I’ve been engaging in some way with EA since 2018. First by helping run EA Oxford. Then by running the group and setting up remote ’fellowships’. Recently I was on the CEA groups team looking into ways to support university groups.
I’m now looking for new things to work on. Hoping to shift towards work which both:
Gives me more sustained motivation (possibly building systems and developing code).
Shifts my focus more sharply towards urgent sources of existential risk
I notice that you quoted:
The big funding bodies (OpenPhil, EA Funds, etc.) should be disaggregated into smaller independent funding bodies within 3 years
This feels like a totally seperate proposal right? Evaluated separately a world like ‘Someone who I trust about as much as I trust Holden or Alex (for independent reasons) is running an independent org which allocates funding’ seems pretty good. Specifically it seems more robust to ‘conflict-of-interest’ style concerns whilst keeping grant decisions in the hands of skilled grantmakers. (Maybe a smaller version of this proposal is splitting up Holden and Alex’s arms of OP although they seem independent enough that I’m not too bothered about this)
I think we could start talking about how feasible it is to construct an organisation like this and whether major funders like Holden would want to funnel money towards it but at the very least it hasn’t been ruled out by any of the reasoning above
I agree with other commenters saying 9mins seems too long but the general idea is good. I think a human read shorter summary would be really good. A lot of the videos like this I’ve seen also have some synthy soundtrack over the top which I’d add, just because I was put off by it being missing
I think this post is my favourite for laying out why a really convincing utilitarian argument for something which is common sense very bad shouldn’t move you. From memory Eliezer says something like ~Thinking there’s a really good utilitarian argument doesn’t mean the ends justify the means, it just means your flawed brain with weird motivations feels like there’s a really good utilitarian argument. Your uncertainty in that always dominates and leaves room for common sense arguments, even when you feel really extra super sure. Common sense morality rules like “the ends shouldn’t justify the means” arose because people in practice are very miscallibrated about when the ends actually justify the means so we should take the outside view and assume we are too.~
(By miscallibrated I think I could defend a claim like “90% of the time people think the ends definitely justify the means and this clashes with common sense morality they are wrong”)
I might be butchering the post though so you should definitely read it.
https://www.lesswrong.com/posts/K9ZaZXDnL3SEmYZqB/ends-don-t-justify-means-among-humans
I was responding mainly to the format. I don’t expect you to get complete answers to your earlier two questions because there’s a lot more rationality methodology in EA than can be expressed in the amount of time I expect someone to spend on an answer
If I had to put my finger on why I don’t feel like the failure to answer those questions is as concerning to me as it seems to be for you I’d say because.
A) Just because it’s hard to answer doesn’t mean EAs aren’t holding themselves and each other to a high epistemic standard
B) Something about perfect not being the enemy of good and about urgency of other work. I want humanity to have some good universal epistemic tools but currently I don’t have them and I don’t really have the option to wait to do good until I have them. So I’ll just focus on the best thing my flawed brain sees to work on at the moment (using what fuzzy technical tools it has but still being subject to bias) because I don’t have any other machinery to use
I could be wrong but my read from your comments on other answers is that we disagree most on B). E.g you think current EA work would be better directed if we were able to have a lot more formally rational discussions. To the point that EA work or priorities should be put on hold (or slowed down) until we can do this.
Easy way for users to blind authorship of posts. And a way to see a trending feed which is only affected by blinded karma
Cross posting to Nathan’s post since it’s pretty recent
(Posting here so people who just read this post can easily see)
From the comments I think the consideration I hadn’t considered was names on posts hold people accountable for the content of their post.
TLDR: We don’t have some easy to summarise methodology and being rational is pretty hard. Generally we try our best and hold ourselves and each other accountable and try to set up the community in a way that encourages rationality. If what you’re looking for is a list of techniques to be more rational yourself you could read this book of rationality advice or talk to people about why they prioritise what they do in a discussion group
Some meta stuff on why I think you got unsatisfactory answers to the other questions
I wouldn’t try to answer either of the previous questions because the answers seem big and definitely incomplete. I don’t have a quick summary for how I would resolve a disagreement with another EA because there are a bunch of overlapping techniques that can’t be described in a quick answer.
To put it into perspective I’d say the foundation to how I personally try to rationally approach EA is in the Rationality A-Z book but that probably doesn’t cover everything in my head and I definitely wouldn’t put it forward as a complete methodology for finding the truth. For a specific EA spin just talking to people about why they prioritise what they prioritise is what I’ve found most helpful and an easy way to do that is in EA discussion groups (in person is better than online).
It is pretty unfortunate that there isn’t some easy to summarise methodology or curriculum for applying rationality for charity current EA curricula are pretty focussed on just laying out our current best guess and using those examples along with discussion to demonstrate our methodology.
How is EA rational then?
I think the main thing happening in EA is that there is a strong personal, social, and financial incentive for people to approach their work “rationally”. E.g people in the community will expect you to have some reasoning which led you to do what you’re doing, and they’ll feedback on that reasoning if they think it’s missing an important consideration. From that spawns a bunch of people thinking about how to reason about this stuff more rationally, and we end up with a big set of techniques or concepts which seem to guide us better.
For times when the authorship of a post probably affected how I interacted with it. I think those effects were negative. (E.g they were closer to biasing against novel ideas from newcomers to the movement than correctly promoting important updates about influential people/organisations in the movement to the frontpage)
I can think of a time where the authorship of a post probably affected how I interacted with it
I’d be interested to hear if my experience is similar to others. Use agree-disagree voting on my replies to this comment to vote in this poll.
