I've promoted Effective Altruism in the past. I will probably continue to promote some EA-related projects. Many individual EAs are well-intentioned, talented, and doing extremely important, valuable work. Many EA organizations have good people working for them, and are doing good work on important problems.
That's why I think Sarah Constantin’s recent writing on Effective Altruism’s integrity problem is so important. If we are going to get anything done, in the long run, we have to have reliable sources of information. This doesn't work unless we call out misrepresentations and systematic failures of honesty, and these concerns get taken seriously.
Sarah's post is titled “EA Has A Lying Problem.” Some people think this is overstated. This is an important topic to be precise on - the whole point of raising these issues is to make public discourse more reliable. For this reason, we want to avoid accusing people of things that aren’t actually true. It’s also important that we align incentives correctly. If dishonesty is not punished, but admitting a policy of dishonesty is, this might just make our discourse worse, not better.
To identify the problem precisely, we need language that can distinguish making specific assertions that are not factually accurate, from other conduct that contributes to dishonesty in discourse. I'm going to lay out a framework for thinking about this and when it's appropriate to hold someone to a high standard of honesty, and then show how it applies to the cases Sarah brings up.
(Disclosure: Sarah's a close friend. I've previously worked at an EA organization that isn't mentioned here, though I have no current affiliation with it.)
Honesty, dishonesty, and lies
Let's look at an example from outside Effective Altruism. Slate Star Codex writes about a New York Times article that claims economists "don't buy" claims of the benefits of education vouchers:
From today’s New York Times: Free Market For Education: Economists Generally Don’t Buy It:
[…] Only a third of economists on the Chicago panel agreed that students would be better off if they all had access to vouchers to use at any private (or public) school of their choice.
36% of economists agree that vouchers would improve education, compared to 19% who disagree. […]
A more accurate way to summarize this graph is “About twice as many economists believe a voucher system would improve education as believe that it wouldn’t.”
By leaving it at “only a third of economists support vouchers”, the article implies that there is an economic consensus against the policy. […] Its title is “Free Market For Education: Economists Generally Don’t Buy It”. But its own source suggests that, of economists who have an opinion, a large majority are pro-voucher.
I think this is really poor journalistic practice and implies the opinion of the nation’s economists to be the opposite of what it really is. I hope the Times prints a correction.
Noah Smith replies that the article was fine because it contained no false factual assertions:
A correction!! Of course no correction will be printed, because no incorrect statements were made. Dynarski said that economists are "far less optimistic" about vouchers than about Uber/Lyft, and this is true. She also reported close to the correct percentage of economists who said they supported the policy in the IGM poll ("a third" for 36%).
Scott is upset because Dynarski left out other information he considered pertinent - i.e., the breakdown between economists who were "uncertain" and those who "disagree". Scott thinks that information is pertinent because he thinks the article is trying to argue that most economists think vouchers are bad.
Scott Alexander thinks the article was deceptive. Noah Smith thinks the article was basically accurate. I think they're applying different standards of honesty.
Factual accuracy vs perjury
Smith appears to be applying the factual accuracy standard of honesty. If your sentences, taken literally, describe a true fact about the world, then they're admissible. If not, not. This is the standard by which Hillary Clinton appeared to be honest, according to her supporters, while Donald Trump did not. When is this standard good enough to lead us to the truth?
A mere factual accuracy standard does not work if you cannot count on a comprehensive presentation of the relevant information. Typically not all the evidence will be on one side of any interesting issue, even ones with a clear answer. If you only listen to one side, a clever arguer can present only the true facts that favor one side, resulting in strongly filtered evidence:
Yesterday I discussed the dilemma of the clever arguer, hired to sell you a box that may or may not contain a diamond. […] What if the only evidence you have is the word of the clever arguer, who is legally constrained to make only true statements, but does not tell you everything he knows? Each statement that he makes is valid evidence—how could you not update your probabilities? […] According to Jaynes, a Bayesian must always condition on all known evidence, on pain of paradox. But then the clever arguer can make you believe anything he chooses, if there is a sufficient variety of signs to selectively report. That doesn't sound right.
