A study should be conducted that records and analyses the reactions and impressions of people when first encountering EA. Special attention should be paid to reactions of underrepresented groups such as groups based on demographics (age, race, gender, etc.), worldview (politics, religion, etc.) or background (socio economic status, major etc.).
Tsunayoshi
Karma: 702
I am hesitant to agree. Often proponents for this position emphasize the value of different outlooks in decision making as justification, but the actual implemented policies select based on diversity in a narrow subset of demographic characteristics, which is a different kind of diversity.
Hey Wil,
as someone who is likely in the “declining epistemics would be bad” camp, I will try to write this reply while mindfully attempting to be better at epistemics than I usually am.
Let’s start with some points where you hit on something true:
However I think the way this topic is being discussed and leveraged in arguments is toxic to fostering trust in our community
I agree that talk about bad epistemics can come across as being unwelcoming to newcomers and considering them stupid. Coupled with the elitist vibe many people get from EA, this is not great.
I also agree that many people will read the position you describe as implying “I am smarter than you”, and people making that argument should be mindful of this, and think about how to avoid giving this impression.
You cite as one of the implied assumptions:
Those with high quality epistemics usually agree on similar things
I think it is indeed a danger that “quality epistemics” is sometimes used as a shortcut to defend things mindlessly. In an EA context, I often disagreed with arguments that defer strongly to experts in EA orgs. These arguments vaguely seem to neglect that these experts might have systematic biases qua working in those orgs. Personally, I probably sometimes use “bad epistemics” as a cached thought internally when encountering a position for which I have mostly seen arguments that I found unconvincing in the past.
Now for the parts I disagree with:
I scrolled through some of the disagreeing comments on Making Effective Altruism Enormous, and tried to examine if any have the implicit assumptions you state:
It is a simple matter to judge who has high quality epistemics This comment argues that broadening the movement too much will reduce nuance by default. While it implies that EA discussions have more nuance than the average discussion, I do not think the poster or anyone else in the thread says it is easy to identify people with good epistemics. Furthermore, many argue that growth should be not too fast to be able to get people used to EA discussion norms, which implies that people do not necessarily think that bad epistemics are fundamental.
Those with high quality epistemics usually agree on similar things
I don’t think the strong version of this statement (“usually”) holds true for most people in the epistemics camp, some people, including me, would probably agree that e.g. disagreeing with “it is morally better to prioritize expected impact over warm feelings” is usually not good epistemics. I.e. there are some few core tenants for which “those with high quality epistemics” usually agree on.
It’s a given that the path of catering to a smaller group of people with higher quality epistemics will have more impact than spreading the core EA messaging to a larger group of people with lower quality epistemics
While many people probably think it is likely , I do not think the majority consider it “a given”. I could not find a comment that argues or assumes it is obvious in the above discussion.
What I personally believe:
My vague position is that one of the core advantages of the EA community is caring about true arguments, and consequently earnest and open-minded reasoning. In so far that I would complain about bad epistemics, it is definitely not that people are dumb. Rather, I think that it is a danger that people engage a bit more in what seems like motivated reasoning in some discussions than the EA average, and seem less interested in understanding other people’s position and changing their mind. These are gradual differences, I do not mean to imply that there is a camp who reasons perfectly and impartially, and another one that does not.
Without fleshing out my opinion too much (the goal of this comment is not to defend my position), I usually point to the thought experiment: “What would have happened if Eliezer Yudkowsky wrote ai safety posts on the machine learning subreddit?” to illustrate how important having an open-minded and curious community can be.For example, in your post you posit three implicit assumptions, and later link to a single comment as justification. And tbf, that comment reads a little bit dismissive, but I don’t think it actually carries the three assumptions you outline, and should not be used to represent a whole “camp”, especially since this debate was very heated on both sides. It is not really visible that you try to charitably interpret the position you disagree with. And while it’s good that you clearly state that something is an emotional reaction, I think it would also be good if that reaction is accompanied with a better attempt to understand the other side.
It’s very bad that the movement is focusing outreach on elite universities. Proximity to them should not be a criterion. We should invest in less elitist communities that can make the movement more diverse.
Very bad is a strong statement. Do you mind elaborating on why you think diversity in itself is important, and what kind of diversity you refer to (e.g. diversity of viewpoints, diversity of ethnicity etc.)? FWIW, Harvard students’ ethnic markup differs somewhat from the US population, but not very much so ( once you factor out non residents, the underrepresentation does not seem to exceed a factor of 2.0).
