> Betting aside, if you really thought the world was likely to end, you should be going into massive debt, secure in the knowledge that you won’t have to repay it. But almost no AI doomster — EA or otherwise — does so. I know, they’ve got just-so stories about why they’re not playing the obvious strategy given their stated beliefs. The best explanation, however, is that their stated beliefs are mere lip service.
This seems only obviously true if you think the probability of doom is very high (80%+?). Most people concerned with AI safety think the risk is more in the 10-30% range. That might justify a slightly greater-than-typical rate of spending, but not going into massive debt.
Some of the non-doom scenarios could involve massive investment returns and the invention of new goods (but lower employment income), which could all suggest saving *more* than typically makes sense.
The overall effect of AI on how much to spend vs. save seems unclear and pretty sensitive to the probability you put on different scenarios.
I don't understand what Caplan is concretely suggesting here. Ways I could go into debt:
1. Buy a house with a mortgage instead of rent. If Caplan is counting this form of debt then sure, lots of doomers do that.
2. Take out a personal loan. A quick search online indicates that this maxes out at $50k, maybe $100k for "exceptionally well qualified" individuals. $50k is < 10% of most AI people's salaries.
Is there some option I'm missing that would let me borrow millions of dollars that I don't have to pay back for 5+ years? I feel unsurprised that this doesn't exist, as I assume adverse selection from e.g. terminally ill borrowers would destroy the market, but maybe something does exist.
Putting a number on this: 20% doom within 10 years is a ~2.2%/yr hazard (0.8^0.1 ≈ 0.978). In a standard consumption model that works like a discount rate 2 points higher: save a bit less, not borrow heavily.
Exactly. There's a small increase in the discount rate. Then risk aversion mutes the effect: spending 30% more before we all die won't make you much happier, but going bankrupt in the other 80% of worlds would be really bad.
It's incredible that even smart people like Bryan Caplan are showing this bugman mentality where the point of spending is just to maximize pleasure before they die.
Bryan, despite his wisdom, is sadly mistaken on animal welfare. He and I would agree that insect welfare is nonobvious, but we would also agree that it is a logical corollary of the wrongness of torturing dogs for fun. I think ‘it is evil to torture a dog for fun because of harm to the dog’ is more obvious than ‘insects never deserve moral consideration.’ Frankly, I think many people would move toward this position just by looking at zoomed-in pictures of bees. They have faces, just like us.
I disagree. Regardless, what am I meant to conclude from this, that it’s not wrong to torture animals? Humans in nature suffer a lot, does this justify torturing a human? I would be interested if you could elaborate.
Animal cruelty is being cruel to an animal. The details and edge cases are probably hard to work out, like with nearly all definitions.
Can you please elaborate on what this has to do with the just treatment of animals? It seems you are saying that it is not wrong to abuse animals because they are often hurt in nature. But humans are often hurt in nature. So it seems like your view implies that it is not wrong to abuse humans.
Dog meat consumption is vanishing in East Asia now that dogs have become popular housepets. Those born after 2000 are more like Westerners than their grandparents in how they treat dogs (i.e. as substitute children, not as tool animals).
Dear Bryan, I appreciate the moderate defense of EAs but this is just pseudo-philosophy: 'Morality rests on the value of human well-being and prima face human rights. If you demand an argument for any of these, the right response is: “It’s obvious. The end.”' See https://walterveit.substack.com/p/most-a-priori-philosophy-is-a-waste
Bryan Caplan: “Here is my opinion. It is what I deem the popular / common sense position, therefore it is right. Deal with it.”
Like I have devoted a good deal of my life to the human wellbeing / global health side of things, but people like you so blatantly stating that their priors make them right when they are filled with common human biases specific to our current era are pushing me more & more to the animal welfare arguments.
As a former pro gambler, I love the instinct, but your end-of-the-world bets are undermined by bet sizes.
A few hundred bucks is ~play money for you and Eliezer and most other folks you might bet with, so everyone can take bets or turn them down without much thought.
But if you raise bet sizes to ~$10k+, you can’t prepay someone unless you really trust them to pay you back.
So structuring a meaningful end-of-the-world bet is more complicated, and having few takers proves little.
The other important point Ben makes above (re: why not go into debt?) is that most people worried about AI think the risk is 10%-30%. So your odds need to reflect that.
E.g. I'd take a bet where for every $100 you prepay, I owe you $125 CPI-adjusted on January 1st, 2036, for any amount up to $100k prepaid.
But if you turn that down because you can't trust I'll pay, or decide it's not worth it all things considered, I can't use that as proof you don't really think AI risk is low!
