialist normative theory, see Kagan's "Kantianism for Consequentialists": https://campuspress.yale.edu/shellykagan/files/2016/07/Kantianism-for-Consequentialists-2cldc82.pdf
Has Singer ever said anything about s-risks? If not, I’m curious to hear his thoughts, especially concerning how his current view compares to what he would’ve thought during his time as a preference utilitarian.
Sorry, I'm a bit confused on what you mean here. I meant to be asking about the prevalence of a view giving animals the same moral status as humans. You say that many might think nonhuman animals' interests are much less strong/important than humans. But I think saying they are less strong is different than saying they are less important, right? How strong they are seems more like an empirical question about capacity for welfare, etc.
Ya, I think 80,000 Hours has been a bit uncareful. I think GPI has done a fine job, and Teruji Thomas has worked on person-affecting views with them.
Woops yeah, I meant to say that GPI is good about this but the transparency and precision gets lost as ideas spread. Fixed the confusing language in my original comment.
In the longtermism section on their key ideas page, 80,000 Hours essentially assumes totalism without making that explicit:
Yeah this is another really great example of how EA is lacking in transparent reasoning. This is especially problematic since many people probably don't have the conceptual resources necessary to identify the assumption or how it relates to other EA ideas, so the response might just be a general aversion to EA.
This article is a bit older (2017) so maybe it's more forgiveable, but their coverage of the asymmetry there is pretty bad.
As another piece of evidence, my university group is using an introductory fellowship syllabus recently developed by Oxford EA and there are zero required readings about anything related to population ethics and how different views here might affect cause prioritization. Instead extinction risks are presented as pretty overwhelmingly pressing.
FWIW, I'm skeptical of this, too. I've responded to that paper here, and have discussed some other concerns here.
Thanks, gonna check these out!
Thanks for this post. Looking forward to more exploration on this topic.
I agree that moral circle expansion seems massively neglected. Changing institutions to enshrine (at least some) consideration for the interests of all sentient beings seems like an essential step towards creating a good future, and I think that certain kinds of animal advocacy are likely to help us get there.
As a side note, do we have any data on what proportion of EA's adhere to the sort of "equal consideration of interests" view on animals which you advocate? I also hold this view, but its rarity may explain some differences in cause prioritization. I wonder how rare this view is even within animal advocacy.
Thanks for writing this up.
These are all interesting thoughts and objections that I happen to find persuasive. But more generally, I think EA should be more transparent about what philosophical assumptions are being made, and how this affects cause prioritization. Of course the philosophers associated with GPI are good about this, but often this transparency and precision gets lost as ideas spread.
For instance, in discussions of longtermism, totalism often seems to be assumed without making that assumption clear. Other views are often misrepresented, for example in 80,000's post "Introducing longtermism" where they say:
This objection is usually associated with a “person-affecting” view of ethics, which is sometimes summed up as the view that “ethics is about helping make people happy, not making happy people”. In other words, we only have moral obligations to help those who are already alive...
But of course person-affecting views are diverse and they need not imply presentism.
From my experience leading an EA university group, this lack of transparency and precision often has the effect of causing people with different philosophical assumptions to reject longtermism altogether, which is a mistake since it's robust across various population axiologies. I worry that this same sort of thing might cause people to reject other EA ideas.
Hi all, I'm sorry if this isn't the right place to post. Please redirect me if there's somewhere else this should go.
I'm posting on behalf of my friend, who is an aspiring AI researcher in his early 20's, and is looking to live with likeminded individuals. He currently lives in Southern California, but is open to relocating (preferably USA, especially California).
Please message jeffreypythonclass+ea@gmail.com if you're interested!
AFAIK the paralysis argument is about the implications of non-consequentialism, not about down-side focused axiologies. In particular, it's about the implications of a pair of views. As Will says in the transcript you linked:
"but this is a paradigm nonconsequentialist view endorses an acts/omissions distinction such that it’s worse to cause harm than it is to allow harm to occur, and an asymmetry between benefits and harms where it’s more wrong to cause a certain amount of harm than it is right or good to cause a certain amount of benefit... And if you have those two claims, then you’ve got to conclude [along the lines of the paralysis argument]".
Also, I'm not sure how Lukas would reply but I think one way of defending his claim which you criticize, namely that "the need to fit all one’s moral intuitions into an overarching theory based solely on intuitively appealing axioms simply cannot be fulfilled", is by appealing to the existence of impossibility theorems in ethics. In that case we truly won't be able to avoid counterintuitive results (see e.g. Arrhenius 2000, Greaves 2017). This also shouldn't surprise us too much if we agree with the evolved nature of some of our moral intuitions.
This was such a fun read. Bentham is often associated with psychological egoism, so it seems somewhat odd to me that he felt a need to exhort readers to fulfill their own pleasure (since apparently all actions are done on this basis anyway).
Could you say more (or work on that post) about why formal methods will be unhelpful? Why are places like Stanford, CMU, etc. pushing to integrate formal methods with AI safety? Also Paul Christiano has suggested formal methods will be useful for avoiding catastrophic scenarios. (Will update with links if you want.)