Nevertheless, it is true that focusing on elite universities is bound to attract students that are in some ways different from the population at large. However, focusing on them has the benefit on finding ambitious students with comparatively larger chances of impacting the world.
Additionally, elite universities just have a higher proportion of students who are even interested in EA in the first place, so network effects mean that these universities will probably have more fruitful and lively EA student groups. As a local group organizer in Germany, where we do not have elite universities, this difference is palpable. It seems local EA groups in Oxford and London are much more vibrant.
But ultimately we’re here to reduce existential risk or end global poverty or stop factory farming or other important work. Not primarily to make each other happy, especially during work hours
You raise many good points, but I would like to respond to (not necessarily contradict) this sentiment. Of course you are right, those are the goals of the EA community. But by calling this whole thing a community, we cannot help but create certain implicit expectations. Namely, that I will not only be treated simply as a means to an end. That means only being assessed and valued by how promising I am, how much my counterfactual impact could be, or how much I could help an EA org. That’s just being treated as an employee, which is fine for most people, as long as the employer does not call the whole enterprise a community.
Rather, it vaguely seems to me that people expect communities to reward and value their engaged members, and consider the wellbeing of the members to be important by itself (and not so the members can be e.g. more productive).
I am not saying this fostering of the community should happen in every EA context, or even at EA globals (maybe a more local context would be more fitting). I am simply saying that if every actor just bluntly considers impact, and at no place community involvement is rewarded, then people are likely and also somewhat justified to feel bitter about the whole community thing.
Very good post! Some potential tips how people who have similar experiences to what you described can feel more included:
Replacing visits to the EA Forum with visits to more casual online places: various EA Facebook groups (e.g. EA Hangout, groups related to your cause area of interest), the EA Discord server, probablygood.org (thanks to another commenter mentioning the site).
Attending events hosted by local EA groups (if close by). These events are in my experience less elite and more communal.
If attending larger EA conferences, understand that many people behave like they are in a job interview situation (because the community is so small, reputation as a smart person can be beneficial), and will consequently e.g. avoid asking questions about concepts they do not know.
AFAIK there is one positive, randomized trial for a nasal spray containing Iota-Carrageenan (Carragelose): “The incidence of COVID-19 differs significantly between subjects receiving the nasal spray with I-C (2 of 196 [1.0%]) and those receiving placebo (10 of 198 [5.0%]). ” It is available at least in Europe, and in the UK I think under the brand name Dual Defence. Why it has not received more attention is beyond me.
Interesting!
Does bullying increase with onset of adolescence? Schools alone cannot be the factor causing the decrease in life satisfaction, since it seems to occur after grade 5, but students have been in school before that already.
(Caveat: Due to space and time constraints, this comment aims to state my position and make it somewhat plausible, but not to defend it in depth. Also, I am unsure as to whether the goal of bioethicists is to come up with their own ethical positions, or to synthesize the ethics of the public in a coherent way)
For most of the post, I draw on decisions made by (bio)ethic committees that advise governments around the world. I believe those are a great basis for doing so, because they are generally staffed by researchers and independent. My cursory searching has found such committees in France and Austria; the members of the Austrian committee are mostly either high ranking bio-ethics professors, or are at least working in the field in some capacity. Their reports and votes are public. The info for the French members is less transparent. I have not looked into the various US ethic commissions because their appointments seem much more influenced by politics.
You make a great disambiguation of different levels of criticism against “bioethics”. The strong version of the view is that bioethicists as academic researchers reach bad conclusions, even compared to the general population.
I believe there is good justification for holding this view. In particular, many of the decisions made by ethic commissions are highly counter-intuitive to me:
Many of the provisions of informed consent differ from what the general public would consider reasonable. For example, in challenge trial protocols, even those created by proponents, payment of participants beyond time compensation was discouraged in order “not to take advantage of the poor”. I believe most people would disagree with that (depending on the framing), as would most EA-types.
The bioethics committee of Austria explicitly speaks out against surrogate motherhood: “In view of the manifold and complex social, mental and legal problems connected with “surrogate motherhood”, the Bioethics Commission recommends that methods of reproductive medicine be denied to male homosexual couples.” (I could not find a poll of the public for Austria, but the public in France is supportive)
The commission in France recommends against physician assisted suicide and euthanasia, the commission in Austria recommends only against the latter. (p.61)
The WHO advisory committee on Covid-19 challenge trials was split on whether it would be ethical to conduct one if there was no available treatment (p.9). Most of the members are however not bioethicists.
No strong evidence, but in reading these reports I have not seen them actually making a cost-benefit calculation or referring to one. I think doing so would be very unusual.