I have been absorbed into the bizarre, intriguing, and sometimes infuriating EA subculture over the past couple of years by mutual friends and by my partner. I quite enjoy the time I spend with them; they are bright, curious, and eager to help, which are traits I respect and value dearly in my chosen company.
However, I am often perplexed and frustrated with their heuristics and metrics. Almost nothing makes me more upset than someone trying to (in dead seriousness) justify anti-natalism and longtermism to me. Why the fuck are we trying to deny the fruition of life itself? If I was a religious figure perhaps 400 years ago, I would burn the anti-natalists at the stake for hearsay and crimes against the sanctity of God.
If I could force every EA to read this post at the threat of an incoming trolley, I would. Thank you for being a vessel for my frustrations and conflicting emotions about the movement.
I'm with you on anti-natalism being deeply anti-human. The EA people I know are pretty passionately pro-natalist. But longtermism seems deeply pro-human. I mean, I'm sure there's some crazy thought experiment where you can wring monstrous conclusions out of it. I'm with Bryan that we should apply common sense to such cases. But caring about the long-term flourishing of humanity is common sense, isn't it?
I was hugely impressed by Toby Ord's The Precipice. It had what felt like a very common-sense case for longtermism, but mostly it was just pure pragmatics -- examining every possible human ~extinction risk over the next 100 years and meticulously working out how much to worry about each one. (Asteroid/comet 0.0001%, nuclear war 0.1%, runaway climate change <0.1%, natural pandemics 0.01%, engineered pandemics 3%, etc.)
If you don't want to commit to the whole book, Scott Alexander's review is great:
It occurs to me there's a possible misreading of the above parenthetical, like "none of these risks are worth much worry!" So I should clarify that I was picking particularly low-risk entries from Ord's list in order to establish credibility with anti-worriers like Bryan Caplan.
Meta-clarification: A 3% risk of extinction in the next 100 years from an engineered pandemic is nothing to sneeze at!
But of course dwarfing all of that is AI risk. I think Ord gained huge credibility in his 2019 book by not overhyping AI. He called AGI decades, not years, away. Even now his median for AGI is 2038. Still, his original estimate of the extinction risk over the next 100 (now 93 I guess) years was 10% and it's worth engaging with the argument directly, which I think Bryan will admit his blog post here does not do.
I'm very worried about the real-world harm of belief in bug welfare. I think AI with bug welfare values is one of the most likely scenarios in which AI would exterminate humanity.
I put 5% odds on AI exterminating humanity in order to create more bugs (or shrimp or computer pleasure machines etc.) at 10% some time between 2050 and 2150.
Brian Tomasik would tell you that as r-strategists, bugs tend to live short and miserable lives. Therefore we don't want to create more of them, we want to create less. And human civilization can be an effective way to do that.
Bryan, I think your case is: EAs are likeable and have improved philanthropy, but a lack of common sense has led many into AI doom and bug welfare, and when a conclusion is absurd enough you should run the argument backwards.
I agree with a fair amount of what you write. EAs are prone to groupthink, partly because the network is small and dense, and EAs often treat common sense as a prior to update away rather than as evidence. Your point about reasoning from obvious premises to non-obvious conclusions is a good one.
However, I have three disagreements:
1. On going into debt. As Ben Todd says above, most put the risk at less than 30%, so there's a greater than 70% chance of having to repay, and many non-doom scenarios involve high returns to capital, which argues for saving and investing more. Paul Christiano is about 2x levered long the market.
2. On whether many EAs have 'lost interest' in global health. Of the roughly $2bn moved by the EA ecosystem in 2025, about half went to global health and development and about 30% to reducing global catastrophic risk.
3. On your point about reasoning from highly obvious premises to not-so-obvious conclusions. I think EA concern passes your test. Suffering is bad, whoever has it; whether something can suffer depends on facts about it, not its species; some insects behave as if they feel pain; there are a lot of insects. Conclusion: bug welfare might matter. Or: transformative technologies can do serious harm; AI is advancing fast; we don't fully control it; competition pushes deployment ahead of safety. Conclusion: there's a real chance it goes badly. The denials, that no insect can suffer and there's zero chance of AI catastrophe, aren't obvious at all. We can argue about how much weight small chances of big harms deserve. But that's a disagreement about expected value, not a failure of common sense.
"if you really thought the world was likely to end, you should be going into massive debt"
That does not seem to apply here as most doomers I think have a P(doom) well under 50%. If you think there are 10% chances AI will end the human race, you believe there is a 90% chance you will have to repay your debts.