Most legal processes work on the theory that every case has exactly two opposed sides and that it is easier to find two biased humans than one unbiased one. Between the prosecution and the defense, someone has a motive to present any given piece of evidence, so the court will see all the evidence; that is the theory. If there are two clever arguers in the box dilemma, it is not quite as good as one curious inquirer, but it is almost as good. But that is with two boxes. Reality often has many-sided problems, and deep problems, and nonobvious answers, which are not readily found by Blues and Greens screaming at each other.
An example of a system where statements are regularly held to this standard is the court system in the US and similar countries. This standard works in the context of such an adversarial system: both sides of the argument get to make any arguments they like, point out flaws in the other side's case, and there's a third party who's explicitly not an advocate, focused on nothing but making sure both sides get a fair chance to present their case, blocking irrelevant shenanigans. Then there's a decision, it's recorded, the matter's settled, and any time the matter comes up again, it's an uphill battle to relitigate things without new evidence.
In a system like this, no one party to the argument bears the burden of bringing up all the relevant facts, or ensuring that each fact is contextualized adequately – as long as witnesses limit themselves to assertions that are literally true, if the facts are selectively represented to favor one side, the other side has both the incentive and the capacity to expose this to the court.
In criminal cases (but not civil ones), the burden of proof is on one side – but even so, the standards of admissibility of evidence and argument are applied symmetrically. Nor can anyone game the system by simply relitigating the same case over and over – the avenues for appeal are finite, there are penalties for legal actions that the court deems frivolous, and at the end of the process some court makes a decision that determines the outcome.
Smith appears to be arguing that the New York Times did not perjure itself. This is true, but the New York Times was not arguing in an adversarial venue in which a reader can count on seeing the other side present counterarguments. It does not present itself as an advocate in a court – it presents itself as a source of reliably informative, objective reporting. It claims that it is going to give you not just the news that favors its side, but all the news that's fit to print.
Alexander's point seems to be that there is some highly relevant news, which is fit to print, that has been omitted. In this case, the New York Times has not in fact presented all the news that's fit to print, belying its ancient motto.
(UPDATE: Empirically, it seems that most casual readers drew the false conclusion that Alexander predicted they would.)
Intent to inform
While adversarial processes help alleviate this problem, they only reliably lead towards the truth if all sides are represented. If one side is entirely omitted, then you will reliably have a point of view slanted against that side. Similarly, if an asymmetrical burden of admissibility is imposed, holding one side to a higher standard than another, then we will see less evidence in that direction.
To the extent to which we are looking at arguments where we don't get all the relevant sides, then we have to rely on some other guarantee that the discourse will converge towards truth. We will have to trust that the parties presenting us with information are doing so with an honest intent to inform us about relevant evidence on all sides.
When you cannot verify every decision someone is making, and are relying on their intent to cooperate, the things you can observe and verify are evidence about their underlying intent. For this reason, once you observe someone selectively withholding or deemphasizing some information, you should trust them less, even if they correct the specific issues you can observe. If you are satisfied with corrections only of the misrepresentations you can verify, you will simply teach them how to be better clever arguers. What you're looking for is an affirmative effort to honestly inform, such that they seem to be trying even in cases where you wouldn't have caught them:
Expending effort to infer others' preferences, needs, and boundaries, and how your actions affect these, is a classic "costly signal" since it's something a friendly person would want to do anyway, and something an unfriendly person would not want to do without norms that force it. […]
An important cost here is that making a scene harder for outsiders to exploit maliciously also makes it harder for harmless strangers to interface with, and imposes some cognitive overhead on all participants. Probably for people sufficiently acculturated (and not neurodivergent in certain ways) the training creates a passive process where they don't even see how they could do otherwise, which is fairly low-effort as long as they don't have to code-switch or do anything unusual. […]
Explicit communication about preferences can be very valuable for bridging cultural gaps, or simply misaligned expectations, and often what we need is simply to tell each other what's wrong. But if someone crosses boundary after boundary with no apparent attempts to generalize, it doesn't matter whether they're malicious, or simply very much unlike you. If the other person isn't able to do some of the work towards being decent towards you by your standards, you can't explicitly tell them how. Not mustn't - can't. There's no litmus test for "try to infer my preferences, rather than directly optimizing on the metrics I can observe directly," unless you set up bullshit tests.
