Saw this quote somewhere and it made me think of this comment again:
Yeah, I haven’t thought about this question previously and am not very familiar with AI safety research/debates (even though I occasionally skim stuff), but one objection that came to my mind when reading the original post/question was “If you aren’t working on it, does that actually mean there will be one whole less person working on it?” Of course, I suppose it’s possible that AI safety is somewhat weird/niche enough (in comparison to e.g., nursing, teaching) where the person-replacement ratio is moderate or low and/or the relative marginal returns of an additional worker are still fairly high, e.g., your individual choice to get a job in AI safety may have the expected average effect of increasing the total amount of people working on the project by, say, 0.75. I don’t have the field knowledge to answer that question, and it’s only one of many factors to consider, but if it is the case that the replaceability ratio is relatively high (e.g., your net average effect is <0.25) then that immediately has a big reduction on the potential for “I am increasing the number of people working on AI which increases the likelihood of bad AI occurring.”
That being said, I’m confident there are much better counterarguments that draw on more knowledge of how working on AI safety can reduce the risk you are talking about without also contributing to this blob concept of “more people working on AI” which you worry could increase the likelihood of AGI, which increases the likelihood of bad AGI.
Ah, I figured it was more of an argument against delayed giving rather than about plainly not giving. To clarify further, is your claim that the price of [saving?] a child’s life is actually doubling every few years (out of proportion to inflation), or is it just supposing a hypothetical world where that is the case?
Average annual stock market returns of roughly 9% while accounting for roughly 2% inflation, would see you double your money in about (70/(9%-2%)=10 years. It’s not exactly fast, but it’s doubling after accounting for inflation so it’s nothing to sneeze at. (Also worth noting, that’s not accounting for the additional deposits you make with your income, which would likely double a given amount much faster)
Of course, anyone should be careful about investing (e.g., potential for downturns or personal inability to make sound investment decisions in line with really basic advice), potential for value drift, and the possibility that some current causes are urgent/warrant immediate funding despite the possibility that a later cause might be even more important that the current one. However, for some people/situations those concerns may also be partially if not wholly offset by the concepts like increased insight/research into charity effectiveness.
Ultimately, I don’t have a strict opinion on which method is generally better, but I don’t think it’s justified to so heavily dismiss/criticize delayed giving.
I think this post was written well and points out a potentially serious effect. I also think there is probably some validity to it—especially for some people it might be very strong—but I also like to wrinkle (or try to wrinkle) neat ideas, so I’ll pose some potential objections I had while reading. In short, do you think it’s possible there’s a lot of correlation-causation confusion going on here? For example:
Bigger and more-complex projects tend to be more important (or else we probably wouldn’t do the project), and the complexity and size tend to make it harder to get started or get into a rhythm. For example, I found writing my senior undergrad thesis to be fairly difficult because it was so daunting and complex: I often would ask myself “where do I start or where do I pick up from my most recent work?” My perfectionist streak definitely hurt me throughout the process, but in the end I don’t think the importance of the thing was the key issue. However, a large project/task may not necessarily be complex: it might be simple data entry, and I feel that I am decently capable of motivating myself to work on such projects in batches (perhaps even more so when it is more important!)
These bigger and more-complex projects also tend to have longer deadlines, and longer deadlines plays into procrastination habits.
If you notice yourself having to force/motivate yourself to work on something, it’s typically because you don’t want to do it despite its importance. Is it possible that there is a degree of observation bias in that we might not notice all the cases where the importance of a project successfully motivated us to work on it? Or, perhaps more importantly/clearly, we might not consider all the times we just gave up on a potential project/task when we deemed it not important (rather than unsuccessfully berating ourselves to complete the task)?
Ultimately, I’m not disputing that there probably is some degree of importance ⇒ avoidance effect, but I think it’s possible to overestimate the causal relationship between importance and avoidance due to other correlated factors and observation biases.