But the argument that bug welfare matters does proceed from premises that are either (a) empirical, or (b) more obvious than the negated conclusion: "bug pain doesn't matter." Here it is.
1. The reason pain is bad is, at least in part, because of how it feels, not because of other features of the subject, like their intelligence or informational similarity to me. (Note: maybe those other factors do matter, and maybe they matter *a lot*!, but it does seem obvious that *at least some of the badness of pain* is due to the negative phenomenal quality of the painful experience).
2. Bugs can feel pain. They are, at least thinly, conscious and capable of negatively valanced qualia. There is something that it's like to be a bug in pain, and that experience has a negative phenomenal quality *of the same kind* as the one we have when we are in pain.
This is the empirical premise, and it might be wrong! But it's wrongness is a matter for science to figure out, not a matter for us to hand-wave away on account of it being inconvenient. At least for some candidates, like bees and ants, the evidence seems overwhelmingly to favor this premise.
3. Therefore, bug pain is bad.
The reason why you reject this is because of the unintuitive conclusion that bug pain, in aggregate, matters more than human pain, but that commits you to the strange position that bug pain can't be bad *because there are so many of them.* If there were just a few of them, you would maybe admit that their pain mattered.
But this is working backwards. Human pain doesn't get less bad as our population increases. Whether or not bug pain matters is a separate question from how many bugs there are, to be resolved first. If it turns out that bug pain really does matter, than the fact that there are so many bugs really does make it matter more.
I was thinking the same thing. I question the premise that pain is bad. Pain has (evolved for) a purpose. To avoid behaviors and situations that shorten life and lessen the chance of reproduction.
You've snuck in Utilitarian premises, like that there are states of affairs ("welfare") which matter morally. A much more plausible alternative view is that conscious creatures have moral rights, but moral conduct is not about achieving the state of affairs which matters the most or has the most value.
Any plausible moral theory thinks welfare matters in some way. A main confusion among non-philosophers and even some philosophers who are non-specialists is that only utilitarianism can think that welfare matters. But that’s clearly false in the case of, e.g., Rossian deontology. Even Kant thought that suffering mattered, as far as I and AI know.
A majority, maybe, but philosophical arguments require decoupling those concerns.
EAs are often utilitarians because effective altruism follows more naturally from it. Utilitarians (at least almost) *only* value the things EA promotes, so it’s quite easy to come to the conclusion. But this is a sociological factor that we have to decouple from the analysis.
Semi-off-topic, I wondered for a long time what exactly was animal cruelty. Is pushing a dog's nose into its pee cruel (quite aside from whether it works)?
The only simple and vaguely objective definition I could come with was "anything worse than life in the wild, which includes being eaten to death or starving to death after an injury". So don't stub out cigarettes on your dog. That sounds like a pretty sensible definition until you realize it forbids chaining dogs up in the back yard, and cooping up birds in cages and clipping their wings so they can't fly away.
I also had numerous side thoughts which amused me.
* Why is it moral for predators to eat prey alive, but immoral for ranchers to slaughter animals "humanely"?
* If humans slaughtering livestock for food is immoral, why is it moral to applaud wolves to eat livestock?
* Are ranchers who execute murdering wolves dispensing justice?
* Is it cruel for cats and other animal mamas to catch and use prey to teach their cubs how to hunt? Why can't humans keep pets and have bear baiting entertainment?
Yeah i like your definition of animal cruelty as "worse than what would have been in the wild"
By that logic tho, bryans arguments against animal welfare fall short though. I have no problem eating a chicken that lived free in the wild and died fighting for its survival. I do have a problem with one cooped up its entire life in a cage, pecking other chickens from frustration and injected with growth hormones to speed up its slaughter.
The bugs argument doesn't apply here because as far as I know, we don't (yet?) have bug farms. Its not eating meat that's cruel/unethical, it's the animal farming industry that has to reform
It was a fun thought experiment that lasted for several years, and convinced me that there can be no objective definition of animal cruelty. It can only be handled by customs and what neighbors or community put up with. I don't think legislation can do anything except inflame passions.
How can someone understand buying home or life insurance, to hedge against unlikely but expensive events, but not understand keeping your 401k intact instead of immediately spending it, to hedge against low probability case of AI stagnation and having a normal retirement? I give 80% p(doom) by end of 2035, but the 'we're saved by stagnation' case is still a long tail to prepare for.
It's not at all intuitive to me that animal suffering doesn't matter or is negligible, and it seems to have little to do with the pattern of behavior for why people eat meat, so I think I'm not alone.