If one accepts these decisions as bad, then I do not believe that the defence of institutional dynamics is sufficient to explain them away. The members are not appointed by a politicized process, but seem to just be experts in their field, and certainly not career bureaucrats.
But they themselves and their decisions are sometimes public, so maybe they fear backlash over some decisions? However often there is a minority opinion advocating for more permissibility, so presumably holding such positions is both possible and does not lead to huge backlash.
“Moreover, I observe that machine-learning or model-based or data-analysis solutions on forecasting weather, pandemics, supply chain, sales, etc. are happily adopted, and the startups that produce them reach quite high valuations. When trying to explain why prediction markets are not adopted, this makes me favor explanations based on high overhead, low performance and low applicability over Robin Hanson-style explanations based on covert and self-serving status moves.”
I agree that the success of bespoke ml tools for forecasting negates some of the Hansonian explanations, but probably not most of them.
As ML tools replace human forecasts, they do not pose a threat to the credibility of executives. They do not have to provide their own forecasts that could later be falsified.
(Speculative) The forecasts produced by such tools are presumably not visible to every employee, while many previous instances of prediction markets had publicly visible aggregate predictions.
These tools forecast issues that managers are not traditionally expected to be able to forecast. Weather and pandemics are certainly not in the domain of executives, and I am unsure whether managers usually engage in supply chain and sales predictions.
These tools do not actually provide answers that could be embarrassing to executives, and for which prediction markets with aggregated human expertise could be useful. For example, machine learning cannot predict “conditional on proposal by CEO Smith, what will our sales be”. A good test for this explanation could be how many companies allow feedback to strategy proposals by employees and visible to all employees.
Thanks for the writeup! This is surely a perspective that we are missing in EA.
I did not have time to read all of the post, so I am not sure whether you address this: The cost-effectiveness estimates of XR are ex-post, and of just one particular organization. To me it seems obvious, that there are some movements/organizations that achieve great impact through protest, it is more difficult to determine that beforehand.
So as far as you propose funding existing projects, do you believe that the impact and behaviour of a movement are stable? Unlike NGOs, movements seem much more amenable to unforeseen (bottom-up) change, as there is inherently less control over it. How stable do you believe these movements to be?
They did not have a placebo-receiving control group.
All the other points you mentioned seem very relevant, but I somewhat disagree with the importance of a placebo control group, when it comes to estimating counterfactual impact. If the control group is assigned to standard of care, they will know they are receiving no treatment and thus not experience any placebo effects (but unlike you write, regression-to-the-mean is still expected in that group), while the treatment group experiences placebo+”real effect from treatment”. This makes it difficult to do causal attribution (placebo vs treatment), but otoh it is exactly what happens in real life when the intervention is rolled out!
If there is no group psychotherapy, the would-be patients receive standard of care, so they will not experience the placebo effect either. Thus a non-placebo design is estimating precisely what we are considering doing in real life: give an intervention to people, who will know that they are being treated and who would just have received standard of care (in the context of Uganda, this presumably means receiving nothing?).
Ofc, there are issues with blinding the evaluators; whether StrongMinds has done so is unclear to me. All of your other points seem fairly strong though.
You’d also expect that class of people to be more risk-averse, since altruistic returns to money are near-linear on relevant scales at least according to some worldviews, while selfish returns are sharply diminishing (perhaps logarithmic?).
It’s been a while since I have delved into the topic, so take this with a grain of salt:
Because of the heavy influence of VCs who follow a hits-based model, startup founders are often forced to aim for 1B+ companies because they lost control of the board, even if they themselves would prefer the higher chances of success with a <1B company. That is to say, there are more people and startups going for the (close to) linear utility curve than you would expect based on founders’ motivations alone. How strong that effect is I cannot say.
This conflict appears well known, see here for a serious treatment and here for a more humorous one.
You mention “It’s probably the case that the biggest harms from immigration come from people irrationally panicking about immigration, but (surprise!) people are in fact irrational.”.
From an EU-perspective, the effect seems pretty clear: After the refugee crisis 2015-2016 there have been numerous cases of populist right-wing parties gaining support or outright winning elections after running on anti-immigration platforms: to name just a few: the Lega Nord in Italy became part of the government, the FPÖ polled at their highest in 2016, and anti-immigration sentiment was at least influential for Brexit. These are arguably outcomes that substantially weaken political institutions and lead to worse governance.
This kind of backlash from some parts of the established population happens at moderate levels of immigration. We should expect it to be much stronger when immigration would be much higher under an Open Borders system, and account for the effects of that.