Case studies of EA's trustworthiness problem
I've distinguished mere factual accuracy from the intent to inform. Before I consider the cases Sarah brings up, I want to lay out another distinction. There are two ways to interpret quoting a case of person X saying Y as an example of a lying problem:
- Y shows that person X lied.
- Y is either a direct contribution to, or evidence of, deception-friendly norms.
The first claim is often dubious, but the second often seems right to me. I don't have an opinion on whether this constitutes "a lying problem," so I'm trying to use more precise language here.
A tax on criticism
The first example Sarah brings up is an EA arguing that criticism should be held to a higher standard than other discourse:
Topics like this are sensitive and complex, so it can take a long time to write them up well. It’s easy to get misunderstood or make the organisation look bad.
At the same time, the benefits might be slight, because (i) it doesn’t directly contribute to growth (if users have common questions, then add them to the FAQ and other intro materials) or (ii) fundraising (if donors have questions, speak to them directly).
Remember that GWWC is getting almost 100 pledges per month atm, and very few come from places like this forum. More broadly, there’s a huge number of pressing priorities. There’s lots of other issues GWWC could write about but hasn’t had time to as well.
If you’re wondering whether GWWC has thought about these kinds of questions, you can also just ask them. They’ll probably respond, and if they get a lot of requests to answer the same thing, they’ll probably write about it publicly.
With figuring out strategy (e.g. whether to spend more time on communication with the EA community or something else) GWWC writes fairly lengthy public reviews every 6-12 months.
He also said:
None of these criticisms are new to me. I think all of them have been discussed in some depth within CEA.
This makes me wonder if the problem is actually a failure of communication. Unfortunately, issues like this are costly to communicate outside of the organisation, and it often doesn’t seem like the best use of time, but maybe that’s wrong.
Given this, I think it also makes sense to run critical posts past the organisation concerned before posting. They might have already dealt with the issue, or have plans to do so, in which posting the criticism is significantly less valuable (because it incurs similar costs to the org but with fewer benefits). It also helps the community avoid re-treading the same ground.
In other words: the CEO of 80,000 Hours thinks that people should “run critical posts past the organization concerned before posting”, but also thinks that it might not be worth it for GWWC to address such criticisms because they don’t directly contribute to growth or fundraising, and addressing criticisms publicly might “make the organization look bad.”
This cashes out to saying “we don’t want to respond to your criticism, and we also would prefer you didn’t make it in public.”
It’s normal for organizations not to respond to every criticism — the Coca-Cola company doesn’t have to respond to every internet comment that says Coke is unhealthy — but Coca-Cola’s CEO doesn’t go around shushing critics either.
Ben Todd's suggestions are good general practice for producing high-quality commentary on organizations' internal affairs. If I were to write something up about 80K’s internal practices, I would want to give 80K plenty of time to take a look at it first, even if it weren’t mainly critical. I would do this because I would prefer not to be embarrassingly wrong.
Similarly, I try to let people and organizations know if I plan to quote them or describe their public conduct, in case there’s context I’m missing, or they want to point me to more recent public information on the topic.
The problem comes when this standard is applied to critics but not to supporters of EA organizations. This is effectively a tax on internal criticism of EA. If you ask that we impose a higher burden on criticism than on praise for you or your organization, you are proposing that we forgo the benefits of an adversarial system, in order to avoid potentially damaging criticism. If we forgo the benefits of an adversarial system, we can only expect to come to the right answers if the parties that are presenting us with information exhibit an exceptionally honest intent to inform.