Is it bad to kick a duckling? I'd say so. Why? Because it would harm my moral character by acting with cruelty? Implausible. If for some reason, kicking ducklings had no impact on how good I am as a person towards humanity, I would still not kick ducklings for fun. I also wouldn't kick a duckling for merely pragmatic reasons like getting from point A to B a little quicker. It's obvious to me that part of why it's wrong is that I'm causing pain to the duckling for no notable gains to myself. Likewise, I've noticed most people take some care to avoid causing needless harm to mammals and birds. This is odd if people genuinely think their lives count for very little.
The account for why most people seem to think eating meat is morally permissible is that there's something about the purpose that makes it okay. Something like "in most circumstances, what we're doing would be wrong, but because the purpose is consumption, it's justified."
Putting aside whether that's correct, this is completely divorced from why Caplan thinks it isn't morally wrong to eat meat. He just thinks animal interests are negligible. But it actually does matter to people whether people are putting pigs in crates they can't turn around in for food or for, say, display in a zoo, where they would want to see the pig be treated much better. Note: there was massive public backlash from attempts to roll back the ability of states to pass farm animal welfare laws recently. It's not just EAs who think even farm animals deserve better treatment than they get.
Here's something I do think most people consider unintuitive in the typical EA position: Peter Singer's thesis that animal interests count equally as human interests. But you don't actually need to accept that thesis to think animal welfare matters to some extent and that it's a worthy cause area for EAs.
It's fine to reject insect welfare by an appeal to common sense and call it case closed. It's a moral question and you can just be a moral intuitionist or be a consequentialist and take it as axiomatic that only humans are moral patients. Maybe dogs and cats too, a little bit at least? But ultimately, yes, there's an arbitrary line to be drawn and common sense is key.
But AI risk isn't like that at all. We have to estimate probabilities of various trajectories and decide whether it's prudent to slow down frontier AI research. I agree the good trajectories are very good. But the bad trajectories are very bad, and as you say, you aren't making an argument against that.
Except the tu-quoque argument that self-consistent doomers would take on massive debt.
Bryan, it's pretty frustrating how thoroughly your commenters have rebutted that, and you brush it all aside with zero engagement other than to call it just-so stories. I'm happy to bite any bullets on my p(doom) of 20% by the end of the 2030s. I'm guessing that doesn't imply it's rational to effectively borrow money from you at a 21% interest rate, which is what I believe your bet with Greg Colburn amounted to. But whatever it does imply, I'm game!
Except there's another rebuttal: I have a high p(doom) *conditional on racing ahead* but I'm hopeful we'll succeed in slowing down.
But, as you know (recall our "Climate Shock" bet) I'm all about finding ways to distill disagreements into wagers, so let's keep trying to do that. Maybe you can identify lines in the sand that, if crossed, would change your mind about AI risk, and we can bet on those?
The AI-2027 scenario (published April 3, 2025) could be a source of ideas for such. On Friday I went through all their predictions through the end of 2026 to see how they're faring and the answer is uncannily well:
Short version: Effective Altruists' casuistry doesn't discredit their assessments of AI risk. For that we have to extrapolate the AI capabilities trend and anticipate the consequences of future superhuman AI.
I'm not a doomer in the sense of having any certainty about either the trend extrapolation or whether superhuman AI will spin out of control. I'm a doomer in the sense of lacking certainty in the other direction.
I'm happy to make wagers that that lack of certainty implies.
> Betting aside, if you really thought the world was likely to end, you should be going into massive debt, secure in the knowledge that you won’t have to repay it. But almost no AI doomster — EA or otherwise — does so. I know, they’ve got just-so stories about why they’re not playing the obvious strategy given their stated beliefs. The best explanation, however, is that their stated beliefs are mere lip service.
This seems only obviously true if you think the probability of doom is very high (80%+?). Most people concerned with AI safety think the risk is more in the 10-30% range. That might justify a slightly greater-than-typical rate of spending, but not going into massive debt.
Some of the non-doom scenarios could involve massive investment returns and the invention of new goods (but lower employment income), which could all suggest saving *more* than typically makes sense.
The overall effect of AI on how much to spend vs. save seems unclear and pretty sensitive to the probability you put on different scenarios.
I don't understand what Caplan is concretely suggesting here. Ways I could go into debt:
1. Buy a house with a mortgage instead of rent. If Caplan is counting this form of debt then sure, lots of doomers do that.
2. Take out a personal loan. A quick search online indicates that this maxes out at $50k, maybe $100k for "exceptionally well qualified" individuals. $50k is < 10% of most AI people's salaries.
