Open Letter to Young EAs

To voters: Please remember that the downvote button is not a disagreevote button. Under the circumstances here, it is mainly a reduce visibility / push off the homepage button. This is an attempt by an important critic to engage with the community. Although my reaction to the letter is mixed, I -- like Owen -- think it is offered in good faith and has some important points to make. I would hesitate before making it harder for others to encounter and decide for themselves whether to read. I would particularly hesitate before downvoting primarily because you didn't like the Wired article (I didn't either) rather than on the merits of this open letter.

To readers: In my view, more of the value is nearer the end, so if you're short on time / feeling frustrated by the first parts, you might skip ahead to "Feedback from reality" on page 12 of 17 rather than jettisoning the whole document. I'd also say that the content gradually moves away from / beyond what I recall of the Wired article and into more practical advice for young EAs.

Hi Leif, I appreciate your sharing this here, and hope that it reflects a willingness to engage in two-way communication (i.e., considering and engaging with our criticisms, as we have considered and engaged with yours). As I replied on twitter:

Something I found puzzling about the WIRED article was that it combined extreme omission bias (urging readers to vividly imagine possible harms from action, but not more likely harms from inaction) with an apparent complete lack of concern for the moral risks posed by your act of publicly discouraging life-saving philanthropy. For example, it's likely that more children will die of malaria as a result of anti-EA advocacy (if it persuades some readers), and I don't understand why you aren't more concerned about that potential harm.

It's unfortunate that you dismiss criticism as "extreme" rather than engaging with the serious philosophical points that were made. (Again, I'd especially highlight the status quo bias implicit in your "tests" that imagine harms from action while ignoring harms from inaction.)

The point of my anti-vax analogy is not name-calling, but to make clear how it can be irresponsible to discourage ex ante helpful actions by highlighting rare risks of harmful unintended side-effects. Your argument seemed to share this problematic structure.

These are substantive philosophical criticisms that are worth engaging with, not just dismissing as "extreme".

The first criticism feels pretty odd to me. Clearly what Singer, MacAskill, GiveWell, etc. are talking about is the counterfactual impact of your donation, since that is the thing that should be guiding your decision-making. And that seems totally fine and in accordance with ordinary English: it is fine to say that I saved the life of the choking person by performing the Heimlich maneuver, that Henry Heimlich saved X number of lives by inventing the Heimlich maneuver, that my instructor saved Y number of lives by teaching people the Heimlich maneuver, and that a government program to promote knowledge of the Heimlich maneuver saved Z lives, even if X + Y + Z + 1 is greater than the overall number of lives saved by the Heimlich maneuver. And if I were, say, arguing for increased funding of the government program by saying it would save a certain number of lives, it would be completely beside the point to start arguing that I should actually divide the marginal impact of increased funding to account for the contribution of Henry Heimlich, etc.

It looks like the total number of lives saved by all Singer- and EA-inspired donors over the past 50 years may be small, or even zero

This conclusion from the first half of the letter seems unjustified by the prior text?

You seem to be arguing that there's a credit allocation problem, where there's actually many actors who contribute to a bednet saving a life, but Givewell style calculations ignore this, and give all the credit to the donor, which leads to over counting. I would describe this as GiveWell computing the marginal impact, which I think is somewhat reasonable (how is the world different if I donate vs don't donate), but agree this has issues and there are arguments for better credit allocation methods. I think this is a fair critique.

But, I feel like at best this dilutes the impact by a factor of 10, maybe 100 at an absolute stretch. If we take rough estimates like 200,000 lives saved via GiveWell (rough estimate justified in footnote 1 of this post), that's still 20,000 or 2,000 lives saved. I don't see how you could get to "small or even zero" from this argument

If you publish a bad pieces and share them with millions of people, I don't really feel obliged to talk or listen to other things you write until you correct the inaccurate piece. I don't think any other community would and I think it's a bad use of our time to extend this absurd level of charity. 

People are free to tell me the wired article wasn't inaccurate or lazy, but scanning it, it looks that way.

Here are quote I could find in 15 minutes from your first article that leave the reader with an inaccurate impression. I have not read this new article.