If you ask people to hold criticism of you to a higher standard than praise, you are either asserting a right to misinform, or implicitly promising to be honest enough that a balanced adversarial system is not necessary. You are promising to be a reliable, objective source of information, not just a clever arguer.
If you're asserting a right to misinform, then it is clear enough why people might not want to trust you.
If, on the other hand, you're promising not to be a clever arguer – if you're promising to be so reliably objective a provider of information that a tax on critics doesn't lead to misinformation - then you are promising to be exceptionally forthcoming about evidence against your views. But in that case, you're promising not to be too busy to present and address potential criticism that your audience is likely to find relevant.
I am not accusing Ben Todd of deliberately, consciously claiming the right to lie, on behalf of Giving What We Can. I don't think this reflects especially poorly on Ben Todd personally – I think he sincerely thinks that his proposed policies cause better outcomes, and that these views are not exceptional in the EA community. I think that he should continue to accurately represent his actual opinions and do what he thinks is best. I do hope that this causes him to either reconsider and change his opinion, publicly – or explain to me what I'm missing and why I'm wrong.
What I am saying is that he has asked us to hold criticism of EA organizations to a higher standard than support of EA organizations, and has also implied that we should excuse these organizations from proactively addressing potential criticism, in cases where this doesn't directly contribute to growth or fundraising. Taken together, this amounts to a claim that EA organizations have the right to distort the public discourse about them to further their organizational aims.
Distortion of evidence
Sarah discusses the case of Animal Charity Evaluators heavily implying that leafleting has a strong evidence base as an effective animal welfare intervention, when in fact there's no experimental evidence for it, and some evidence against. On the impact tab on their leafleting page they wrote, "the existing evidence on the impact of leafleting is among the strongest bodies of evidence bearing on animal advocacy methods," which might lead a reader to conclude that there is strong evidence for this intervention. Elsewhere they admit that the evidence is extremely weak.
This problem has been pointed out publicly multiple times. (For an example, see Aceso Under Glass's 2015 post on the subject.) ACE staff have repeatedly said they understand the concerns and are revising the page.
ACE finally, after Sarah's post sensationalized the issue, added a disclaimer to the impact page to this effect. But the problem is not that this particular page is telling a very harmful lie, it’s evidence about a lack of intent to honestly inform.
On a Facebook thread on the topic, an ACE board member asked for suggestions for what ACE might do to repair this breach of trust. Here's how I responded:
I'm not going to propose a bunch of hoops ACE can jump through to make things better here. Just like the problem isn't that ACE got caught here, the problem can't be solved by a bunch of easily verified things.
I think ACE staff should sit down and have an honest internal discussion about why it seemed fine to ignore this issue for so long, and why the controversy's surprising - if it is in fact surprising. They should seriously think through the perceived conflict between prioritizing work that promotes the cause, and reassess whether perhaps they've underestimated the costs of overenthusiastic framings. And then they should do whatever they think is right after that.
If they end up deciding on a communication strategy that involves more forthright, prominent communication about the limits of the existing evidence base for animal charities, and why they think they're still worth supporting, then I'll be happy - and if they don't think that's the best strategy, then they aren't so worried about making people like me happy.
A good way to realize some of the benefits of a more honesty-centric strategy would be to write openly about the mistakes they made with their current communication strategy, why they made those mistakes, and what they intend to do differently. GiveWell put its astroturfing misstep on its mistakes page, and I think this was a good move. 80,000 Hours, Ben Todd's employer, also maintains a mistakes page, to its credit.
Cavalier attitude towards promises
Finally, there's the case of the Giving What We Can pledge:
I recognise that I can use part of my income to do a significant amount of good. Since I can live well enough on a smaller income, I pledge that for the rest of my life or until the day I retire, I shall give at least ten percent of what I earn to whichever organisations can most effectively use it to improve the lives of others, now and in the years to come. I make this pledge freely, openly, and sincerely.