Is there some option I'm missing that would let me borrow millions of dollars that I don't have to pay back for 5+ years? I feel unsurprised that this doesn't exist, as I assume adverse selection from e.g. terminally ill borrowers would destroy the market, but maybe something does exist.
Putting a number on this: 20% doom within 10 years is a ~2.2%/yr hazard (0.8^0.1 ≈ 0.978). In a standard consumption model that works like a discount rate 2 points higher: save a bit less, not borrow heavily.
Exactly. There's a small increase in the discount rate. Then risk aversion mutes the effect: spending 30% more before we all die won't make you much happier, but going bankrupt in the other 80% of worlds would be really bad.
It's incredible that even smart people like Bryan Caplan are showing this bugman mentality where the point of spending is just to maximize pleasure before they die.
Bryan, despite his wisdom, is sadly mistaken on animal welfare. He and I would agree that insect welfare is nonobvious, but we would also agree that it is a logical corollary of the wrongness of torturing dogs for fun. I think ‘it is evil to torture a dog for fun because of harm to the dog’ is more obvious than ‘insects never deserve moral consideration.’ Frankly, I think many people would move toward this position just by looking at zoomed-in pictures of bees. They have faces, just like us.
Nature is full of predators and prey. Natural death is being eaten alive or starving to death. Trying to define animal cruelty is a lost cause.
I disagree. Regardless, what am I meant to conclude from this, that it’s not wrong to torture animals? Humans in nature suffer a lot, does this justify torturing a human? I would be interested if you could elaborate.
Disagree with what? My conclusion is that laws against animal cruelty are as ill-defined as dress code laws. If you think otherwise, try defining one.
Animal cruelty is being cruel to an animal. The details and edge cases are probably hard to work out, like with nearly all definitions.
Can you please elaborate on what this has to do with the just treatment of animals? It seems you are saying that it is not wrong to abuse animals because they are often hurt in nature. But humans are often hurt in nature. So it seems like your view implies that it is not wrong to abuse humans.
Well said
I don't think it's obvious, and my understanding is that in east Asia people not only eat dogs but think they taste better as a result of suffering.
I do value dogs above most other non-human species of animals, because they can reciprocate to a certain extent, even if not on the level of Katja Grace's hypothetical of trading with ants https://worldspiritsockpuppet.substack.com/p/we-dont-trade-with-ants
They certainly have more ass-kicking ability https://www.lesswrong.com/posts/Dvc7zrqsdCYy6dCFR/suffering?commentId=pjPbCamDYMNsPf9HB than shrimp.
Dog meat consumption is vanishing in East Asia now that dogs have become popular housepets. Those born after 2000 are more like Westerners than their grandparents in how they treat dogs (i.e. as substitute children, not as tool animals).
Define torture.
I define torture as needless infliction of pain. So, it is scaled against need and human need at that.
Dear Bryan, I appreciate the moderate defense of EAs but this is just pseudo-philosophy: 'Morality rests on the value of human well-being and prima face human rights. If you demand an argument for any of these, the right response is: “It’s obvious. The end.”' See https://walterveit.substack.com/p/most-a-priori-philosophy-is-a-waste
I've written up my objection as a reply here: https://walterveit.substack.com/p/bryan-caplan-vs-effective-altruism
Bryan Caplan: “Here is my opinion. It is what I deem the popular / common sense position, therefore it is right. Deal with it.”
Like I have devoted a good deal of my life to the human wellbeing / global health side of things, but people like you so blatantly stating that their priors make them right when they are filled with common human biases specific to our current era are pushing me more & more to the animal welfare arguments.
As a former pro gambler, I love the instinct, but your end-of-the-world bets are undermined by bet sizes.
A few hundred bucks is ~play money for you and Eliezer and most other folks you might bet with, so everyone can take bets or turn them down without much thought.
But if you raise bet sizes to ~$10k+, you can’t prepay someone unless you really trust them to pay you back.
So structuring a meaningful end-of-the-world bet is more complicated, and having few takers proves little.
The other important point Ben makes above (re: why not go into debt?) is that most people worried about AI think the risk is 10%-30%. So your odds need to reflect that.
E.g. I'd take a bet where for every $100 you prepay, I owe you $125 CPI-adjusted on January 1st, 2036, for any amount up to $100k prepaid.
But if you turn that down because you can't trust I'll pay, or decide it's not worth it all things considered, I can't use that as proof you don't really think AI risk is low!