• "Elon Musk has said that EA is close to what he believes" - has Musk acted on these supposed beliefs or is this just guilt by association?
• "comparable to what it’s estimated the Saudis spent over decades to spread Islamic fundamentalism around the world" - I can find many things that cost $46 but I note that you chose a terrorist ideology
• ""Insecticide-treated bed nets can prevent malaria, but they’re also great for catching fish. In 2016, The New York Times reported that overfishing with the nets was threatening fragile food supplies across Africa." " - my sense is that this is widely debunked. As a result of your article it was shared by Marc Adreessen. As you yourself note we should count harms as well as benefits. I could this as a harm to what I am confident is an effective way to stop malaria.
• 'In a subsection of GiveWell’s analysis of the charity, you’ll find reports of armed men attacking locations where the vaccination money is kept—including one report of a bandit who killed two people and kidnapped two children while looking for the charity’s money.' " - I think it's a bit absurd to imply that the norm is to count this stuff as the costs of aid. Perhaps it should be (and it's good that givewell mentioned it) but the implication that they are unusually bad for not doing so? 

I could go on. 

Leif, we do not owe you our time. You had the same social credit that all critics have and a large platform. You could have come here and argued your case. I am sure people would have engaged. But for me, you have burned that credit, sharing inaccuracies to millions of people. Your piece started a news cycle about the harms of bednets based on inaccurate information. That has real harms. So I don't care to read your piece.

I don't know whether I am the hero in my own story - I have done many things I regret - but I do know a thing or two about dealing with those I disagree with. I would not publish a piece with this amount of errors and if I did I wouldn't expect people to engage with me again. I do not understand why you think we would.

I hope you are well, genuinely. 

The criticism of the concept of "effective altruism," and the second main criticism to the extent that it's related to it, also feels odd to me. Altruism in the sense of only producing good is not realistically possible. By writing your Wired article, say, it is overwhelmingly likely that you set off a chain of events that will cause a different sperm to fertilize a different egg, which will set off all sorts of other chains of events meaning that in hundreds of years we will have different people, different weather patterns, different all sorts of stuff. So writing this will cause untold deaths, untold suffering, that wouldn't have occurred otherwise. So too for your friend Aaron in the Wired article helping the people on the island, and so too for anything else you or anyone might do.

So either altruism is something other than doing only good, or altruism is impossible and the most we can hope for is some kind of approximation. It wouldn't follow that maximizing EV is the best way to be (or approximate being) altruistic, but the mere fact that the actions EAs take are like all other actions in that they have some negative consequences is not in itself much of a criticism.

Imagine a relay race team before a competition. The second-leg runner on the team thinks—let us assume correctly—‘If I run my leg faster than 12 seconds, then my team will finish first; if I don’t, then my team won’t finish first.’ She then runs her leg faster than 12 seconds. As the fourth-leg runner on her team crosses the finish line first, the second-leg runner thinks, ‘I won the race.’ Is she right?

Yes, of course she's right. Even if she's the weakest member of the team. They don't give Olympic relay teams 1/4 of a medal each.

-

For the record I don't describe myself as an EA and don't really hang out in EA circles. I'm far too old to be susceptible to arguments that I'm going to save the world with the power of my intellect and good intentions. If the founding fathers of EA's bios are accurate I discovered Peter Singer's solution to world poverty slightly before them, thought he had a [somewhat overstated] point and haven't done anywhere near enough to suggest I absorbed the lesson. I think utilitarianism's utility is limited but don't have the academic pedigree to argue about it for any length of time, and I think a lot of EA utilitarian maths is a bit shoddy.^[1] So I don't think I'm making a particularly partisan argument here.

But you aren't half leading with your weakest arguments^[2] GiveWell's estimation that if x bednets are distributed, on average about y% Malawian mothers receiving the nets will succeed in using them to protect their kids, so z% fewer kids will die isn't stealing credit from Malawian mothers or Chinese manufacturers in a zero sum karmic accounting game, it's a simple counterfactual (with or without appropriate sized error bars). Or put another way, if a Malawian kid thanks her mother for going hungry for two days to pay for a malaria net herself,^[3] the mother shouldn't feel obliged to say "no, don't thank me, thank the Chinese people that manufactured it and the supply chain that brought it all the way here, and the white Westerners for doing enough research into malaria nets to convince vendors in my village to stock it."  The argument that installing a few more stakeholders in the way introduces a qualtitative difference between donating and diving into a pond might make Peter Singer's thought experiment a little bit trite, but it isn't an argument against the quantitative outcomes of donating at all.