I certainly would be interested in seeing such a system go into place—I think it would probably be beneficial—the main issue is just whether something like that is likely to happen. For example, it might be quite difficult to establish agreement between Charity Evaluator and GiveWell when it comes to the benefits of certain charities. Additionally, there may be a bit of survivor bias when it comes to organizations that have worked like FIDE, although I still think the main issue is 1) the analysis/measurement of effectiveness is difficult (requiring lots of studies vs. simply measuring album downloads/streams); and 2) the determination of effectiveness may not be widely agreed upon. That’s not to say it shouldn’t be tried, but I think that might contribute to limiting the effectiveness relative to the examples you cite.
I did also initially think that it might be good to try to change the lumberjack instance, if possible, although it wasn’t for the same reason: I just feel that there is much more of a case to make that the lumberjack deserves a whole-of-community effort since there is a plausible chance the extra bird could make a difference. But after considering this about the non-urgency of the sprout vs the lumberjack, I especially feel it may not the best example. Still, I understood the message/idea, and it’s hard to know how non-EAs might react to the situation. Just something to keep in mind.
“But it’s got to confer status to the charity and people like Jay Z can gain more status by donating to it”—I think this brushes a good point which I’d like to see fleshed out more. On some level I’m still a bit skeptical in part because I think it’s more difficult to make these kinds of designations/measurements for charities whereas things like album statuses are very objective (i.e., a specific number of purchases/downloads) and in some cases easier to measure. Additionally, for some of those cases there is a well-established and influential organization making the determination (e.g., football leagues, FIDE for chess). I definitely think something could be done for traditional charities (e.g., global health and poverty alleviation), but it would very likely be difficult for many other charities, and it still would probably not be as widely recognized as most of the things you mentioned.
I can’t say I’ve read that many of the creative fiction yet, but of those that I have read this is probably my favorite. It’s simple, has nice/sweet illustrations, and not too heavy while also conveying some basic ideas. +2
(Or maybe, “I invested the money and made above-inflation returns while also waiting to see further research on the most cost-effective interventions. I did occasionally make large donations with the investments when particularly opportune moments arose.”)
I see what you mean, and again I have some sympathy for the argument that it’s very difficult to be confident about a given probability distribution in terms of both positive and negative consequences. However, to summarize my concerns here, I still think that even if there is a large amount of uncertainty, there is typically still reason to think that some things will have a positive expected value: preventing a given event (e.g., a global nuclear war) might have a ~0.001% of making existence worse in the long-term (possibility A), but it seems fair to estimate that preventing the same event also has a ~0.1% chance of producing an equal amount of long-term net benefit (B). Both estimates can be highly uncertain, but there doesn’t seem to be a good reason to expect that (A) is more likely than (B).
My concern thus far has been that it seems like your argument is saying “(A) and (B) are both really hard to estimate, and they’re both really low likelihood—but neither is negligible. Thus, we can’t really know whether our interventions are helping. (With the implicit conclusion being: thus, we should be more skeptical about attempts to improve the long-term future)” (If that isn’t your argument, feel free to clarify!). In contrast, my point is “Sometimes we can’t know the probability distribution of (A) vs. (B), but sometimes we can do better-than-nothing estimates, and for some things (e.g., some aspects of X-risk reduction) it seems reasonable to try.”
This seems to be an issue of only considering one side of the possibility distribution. I think it’s very arguable that a post-nuclear-holocaust society is just as if not more likely to be more racist/sexist, more violent or suspicious of others, more cruel to animals (if only because our progress in e.g., lab-grown meat will be undone), etc. in the long term. This is especially the case if history just keeps going through cycles of civilizational collapse and rebuilding—in which case we might have to suffer for hundreds of thousands of years (and subject animals to that many more years of cruelty) until we finally develop a civilization that is capable of maximizing human/sentient flourishing (assuming we don’t go extinct!)
You cite the example of post-WW2 peace, but I don’t think it’s that simple:
there were many wars afterwards (e.g., the Korean War, Vietnam), they just weren’t all as global in scale. Thus, WW2 may have been more of a peak outlier at a unique moment in history.