In a post on the Effective Altruism forum about the GWWC pledge, Alyssa Vance discussed, among other things, possible circumstances in which giving away ten percent of one's income might not be compatible with the highest-impact strategy. In the next paragraph, she discussed how the pledge's lifelong nature might not be advisable, since it can be hard to predict one's future opportunities, especially as a student, which many pledge-takers are. At the end of this paragraph, she wrote:
In response to this argument, pledge taker Rob Wiblin said that, if he changed his mind about donating 10% every year being the best choice, he would simply un-take the pledge. However, this is certainly not encouraged by the pledge itself, which says "for the rest of my life" and doesn't contemplate leaving.
Rob Wiblin responded by quoting GWWC's sitewide FAQ (not on the pledge page or prominently featured there), which does talk about the prospect of leaving. The FAQ mentions that the pledge is not legally binding, and says that you “can” cease to be a member, specifically bringing up the prospect of “serious unforeseen circumstances.”
Overall, this seems entirely consistent to me with a reading whereby the pledge is morally binding barring circumstances that make it pretty much infeasible to fulfill. However, I tried to steelman this a little. Perhaps it was an honest misunderstanding. Some promises are like wedding vows, binding enough that the prospect that it might turn out not to be a good idea is a reason to be reluctant to commit. Others are like fitness goals, where if it turns out that running twice a week hurts your knees, you'll just do something else instead. If the pledge is meant more like a fitness goal than like a wedding vow, then the problem is just lack of clarity.
But Wiblin continues:
I think we should use the interpretation of the pledge that produces the best outcome. The use GWWC and I apply is completely mainstream use of the term pledge (e.g. you 'pledge' to stay with the person you marry, but people nonetheless get divorced if they think the marriage is too harmful to continue).
One thing I suspect is going on here is that people on the autism spectrum interpret all kinds of promises to be more binding than neurotypical people do (e.g. https://www.reddit.com/r/aspergers/comments/46zo2s/promises/). I don't know if that applies to any individual here specifically, but I think it explains how some of us have very different intuitions. But I expect we will be able to do more good if we apply the neurotypical intuitions that most people share.
Of course if you want to make it fully binding for yourself, then nobody can really stop you.
This reads to me like an implied claim – using the Gricean maxim of relation – that the Giving What We Can pledge, like wedding vows, is not serious enough that the prospect that it might not work out should make you reluctant to make the promise in the first place.
Paul Christiano's response says it as well as I could:
Usually we promise to do something that we would not have done otherwise, i.e. which may not be in line with our future self's interests. The promise "I will do X if my future self wants to" is gratuitous.
It seems reasonable for GWWC to say that the GWWC pledge is intended more as a statement of intent than as a commitment; it would be interesting to understand whether this is how most people who come into contact with GWWC perceive the pledge. If there is systematic misperception, it seems like the appropriate response is "oops, sorry" and to fix the misperception.
Instead, the initial response is to diagnose us with a mental disorder. To medicalize integrity.
There is not uniformly strong social pressure to take the pledge in the Effective Altruism community, but there is definitely some pressure, and from time to time people talk about whether people who haven't taken the pledge count as real EAs. There is also plenty of social proof. This will lead people who are especially agreeable to take the pledge when they might otherwise not have done so.
If these agreeable pledge-takers understand the pledge to be truly binding, when their peers who have taken the pledge have not, then they've been manipulated into devoting a disproportionate amount of resources through a misleading presentation. This is the sort of thing that can easily happen by accident. I don't blame Giving What We Can for the initial situation.
I blame the EA culture. If you accidentally mislead someone, and when this is pointed out to you, your response is not to seek to prevent the miscommunication is recurring, then it becomes your fault. Rob Wiblin is affiliated not with Giving What We Can, but with 80,000 Hours, a different subsidiary of the Center for Effective Altruism, so it's hard to assign responsibility to one particular actor here. But it says something bad about the EA culture's attitude towards trust that when this problem came up, the first response by an EA organization leader was to dismiss people who "misunderstood" as probably just autistic, rather than to ask how the pledge might be clarified.