1. ^{^}

   in particular, the tendency to confuse marginal and average costs and wild speculative guesses with robust expected value estimation. I don't actually think this is bad per se: people overestimating how much their next fiver can help a chicken or prevent Armageddon certainly isn't worse than people overestimating how much they want the next beer. I just think it looks a lot like the "donor illusion" certain leading EAs used to chastise mainstream charity for; actually the average "child sponsorship" scheme is probably more accurate, in accounting terms, about how much your recurring contribution to the charity pool is helping Jaime from Honduras than many EA causes. (I guess not liking that type of charity either is where you and the median EA agree and I differ :)) 

2. ^{^}

   Judging by your book reviews, you've researched sufficiently to be able to offer more nuanced criticisms of development aid. So I'm not sure why you'd lead with this, or in other articles with anecdotes how about profoundly the whinging of a single drunk teenage voluntourist crushed your dreams of changing the world. It's not even like there aren't much better glib criticisms of EA or charity in general....

3. ^{^}

   maybe because donations dried up...

Disclaimer: My comment is less concise, pondered, and polished than I'd hope for under normal circumstances (for personal/family reasons) but I think the pros of substantive engagement outweigh the cons (and there hasn't been a ton of substantive engagement).

In short:

• I continue to essentially disagree with Leif's criticisms of GiveWell, although some observations about moral credit are worth reiterating in my view;
• I think there's something to Leif's arguments insofar as EA is viewed as something like a Grand Unified Theory of altruism, but they are less compelling against (my own) view of EA as something we do on the margins because we think the EA tools are underutilized in the altruistic world as a whole;
• I generally agree that young EAs should devote more time and energy toward broadening experiences that promote what Leif terms "good judgment;" and
• While endorsing the importance of asking whether one knows enough to be making recommendations, I would place more emphasis on the costs of inaction than I think the Open Letter does.

On Moral Credit and Counterfactuals

• What I liked: I think the emphasis on counterfactuals in EA creates risks of giving donors too much moral credit / patting ourselves on the back too much for donating if we are not careful. As a (small time) donor, I get the privilege of being a counterfactually important part of a team that gets a bednet over a child's head and (given enough bednets) prevents a fatal malaria infection. There's a lot of credit to share, and it is not virtuous to claim it all for oneself. I think there is value to reiterating that to guard against people implicitly equating counterfactual impact with being the sole cause.
• What I didn't like: I don't think a reasonable prospective donor reading GiveWell's materials would get the impression that they were solely responsible for saving a child's life by donating $5000 and that no one else gets moral credit for that. The counterfactual impact is important to convey; because someone donated ~$5,000 to bednets instead of to opera, a child in a developing country lived who wouldn't have otherwise made it. There are ways to manipulate impact that mislead donors; I generally don't see evidence of that here.^[1] Finally, all charity involves working with a team; even if it were somehow possible to be a truly solo altruist it  probably wouldn't be pretty!
• I often analogize to what I would expect out of a non-EA organization in an attempt to avoid both demands for extra rigor or giving the EA org a pass. If a university were pitching a donor to endow a new chair in the philosophy department, I wouldn't expect them to remind me that my donation would be worthless but for (e.g.) the ancient Greeks, the institutions that trained the new hire, the university's general counsel, and the students.
• What would move the needle: I can't get back into the headspace of a new GiveWell donor at this point. If someone ran a study of people new to GiveWell, and a significant number of people came away believing something like the donor was the sole cause of saving someone's life, that would make me think that GiveWell was unreasonably communicating to donors here. 

On Non-Exceptionalism

At least in global health, I think much of what EAs are doing is broadly in line with the broader global health movement. For instance, the IFRC explains:

• These [insectide-treated bednets] are estimated to be responsible for two-thirds of the reduction in malaria cases over the past decades. Thanks to the efforts of national malaria programmes and partners, about 68% of households across sub-Saharan Africa own at least one net. Most of these nets have been bought via funds from The Global Fund to Fight AIDS, Tuberculosis and Malaria, the United States President’s Malaria Initiative, UNICEF and the Against Malaria Foundation (AMF).  