Dear Bryan,
I have been absorbed into the bizarre, intriguing, and sometimes infuriating EA subculture over the past couple of years by mutual friends and by my partner. I quite enjoy the time I spend with them; they are bright, curious, and eager to help, which are traits I respect and value dearly in my chosen company.
However, I am often perplexed and frustrated with their heuristics and metrics. Almost nothing makes me more upset than someone trying to (in dead seriousness) justify anti-natalism and longtermism to me. Why the fuck are we trying to deny the fruition of life itself? If I was a religious figure perhaps 400 years ago, I would burn the anti-natalists at the stake for hearsay and crimes against the sanctity of God.
If I could force every EA to read this post at the threat of an incoming trolley, I would. Thank you for being a vessel for my frustrations and conflicting emotions about the movement.
Best,
Merlin Rojas
I'm with you on anti-natalism being deeply anti-human. The EA people I know are pretty passionately pro-natalist. But longtermism seems deeply pro-human. I mean, I'm sure there's some crazy thought experiment where you can wring monstrous conclusions out of it. I'm with Bryan that we should apply common sense to such cases. But caring about the long-term flourishing of humanity is common sense, isn't it?
I was hugely impressed by Toby Ord's The Precipice. It had what felt like a very common-sense case for longtermism, but mostly it was just pure pragmatics -- examining every possible human ~extinction risk over the next 100 years and meticulously working out how much to worry about each one. (Asteroid/comet 0.0001%, nuclear war 0.1%, runaway climate change <0.1%, natural pandemics 0.01%, engineered pandemics 3%, etc.)
If you don't want to commit to the whole book, Scott Alexander's review is great:
https://slatestarcodex.com/2020/04/01/book-review-the-precipice/
It occurs to me there's a possible misreading of the above parenthetical, like "none of these risks are worth much worry!" So I should clarify that I was picking particularly low-risk entries from Ord's list in order to establish credibility with anti-worriers like Bryan Caplan.
Meta-clarification: A 3% risk of extinction in the next 100 years from an engineered pandemic is nothing to sneeze at!
But of course dwarfing all of that is AI risk. I think Ord gained huge credibility in his 2019 book by not overhyping AI. He called AGI decades, not years, away. Even now his median for AGI is 2038. Still, his original estimate of the extinction risk over the next 100 (now 93 I guess) years was 10% and it's worth engaging with the argument directly, which I think Bryan will admit his blog post here does not do.
I'm very worried about the real-world harm of belief in bug welfare. I think AI with bug welfare values is one of the most likely scenarios in which AI would exterminate humanity.
I put 5% odds on AI exterminating humanity in order to create more bugs (or shrimp or computer pleasure machines etc.) at 10% some time between 2050 and 2150.
Brian Tomasik would tell you that as r-strategists, bugs tend to live short and miserable lives. Therefore we don't want to create more of them, we want to create less. And human civilization can be an effective way to do that.
"Convert Grass Lawns to Gravel to Reduce Insect Suffering": https://reducing-suffering.org/convert-grass-lawns-to-gravel-to-reduce-insect-suffering/
People aren't going to invest as much money in AIs that value bugs vs AIs that value humans.
So far that doesn't seem to be the case. EAs and EA-adjacent people dominate both OpenAI and Anthropic.
Bryan, I think your case is: EAs are likeable and have improved philanthropy, but a lack of common sense has led many into AI doom and bug welfare, and when a conclusion is absurd enough you should run the argument backwards.
I agree with a fair amount of what you write. EAs are prone to groupthink, partly because the network is small and dense, and EAs often treat common sense as a prior to update away rather than as evidence. Your point about reasoning from obvious premises to non-obvious conclusions is a good one.
However, I have three disagreements:
1. On going into debt. As Ben Todd says above, most put the risk at less than 30%, so there's a greater than 70% chance of having to repay, and many non-doom scenarios involve high returns to capital, which argues for saving and investing more. Paul Christiano is about 2x levered long the market.
2. On whether many EAs have 'lost interest' in global health. Of the roughly $2bn moved by the EA ecosystem in 2025, about half went to global health and development and about 30% to reducing global catastrophic risk.
3. On your point about reasoning from highly obvious premises to not-so-obvious conclusions. I think EA concern passes your test. Suffering is bad, whoever has it; whether something can suffer depends on facts about it, not its species; some insects behave as if they feel pain; there are a lot of insects. Conclusion: bug welfare might matter. Or: transformative technologies can do serious harm; AI is advancing fast; we don't fully control it; competition pushes deployment ahead of safety. Conclusion: there's a real chance it goes badly. The denials, that no insect can suffer and there's zero chance of AI catastrophe, aren't obvious at all. We can argue about how much weight small chances of big harms deserve. But that's a disagreement about expected value, not a failure of common sense.