I think it is fair to say that so far alignment research is not a standard research area in academic machine learning, unlike for example model interpretability. Do you think that would be desirable, and if so what would need to happen?
In particular, I had this toy idea of making progress legible to academic journals: Formulating problems and metrics that are “publishing-friendly”could, despite the problems that optimizing for flawed metrics bring, allow researchers at regular universities to conduct work in these areas.
Looking forward to the posts, and happy to postpone further discussion to when they are published, but to me the question and your alluded to answer has enormous implications for our ability to raise life satisfaction levels.
Namely: very rough estimates suggest that we are now 100x-1000x richer than in the past, and our lives are in the range [good-ok], but generally not pure bliss or anything close to it. If we extend reasonable estimations for the effect of material circumstances on wellbeing (i.e. doubling of wealth increases satisfaction by 1 point on a 10 point scale) , we should then expect past humans to have been miserable.
I am skeptical that this was the case: On the one hand, belief systems like Buddhism clearly espouse that life is suffering. On the other hand, other religions are arguably not that pessimistic about life. Furthermore, folk tales and historic accounts generally do not seem to support that people were looking forward to their death (with some exceptions, e.g. spirituals from African-American slaves, that show that life can get that bad.) Also, existing hunter-gatherer tribes seem as satisfied as modern people. (which I guess you already incorporated into your chart somewhat).
To me, it is not surprising that some of the material gain gets eaten by the treadmill-effect (for example status symbols like flashier cars), but we have to remember that pre-modern people had no access to modern medicine to relieve pain (teeth pain can be horrible), far less delicious food and comfort, etc.. This could suggest that life satisfaction has, maybe not a a set-point, but rather a narrow range where it can move under realistic conditions.
This is a very comprehensive report, thanks for posting.
Given that education is seen as a strong predictor of populist attitudes, it is interesting that many interventions listed on the demand side seem to target highly educated people (e.g. Our World in Data, Factfullness, Journalism, Fact checking in general, BPB). The Youtube channel Kurzgesagt and some things Last week tonight comes up with (e.g. the wrestler John Cena warning against conspiracy theories) seem a bit better. You mention research how they might affect policy, but it would also be interesting how they affect the attitudes of the broader audience in general.
Still, information spreading explicitly and effectively targeted towards people at risk of populist attitudes (older, less educated) seems kind of rare? Where are the civic education memes that can be shared in boomer facebook groups? E.g. to combat conspiracy theories, I see a lot of videos with experts explaining the issue in simple terms, when this is exactly the kind of people that populists consider to be the elite not to be trusted.
Chomsky publishing his new book, The Precipice, mere months after Long Story Short clearly indicates that he and Taylor must be closely working together. I look forward to the surely upcoming 80000 hours joint appearance of Taylor Swift and Noam Chomsky.
But shouldn’t this update our priors towards mostly being on the happy timeline, in the West as well? Given that it took Sinovac/China one year from last March to this March to scale up, and that their vaccines are easier to manufacture than mRNA vaccines, and if we assume high investment from the start in China (so their timeline is close to optimal), it really starts to look like we could not have done much better on manufacturing (because the West does not differ strongly in available doses compared to China)?
I.e. we could have approved a few months earlier, but even in December the UK and the US (I think?) were mostly bottlenecked by supply issues, so an earlier approval should not have changed much by this intuition.
Thanks for the reply! I had not considered how easily game-able some selection criteria based on worldviews would be. Given that on some issues the worldview of EA orgs is fairly uniform, and the competition for those roles, it is very conceivable that some people would game the system!
I should however note that the correlation between opinions on different matters should apriori be stronger than the correlation between these opinions and e.g. gender. I.e. I would wager that the median religious EA differs more from the median EA in their worldview than the median woman differs from the median EA.
Your point about unknown unknowns is valid. However, it must be balanced against known unknowns, i.e. when an organization knows that its personnel is imbalanced in some characteristic that is known or likely to influence how people perform their job. It is e.g. fairly standard to hire a mix of mathematicians, physicists and computer scientists for data science roles, since these majors are known to emphasize slightly different skills.
I must say that my vague sense is that for most roles the backgrounds that influence how people perform in a role are fairly well known because the domain of the work is relatively fixed.
Exceptions are jobs where you really want decisions to be anticorrelated and where the domain is constantly changing, like maybe an analyst at a venture fund. I am not certain at all however, and if people disagree would very much like links to papers or blog posts detailing to such examples.