If the list of actors who don't know enough to be doing anything in this field include the Global Fund, the US Government, and UNICEF too, then the problem goes way deeper than EA!

Likewise, looking at Vitamin A supplementation, my first Google hit was Nutrition International, an organization whose work (in general) is sponsored by Gates, UNICEF, the governments of France, Australia, Canada, and the UK, and others.

So my looking for sources outside EA (as the Open Letter suggests) produced reasons to believe these are not things EAs are doing on half-baked EA theories. In contrast, I've read enough about the (copyrighted yet publicly available) Scientology scriptures to know that there are relatively few things in life that can be fairly compared to Scientology. 

Grand Unified Theories vs. Plays on the Margin

I think there's a difference between EA as something like a Grand Unified Theory (GUT) of altruism, or even a GUT of global health/wellbeing altruism, and EA on the margins (a belief that more of the charitable pie should be distributed in general accord with EA principles than is currently the case). 

• Certainly some philosophical writings tend to set EA up as a GUT, and so critiquing EA-as-GUT is a totally fair exercise. However, EA controls only a tiny fraction of even US charitable spend, and a small fraction of charitable spend on global health/wellbeing (especially compared to governments).
  • For an ordinary practitioner / donor in many cause areas, philosophical arguments about EA-as-GUT may be interesting but are of relatively little practical import.
  • Evaluating EA-as-something-like-a-GUT may be more important where a substantial fraction of money or other inputs coming into a field show strong EA influences (e.g., farmed animal welfare).
  • Moreover, while I think most social movements have a tendency to view themselves as GUTs on their topic of interest, it is probably even more likely for people in the EA community to think that math-y spreadsheets are a GUT. One can believe spreadsheets are important while also believing that there are more things under heaven and earth than are dreamt of in EA spreadsheets (and that having 100% of charitable spend flow from current EA methodology might not be such a grand thing!).
    • Therefore, while my view of EA / math-y spreadsheets is more favorable than the Open Letter's, I think there is value in reiterating the problems with treating it as a GUT.

Baskets and Tools

Let me now offer a point that’s really important, especially if you’re in a STEM field and have never heard it before. There are many definitions of ‘rational action.’ EA is centered on one understanding of ‘rational action’ but there are lots of others.

So why believe that EA’s particular understanding of ‘rationality’—just one amongst many—is the best one? Why believe that thinking in terms of EV, marginality, and so on really is a way of always being smart, instead of a way of sometimes being dumb?

I think there may be something here but would put a somewhat different spin on it. Flaws are something I expect to see in the EA paradigm and in every other altruistic paradigm. I suggest that putting all of the world's altruistic eggs into one basket would be an inherently perilous exercise, and so the question is more "are we putting too many / too few eggs in this altruistic paradigm's basket"? 

There's a tension here -- on the one hand, I think we want to have specialist communities rather than having everyone who seeks to do altruism with a shallow knowledge of how to use three dozen types of tools. I want my urologist to have taken a deep dive into urology, but I also want him to know enough about other ways of practicing medicine to know when they don't have the right tools for the job. They also need to know enough about other ways of practicing medicine to be a good urologist.

• Another analogy: Hammers are important tools, but not every problem is a nail. Moreover, other tools are often necessary to take action on a nail. I do think there is a tendency for EA (and probably most other altruist movements) to see ambiguous hardware as nails. Someone with a toolbox of almost exclusively hammers is not going to be as useful in the world as someone who has at least a basic appreciation for and ability to use other major hand tools.

Technicians

In American medicine, we have physicians who have very broad general training plus years of onerous training in their specialty, as well as "midlevels" (like nurse practitioners) who have much less education and training. From what I hear,^[2] midlevels can be very good when operating in their comfort zone (especially if narrowly specialized) but often lack the ability to respond well to curveballs. I do worry that the current recommended "education" of EAs veers too much toward creating narrow technicians rather than the equivalent of residency or even fellowship trained physicians. Cf. the Open Letter's discussion of "good judgment." 

To be clear, midlevels have an important role to play in the US healthcare system; producing residency-trained physicians is extremely expensive. I didn't need an MD when I went in this week for a suspected sinus infection. The problems often come in when midlevels forget that they are midlevels and lack the broad training and experience of residency-trained physicians, so they take on workloads beyond their capacities.