"if you really thought the world was likely to end, you should be going into massive debt"
That does not seem to apply here as most doomers I think have a P(doom) well under 50%. If you think there are 10% chances AI will end the human race, you believe there is a 90% chance you will have to repay your debts.
But the argument that bug welfare matters does proceed from premises that are either (a) empirical, or (b) more obvious than the negated conclusion: "bug pain doesn't matter." Here it is.
1. The reason pain is bad is, at least in part, because of how it feels, not because of other features of the subject, like their intelligence or informational similarity to me. (Note: maybe those other factors do matter, and maybe they matter *a lot*!, but it does seem obvious that *at least some of the badness of pain* is due to the negative phenomenal quality of the painful experience).
2. Bugs can feel pain. They are, at least thinly, conscious and capable of negatively valanced qualia. There is something that it's like to be a bug in pain, and that experience has a negative phenomenal quality *of the same kind* as the one we have when we are in pain.
This is the empirical premise, and it might be wrong! But it's wrongness is a matter for science to figure out, not a matter for us to hand-wave away on account of it being inconvenient. At least for some candidates, like bees and ants, the evidence seems overwhelmingly to favor this premise.
3. Therefore, bug pain is bad.
The reason why you reject this is because of the unintuitive conclusion that bug pain, in aggregate, matters more than human pain, but that commits you to the strange position that bug pain can't be bad *because there are so many of them.* If there were just a few of them, you would maybe admit that their pain mattered.
But this is working backwards. Human pain doesn't get less bad as our population increases. Whether or not bug pain matters is a separate question from how many bugs there are, to be resolved first. If it turns out that bug pain really does matter, than the fact that there are so many bugs really does make it matter more.
You first need to explain why pain is universally a bad thing. I'm serious, and I doubt you can do it.
I was thinking the same thing. I question the premise that pain is bad. Pain has (evolved for) a purpose. To avoid behaviors and situations that shorten life and lessen the chance of reproduction.
You've snuck in Utilitarian premises, like that there are states of affairs ("welfare") which matter morally. A much more plausible alternative view is that conscious creatures have moral rights, but moral conduct is not about achieving the state of affairs which matters the most or has the most value.
Any plausible moral theory thinks welfare matters in some way. A main confusion among non-philosophers and even some philosophers who are non-specialists is that only utilitarianism can think that welfare matters. But that’s clearly false in the case of, e.g., Rossian deontology. Even Kant thought that suffering mattered, as far as I and AI know.
In practice, nearly everyone who uses the vocabulary of the previous commenter is a Utilitarian.
A majority, maybe, but philosophical arguments require decoupling those concerns.
EAs are often utilitarians because effective altruism follows more naturally from it. Utilitarians (at least almost) *only* value the things EA promotes, so it’s quite easy to come to the conclusion. But this is a sociological factor that we have to decouple from the analysis.
Pain and infliction of pain to other creatures is not immoral or wrong. Cruelty is infliction of needless pain so it is scaled to human need.
That bugs can feel pain--it is not empirical in the sense that pain itself is subjective. Pain can not be measured and hence science can not help.
Semi-off-topic, I wondered for a long time what exactly was animal cruelty. Is pushing a dog's nose into its pee cruel (quite aside from whether it works)?
The only simple and vaguely objective definition I could come with was "anything worse than life in the wild, which includes being eaten to death or starving to death after an injury". So don't stub out cigarettes on your dog. That sounds like a pretty sensible definition until you realize it forbids chaining dogs up in the back yard, and cooping up birds in cages and clipping their wings so they can't fly away.
I also had numerous side thoughts which amused me.
* Why is it moral for predators to eat prey alive, but immoral for ranchers to slaughter animals "humanely"?
* If humans slaughtering livestock for food is immoral, why is it moral to applaud wolves to eat livestock?
* Are ranchers who execute murdering wolves dispensing justice?
* Is it cruel for cats and other animal mamas to catch and use prey to teach their cubs how to hunt? Why can't humans keep pets and have bear baiting entertainment?
Cruelty is needless infliction of pain.
Now put that into legalese.
Yeah i like your definition of animal cruelty as "worse than what would have been in the wild"
By that logic tho, bryans arguments against animal welfare fall short though. I have no problem eating a chicken that lived free in the wild and died fighting for its survival. I do have a problem with one cooped up its entire life in a cage, pecking other chickens from frustration and injected with growth hormones to speed up its slaughter.