It’s entirely possible WW2 could have led to another, even worse war—we just got lucky. (consider how people thought WW1 would be the war to end all wars because of its brutality, only for WW2 to follow a few decades later)
Inventions such as nuclear weapons, the strengthening of the international system in terms of trade and diplomacy, the disenchantment with fascism/totalitarianism (with the exception of communism), and a variety of other factors seemed to have helped to prevent a WW3; the brutality of WW2 was not the only factor.
Ultimately, I still consider that the argument that seemingly horrible things like nuclear holocausts (or The Holocaust) or world wars are more likely to produce good outcomes in the long term just generally seems improbable. (I just wish someone who is more familiar with longtermism would contribute)
(Note: I’m not well-steeped in the longtermism literature, so don’t look to me as some philosophical ambassador; I’m only commenting since I hadn’t seen any other answers yet.)
I get lost with your argument when you say “the standard deviation as a measure of uncertainty [...] could be so large that the coefficient of variation is very small” What is the significance/meaning of that?
I read your Medium post and I think I otherwise understand the general argument (and even share similar concerns at times). However, my response to the argument you lay out there would mainly be as follows: yes, it technically is possible that a civilizational nuclear reset could lead to good outcomes in the long term, but it’s also highly improbable. In the end, we have to weigh what is more plausible, and while there will be a lot of uncertainty, it isn’t fair to characterize every situation as purely/symmetrically uncertain, and one of the major goals of longtermism is to seek out these cases where it seems that an intervention is more likely to help than to hurt in the long term.
One of the major examples I’ve heard longtermists talk about is reducing x-risk. You seem to take issue with this point, but I think the reasoning here is tenuous at best. More specifically, consider the example you give regarding “what if nuclear reset leads to a society that is so “enlightened… that they no longer farm animals for food.” Does it seem more plausible that nuclear reset will lead to an enlightened society or a worse society (and enormous suffering in the process)? As part of this, consider all the progress our current society has made in the past ~60 years regarding things like lab-grown meat and veganism—and how much progress in these and other fields would be lost in such a scenario. In this case, it seems far more plausible that preventing a nuclear holocaust will be better for the long term future.
Haha, I certainly wouldn’t label what you described/presented as “timecube weird.” To be honest, I don’t have a very clear cut set of criteria, and upon reflection it’s probable that the prior is a bit over-influenced by my experiences with some social science research and theory as opposed to hard science research/theory. Additionally, it’s not simply that I’m skeptical of whether the conclusion is true, but more generally my skepticism heuristics for research is about whether whatever is being presented is “A) novel/in contrast with existing theories or intuitions; B) is true; and/or C) is useful.” For example, some theory might be basically rehashing what existing research already has come to consensus on but simply worded in a very different way that adds little to existing research (aside from complexity); alternatively, something could just be flat out wrong; alternatively, something could be technically true and novel as explicitly written, but that is not very useful (e.g., tautological definitions), whereas the common interpretation is wrong (but would be useful if it were right).
Still, two of the key features here that contributed to my mental yellow flags were:
The emphasis on jargon and seemingly ambiguous concepts (e.g., “harmony”) vs. a clear, lay-oriented narrative that explains the theory—crucially including how it is different from other plausible theories (in addition to “why should you believe this? / how did we test this?”). STEM jargon definitely seems different from social science jargon in that STEM jargon seems to more often require more knowledge/experience to get a sense of whether something is nonsense strung together or just legitimate-but-complicated analyses, whereas I can much more easily detect nonsense in social science work when it starts equivocating ideas and making broad generalizations.
(To a lesser extent) The emphasis on mathematical analyses and models for something that seemed to call for a broader approach/acceptance of some ambiguity. (Of course, it’s necessary to mathematically represent some things, but I’m a bit skeptical of systems that try to break down such complex concepts as consciousness and affective experience into a mathematical/quantified representation, just like how I’ve been skeptical of many attempts to measure/operationalize complex conceptual variables like “culture” or “polity” in some social sciences, even if I think doing so can be helpful relative to doing nothing—so long as people still are very clear-eyed about the limitations of the quantification)
In the end, I don’t have strong reason to believe that what you are arguing for is wrong, but especially given points like I just mentioned I haven’t updated my beliefs much in any direction after reading this post.