I would guess that, in this analogy, Leif thinks most EA leaders are midlevels who are think they are fellowship-trained physicians. This response is already long, so I'm not going to evaluate that possibility further here.

Finally, I do think that both EA and broader Western society tend to undervalue diverse life experience. I also think the tech industry as an exemplar has some of the pitfalls that the Open Letter alludes to. I would generally endorse Leif's recommendations about good judgment insofar as they are not particularly costly. If I knew that someone had been preselected to be a philosopher-king, I'd be more inclined to endorse the more costly parts too. But I feel a bit of unease with globally recommending that young people pour a lot of time and energy into stuff that the broader society doesn't seem to value that much.

The Costs of Inaction

This is a critical flaw in EA leadership and in the culture they’ve created. The crucial-but-absent Socratic meta-question is, ‘Do I know enough about what I’m talking about to make recommendations that will be high stakes for other people’s lives?’ 

This is an important question, but I don't think it can be asked without considering other actors and the costs of inaction. If we could enforce a global norm that no one should be doing anything in AI without being able to answer this question in the affirmative (and have their answer verified by disinterested outside experts), that would be simply amazing. In fact, one can view the idea of AI policy as directed significantly toward that goal.

One problem is that other actors aren't asking this question. Tyson Foods (the meat conglomerate) is going to do factory-farm-conglomerate things like their executives' paychecks depend on it (and they do). Even laying aside EA-type AI safety worries completely, the financial interests of big AI companies and their employees are a powerful incentive to moving fast even though they don't know nearly enough about the civilization disrupting and defining functions AGI would bring. So recommendations and actions will happen whether we are making them or not.

EA often works in neglected fields, and I think that can play into the calculus as well. A recommendation based on sketchy information isn't necessarily inappropriate if that's the information that is on the table.

So I would phrase the question differently: In light of both the possibilities and opportunity costs of gaining better knowledge, and the effects of other actors who would occupy the field in my absence, do I know enough about what I'm talking about to make high-stakes recommendations?

1. ^{^}

   For instance, touting impact on an organization's most effective program even though it can only absorb a portion of the funding obtained as a result of the appeal.

2. ^{^}

   This is a simplified example, and I've heard about this mostly from the physician standpoint rather than the midlevel one.  

This comment is mostly about the letter, not the wired article. I don't think this letter is particularly well argued (see end of article for areas of disagreement), but I'm surprised by the lack of substantive engagement with it. 

This is fairly rough, i'm sure i've made mistakes in here, but figured it's better to share than not. 

Here’s some stuff i think is reasonable (but would love for folks to chime in if i'm missing something) 

• Questioning GiveWell's $4500 estimate - seems worth questioning! I am no expert in developmental economics, but it seems like Leif made some reasonable points regarding how to measure counterfactuals, Shapley values, the other non-AMF actors who helped get the bednets from conception to creation. Maybe all these points have been covered in places, but I'd be really surprised if everyone reading and engaging in his writing knows the answers (at least for myself, I don't!)
  • I'd be curious how many of people who donate to GiveWell take this number at face value / how many of us have really thought critically about this number
  • My friend with a background in dev econ says that EAs seem to take the $ estimates a lot more seriously than development economists in general. That seems worth reflecting on and learning more about. 
• He suggested GiveWell ask external developmental economists to evaluate their research / impact
  • this sounds great! +1 to this. 
• I think his red flags were all reasonable and things I'd be on the lookout for (the red flags were: seeking confirmation from within the group, attacking credibility of outsiders, distorting criticisms/vilifying critics,  ‘Am I afraid that I’ll lose friends if I question certain things about EA?’)
  • I don't think all of those things are all as big risks as Leif perhaps does - but I do think that these things are present to some degree, and we should take them seriously.
  • I think, on average, we should be encouraging people to form independent beliefs and testing their theories on the world (and getting it wrong! and learning from the mistakes! and not worrying so much about maximizing at all costs!) much more than we currently do.
  • Also, others have written with similar perspectives (e.g. On Living Without Idols  and It's ok to leave EA )
• I am sympathetic to his claim that there is a lot of focus on intermediate steps for LT & meta impact, but that's not really the claim he's making.
  • However, GH&D and animal orgs do often communicate actual impact. You can always argue that they calculated the impact incorrectly - but I don't think they look at intermediate steps.
• i think the nerd scouts suggestion is understandable given that that the Atlas fellowship was a very prominent high school outreach program, and a very large % of CB resources go to uni outreach, and Leif is a Stanford professor where there has been a lot of EA activity for many years.
  • I think this section could have been written with a bit less attitude
  • I am actually pretty bullish on getting younger people (college/ late high school) to really focus more on scout mindset / general decision-making tools & solid career advice. IDK if Leif is suggesting exactly that, but I do think that's pretty important
  • Otherwise, I think I mostly disagree with him on his higher level strategy. It would be hard to know from the outside, but a number of projects have been started doing mid- and late-career outreach that are pretty exciting, and I hope more will be started in the coming years.  
• I read GiveWell's response to the wired article, and I do think overall it's not as direct as it could be.
  • both he and GW are weirdly not addressing the fishing thing directly - GW did not directly link their 2016 response in their notes (it's linked from the info page on bednets, but you have to search for it), and Leif didn't in either of his posts. am slightly confused on both parties' actions here. 