The bugs argument doesn't apply here because as far as I know, we don't (yet?) have bug farms. Its not eating meat that's cruel/unethical, it's the animal farming industry that has to reform
It was a fun thought experiment that lasted for several years, and convinced me that there can be no objective definition of animal cruelty. It can only be handled by customs and what neighbors or community put up with. I don't think legislation can do anything except inflame passions.
How can someone understand buying home or life insurance, to hedge against unlikely but expensive events, but not understand keeping your 401k intact instead of immediately spending it, to hedge against low probability case of AI stagnation and having a normal retirement? I give 80% p(doom) by end of 2035, but the 'we're saved by stagnation' case is still a long tail to prepare for.
It's not at all intuitive to me that animal suffering doesn't matter or is negligible, and it seems to have little to do with the pattern of behavior for why people eat meat, so I think I'm not alone.
Is it bad to kick a duckling? I'd say so. Why? Because it would harm my moral character by acting with cruelty? Implausible. If for some reason, kicking ducklings had no impact on how good I am as a person towards humanity, I would still not kick ducklings for fun. I also wouldn't kick a duckling for merely pragmatic reasons like getting from point A to B a little quicker. It's obvious to me that part of why it's wrong is that I'm causing pain to the duckling for no notable gains to myself. Likewise, I've noticed most people take some care to avoid causing needless harm to mammals and birds. This is odd if people genuinely think their lives count for very little.
The account for why most people seem to think eating meat is morally permissible is that there's something about the purpose that makes it okay. Something like "in most circumstances, what we're doing would be wrong, but because the purpose is consumption, it's justified."
Putting aside whether that's correct, this is completely divorced from why Caplan thinks it isn't morally wrong to eat meat. He just thinks animal interests are negligible. But it actually does matter to people whether people are putting pigs in crates they can't turn around in for food or for, say, display in a zoo, where they would want to see the pig be treated much better. Note: there was massive public backlash from attempts to roll back the ability of states to pass farm animal welfare laws recently. It's not just EAs who think even farm animals deserve better treatment than they get.
Here's something I do think most people consider unintuitive in the typical EA position: Peter Singer's thesis that animal interests count equally as human interests. But you don't actually need to accept that thesis to think animal welfare matters to some extent and that it's a worthy cause area for EAs.
I wrote a response. Feedback will be appreciated! Hopefully I have not misunderstood your claims. https://lovkush.substack.com/p/contra-bryan-caplan-on-effective
A whole $250?? Nobody is being remotely serious.
It's fine to reject insect welfare by an appeal to common sense and call it case closed. It's a moral question and you can just be a moral intuitionist or be a consequentialist and take it as axiomatic that only humans are moral patients. Maybe dogs and cats too, a little bit at least? But ultimately, yes, there's an arbitrary line to be drawn and common sense is key.
But AI risk isn't like that at all. We have to estimate probabilities of various trajectories and decide whether it's prudent to slow down frontier AI research. I agree the good trajectories are very good. But the bad trajectories are very bad, and as you say, you aren't making an argument against that.
Except the tu-quoque argument that self-consistent doomers would take on massive debt.
Bryan, it's pretty frustrating how thoroughly your commenters have rebutted that, and you brush it all aside with zero engagement other than to call it just-so stories. I'm happy to bite any bullets on my p(doom) of 20% by the end of the 2030s. I'm guessing that doesn't imply it's rational to effectively borrow money from you at a 21% interest rate, which is what I believe your bet with Greg Colburn amounted to. But whatever it does imply, I'm game!
Except there's another rebuttal: I have a high p(doom) *conditional on racing ahead* but I'm hopeful we'll succeed in slowing down.
But, as you know (recall our "Climate Shock" bet) I'm all about finding ways to distill disagreements into wagers, so let's keep trying to do that. Maybe you can identify lines in the sand that, if crossed, would change your mind about AI risk, and we can bet on those?
The AI-2027 scenario (published April 3, 2025) could be a source of ideas for such. On Friday I went through all their predictions through the end of 2026 to see how they're faring and the answer is uncannily well:
https://agifriday.substack.com/i/217451314/ai-2027-scorecard-as-of-september-2026
Short version: Effective Altruists' casuistry doesn't discredit their assessments of AI risk. For that we have to extrapolate the AI capabilities trend and anticipate the consequences of future superhuman AI.
I'm not a doomer in the sense of having any certainty about either the trend extrapolation or whether superhuman AI will spin out of control. I'm a doomer in the sense of lacking certainty in the other direction.
I'm happy to make wagers that that lack of certainty implies.