I’m a bit hesitant to upvote this comment given how critical it is [was] + how little I know about the field (and thus whether the criticism is deserved), but I’m a bit relieved/interested to see I wasn’t the only one who thought it sounded really confusing/weird. I have somewhat skeptical priors towards big theories of consciousness and suffering (sort of/it’s complicated) + towards theories that rely on lots of complicated methods/jargon/theory (again, sort of/with caveats)—but I also know very little about this field and so I couldn’t really judge. Thus, I’m definitely interested to see the opinions of people with some experience in the field.
Just to be clear: is your point that because the industry is already/now getting lots of money from billionaires the marginal value of donating additional money is smaller? And/or is it (also) that donating will lead to billionaires donating less?
I was hoping to see some more answers by now, but seeing none I’ll provide some initial points that I expect people who are more familiar with the field could flesh out much better than I. I’m not familiar with the full case for biodiversity, but just based on generic reasoning about cause areas in general, I’d suspect some of the major reasons EAs do not seem to consider it as important as factory farming, wild animal welfare, or even climate change (to name a few environmentally-oriented cause areas) include:
It’s unclear how significant the extrinsic, welfare-oriented value of biodiversity even is. Put most bluntly, “if some random frog species goes extinct, so what?” I understand there are arguments like “what if this species proves to be very useful for e.g., pharmaceuticals or the broader ecosystem”, but I’d want to see some quantification and comparison with other cause areas. Is it worse for the last 100 fish of some random species to die (=> extinction) than for 100 chickens to suffer factory farming conditions? Maybe, but I could also definitely make the case for the opposite conclusion. And as a few other answers/comments pointed out, it definitely seems to be much smaller in significance than human X-risks.
(On a smaller note since I am not very familiar with this field, I’ll add that it’s a bit difficult to understand what all of the charts, figures, and bullet points on the link you shared amount to. Lots of things may qualify as endangered, but how many things actually go extinct? How is this impacting the environment? It’s all just much more difficult to neatly quantify the situation than it is to put a probability estimate on X-risk, preventable mortality estimate for malaria, number of animals in factory farming conditions, etc.)
It’s unclear whether there are cost-effective interventions (in conjunction with point 4): if you would like to propose/list some, that would help, but if it’s the case that vast numbers of these species are going extinct/becoming endangered due to a wide number of environmental changes, it seems more difficult to identify highly cost-effective interventions (in contrast to, e.g., bed nets for malaria, surgery for trachoma).
Some (non-EA) organizations are already working on the field of biodiversity and conservation. This certainly doesn’t mean that it’s already fully saturated, but in order to determine whether there are cost-effective interventions that need funding, one must also consider what kinds of efforts/funding already exist.
In the end, this isn’t to say that biodiversity is necessarily “unimportant” and completely intractable, but given the combined issues surrounding comparative importance, tractability, and neglectedness I don’t know if it’s really an area where EA would have a comparative advantage in relative to e.g., AI safety/alignment (which itself may prove very impactful for the environment and many other problems), cost-effective health interventions (e.g., bed nets, deworming), and general long-termism. I would be willing to read an argument for why it is overly neglected, but I haven’t seen that argument made yet.
I’m still not a huge fan of the way it is written—it sounds almost like a strawman description of wild animal welfare work. In particular, I don’t think adding “maybe” does enough blunt/caveat the second part of the sentence, which is not presented very delicately.
I’m not an academic nor do I otherwise have much familiarity with academia careers, but I have occasionally heard people talk about the importance of incentive structures like tenure publication qualification, the ease of staying in established fields, the difficulty/opportunity cost of changing research fields later in your career, etc. Thus I think it would be helpful/interesting to look at things more from that incentive structure side of things in addition to asking “how can we convince people AI safety is important and interesting?”
Those equations are definitely less intimidating / I could understand them without an issue. (And that’s totally fine, I hope it’s helpful)