Things I'm not very sympathetic to in his argument: 

• overall, there are a lot of points, and not all of them are made equally well
  • I think if any of the points i mention above were the focus and written with a tiny bit more rigor, they would be valuable additions to discourse
• little quibbles
  • Comparing EA to scientology. Feels like there were better and more fair analogies. I think overall this is a theme with his writing, which is unfortunate because it upsets people.
  • some of the narrative about lots of money / EA funding lots of prizes and competitions would have been more relevant pre-FTX collapse)
• The effective vs altruism section feels a bit too abstract to be useful to me, and similar arguments have been made within the community numerous times, especially post FTX. But, fair critique given FTX. (Maybe there's a lesson to be learnt here about how these ideas are perceived externally, clearly there's a lot of comms work we need to do)
• the tone (of the article more so than the letter) is a bit aggressive - Leif is aware of this but i disagree with his reasons. I think he could have made a much stronger and more convincing argument without being that aggressive. In fact, if such an article existed I would share it widely and want to engage in discussions on it !
• didn't feel like a very balanced view of the impact EA has had - he does mention that farmed animal welfare + lead projects, but i think he could have done a little more digging, especially when making some really deep critiques. I would be curious as to what he thinks when looking across the board at what has happened (another comms improvement note, perhaps) 

I was surprised to kind of like this letter (I've not read the Wired article people complained about). I disagreed with a good amount of it, including the overall conclusions, but:

1. I felt it was earnest, and I appreciated that.

2. I think it's asking a number of good questions. EA as a frame doesn't naturally generate some of these questions, but it's generally healthy to look at things from lots of angles -- and I suspect that in particular that in places it's managing to highlight what may be relative blindspots of EA: and it's great to have attention on what the blindspots may be, even if on net you don't think they're necessarily worth correcting.

I'd feel kind of good if this was included in some reading lists for people exploring EA? (Probably along with a good reply that I expect someone will write.)

Two separate points:

1. I am one of those people who, having seen the Twitter post with the letter, scanned the Forum home page for the letter and didn't see it! And regardless of what you think of the letter, I think the discussion in the comments here is useful; I am glad I did not miss it. So I agree with what others have said—there are real downsides to downvoting things just because you disagree with them; I would encourage people not to do this. (And if you downvoted this because you don't think a Stanford professor making a sincere effort to engage with EA ideas is valuable/warrants engagement then... yeah, I just disagree. But I would be eager to hear downvoters' best defense of doing this.)
2. Regarding the letter itself: one thing I am struck by is the number of claims in this letter that go without citations. This is frustrating to me, especially given the letter repeatedly appeals to academic authority. As just one example, claims like "It has lots of premises that GiveWell says depend on guesswork, and it runs against some of the literature in fields like development economics" warrant a citation—what literature in development economics?

Leif - thanks for sharing this! I appreciate your explanation of the Wired article, which frankly did convey the impression, for example, that you were arguing for a certain method of aid delivery.

In the letter, you provide some 'false objections' to watch out for.

Could you please provide what you think would be the strongest objections to your arguments, and what your response in turn to those would be?

Would you be up for making your "deaths of effective altruism" article available in a way that isn't paywalled?

Thanks for being willing to share criticism!