In the comments to Sarah's post, Centre for Effective Altruism head Will MacAskill said he intend to write something up clarifying the pledge. He also endorsed the notion that there are seriously but not infinitely binding pledges such that the prospect that it might not work out is a serious concern:
There does seem to me to be more confusion around the pledge than I had realized, and I’ll write up something soon on how to understand the pledge, but it’ll take me a little bit to write it up in a way I’m happy with. In the meantime: I think that the analogy with wedding vows is pretty good – marriage is a firm intent, it’s not a trivial commitment, you’re doing a bunch of stuff to make it hard to back out of this commitment at a later date, you’re going to make life plans on the assumption that you’ll keep the commitment, and you’re making a commitment to act in a certain way even when its against your own later interest to do so. But, at the same time, you know that there’s a chance you won’t live up to the commitment, or that things will change sufficiently that being married to this person really really won’t make sense any more. In general, there’s quite a bit of space between “this is an absolutely binding commitment that I can never renege on” and “I can quit any time I feel like it”. I think that most promises, including serious promises like marriage, are in that space in between the two – having promised to do X provides a very significant, but not overriding, reason to do X.
This is encouraging, and I look forward to his writeup. But I'm disappointed that it seems to have taken this level of fuss to get this prioritized.
UPDATE: Since it may have been unclear to some - I do not mean to say that either Rob Wiblin or William MacAskill is personally to blame for this. This is a cultural divide, not a matter of individual bad actors. I fully believe both of them that the notion that the Pledge might be misleading never crossed their minds until the recent discussion I've referenced.
I'm not going to go into much depth on the case of Intentional Insights beyond Sarah's description (follow the first link for a more detailed account, "paid for likes" is not the best summary of what happened):
Intentional Insights, a supposed “EA” organization led by history professor Gleb Tsipursky, used astroturfing, paid for likes and positive comments, made false claims about his social media popularity, falsely claimed affiliation with other EA organizations, and may have required his employees to “volunteer” large amounts of unpaid labor for him.
To their credit, CEA repudiated Intentional Insights; Will McAskill’s excellent post on the topic argued that EA needs to clarify shared values and guard against people co-opting the EA brand to do unethical things.
People kept pointing out problems, and Gleb kept responding to the minimum possible construal of that problem - e.g. someone asked him not to use paid promoters, and point to an example, and he’d stop using that paid promoter on that forum, but use the same one on other forums, and other ones on that forum. He was not trying to honestly inform. He responded to criticism with “I’ve updated,” but didn't seem to be trying to make inferences about why his pattern of misbehavior was objectionable. Eventually, we needed public judgment – and eventually we got it.
I used to be happy with this outcome. But now I’m worried about our treatment of Gleb. At first I felt good that we’d identified and sort-of-expelled a bad actor, even if I was annoyed with some people for what I thought was extremely obsessive give-him-another-chance-ism. But I’m not so sure Gleb wasn’t just sincerely emulating the endorsed behavior and ideals of respected community leaders. Maybe the reason I don't see similarly outrageous behavior among other EA organizations – except the few moments where the curtain slips, like above – is because I accept them as authorities. Maybe Gleb's crime was not that he was dishonest, but that he was dishonest in poor taste.
Even one of the most ridiculous things Gleb did seems less exceptional now: A leader in an EA organization emailed him asking him to stop claiming to be supported by their organization. Later, this leader stated publicly that she'd sent this email. Gleb then complained that her public disclosure of the email she had sent him was an "unethical disclosures of my private messages with others."
This definitely seems different from the proposal that EA organizations should get to look at criticism before it's made public. But now I'm less sure that I'm justified in thinking that while the former is cartoonishly demanding, the latter is an understandable mistake. Maybe I'm just blinded by the respectability and good taste of the "better" EA organizations. Maybe I've been a self-righteous dupe. Maybe I still am.
I'm genuinely unsure.