I recently applied for funding and found it helpful to look at my month to month spending over the last few months. I guessed at a rough mean monthly spending over 6 months but I might have been better off picking a median. I also forgot to account for tax!
I’m a bit put off by the claim that the student focus is an historical coincidence. It seems to me that people have really doubled down on that commitment because of the career flexibility students have (as Thomas mentioned). Maybe in the early years of the movement this was a historical coincidence but I don’t think ‘historical coincidence’ is an honest answer to the question “Why has EA community building been focused on students?” nowadays.
Fwiw I don’t think we should focus only on students and agree with a lot of the rest of the post
In the scenario where AGI would 100% be malevolent it seems like slowing progress is very good and all AIS people should pivot to slowing or stopping AI progress. Unless we’re getting into “is xrisk bad given the current state of the world” arguments which become a lot stronger if there’s no safe AI utopia at the end of the tunnel. Either way it seems like it’s not irrelevant
Hi Thomas, great question. I’ve included a list below for our records as of today (mid Nov).
It’s worth noting that we think any of these groups could absorb at least 2 FTE at a minimum, so I’d like people looking at these numbers to not be put off applying for the Campus Specialist Internship or Campus Specialist Programme based on the amount of FTE they currently have (although if you’d want to work on some of the under supported groups that would be amazing).
Berkeley 1.9 FTE (of which 1 FTE is EAIF funded)
Brown 0.25 FTE
Caltech 0 FTE
Cambridge 3.5 FTE (grant transfer in progress)
Columbia 1 FTE
Georgetown 0.25 FTE
Harvard 1.375 FTE
Of that:
Harvard Undergrad 0.925 FTE
Harvard Grad 0.25 FTE
Harvard Law 0.2 FTE
LSE 0.75 FTE
MIT 1.125 FTE
Oxford 2 FTE
Princeton 0 FTE
Stanford 1.3 FTE
Swarthmore College 0.75 FTE
UChicago 0.15 FTE
UHong Kong 0 FTE
UPenn 0.875 FTE
Yale 1.9 FTE
Also worth noting that there are a bunch of other more accessible descriptions of longtermism out there and this is specifically a formal definition aimed at an academic audience (by virtue of being a GPI paper)
Once again I really like this model Bob. I’m pretty excited to see how this model changes with even more time to iterate. I’d never come across the formalised idea of slack before and I think it describes a lot of what was in my head when responding to your last post!
I’m wondering how you’ve been thinking about marginal spending in this model? I.e, If we’re patient philanthropists, which choices should we spend money on and which should we save once we factor in that some choices are easier to affect than others? For example, one choice might be particularly hingey under any of your proposed definitions but be very hard for a philanthropist to affect; e.g the decisions made by one person who might not have heard of EA (either as a world leader or just being coincidentally important). We probably won’t get a great payoff from spending a load of money/effort identifying that person and would prefer to avoid that path down the decision tree entirely.
I guess the thrust of the question here is how we might account for tractability of affecting choices in this model. Once tractability is factored in me might prefer to spend a little money affecting lost of small choices which aren’t as hingey under this definition rather than spending a lot affecting one very hingey choice. If this is the case I think we’d want to redefine hingeyness to match our actual decision process.
It seems like each edge on the tree might need a probability or cost rating to fully describe real-world questions of traceability but I’d be very interested in yours or others thoughts.
(maybe call it ‘hinge precipiceness’?)
(Or maybe precipicipality?)
I really like this model and will probably use it to think about hingeyness quite a lot now!
I’ll make an attempt to give my idea of hingeyness, my guess is that the hingeyness is new enough an idea that there isn’t really a correct answer out there.
You can think of every choice in this model as changing the distribution of future utilities (not just at the next time step but the sum across all time). Hingier choices are choices which change this distribution more than any other. For example a choice where one future branch includes −1000 and a bunch of 0s and the other includes 1000 and a bunch of 0s is really hingey as it changes the portfolio from [-1000, many 0s, 1000] to [-1000, many 0s] or [many 0s, 1000]. A choice between [many 0s, 10] and [more 0s, another 10], is not hingey at all and has no effect on history. A good rule of thumb is to think of a choice as hingier the more it reduces the range of possible utilities.
As an extreme example, choosing between powerful world governments where one values utopias and the other values torture seems very hingey as before we had a large range of possible futures and after we’re irreversibly in a really positive or negative state.
I’ll apply this model to some of your questions below.
All older years have more ripple effects, but does that make 1 more hingey?
I think in the diagram above 1 is coincidentally quite hingey because you choose between a world where you’re guaranteed a 7 or 8 utility down the line or a world where you have a range between 0 or 6. The range of possible choices for one option is very different than the range for the other. You can imagine a similar timeline where the hingy moments are somewhere else (I’ve done an ugly drawing of one such world) As you can see in this timeline choice 1 doesn’t matter at all in the long run because the full range of final options is still open to you but the second tier choices (and the one labelled 0 in the third tier) matter a lot as once you make them your range changes in big ways.
The absolute utility that 1 and 2 could add are the same, but the relative utility is very different. So, what is more important for the hingeyness?
I think neither are the key values here. Hingeyness is about leverage over the whole human trajectory so the immediate changes in utility are not the only thing we should consider. We care more about how this affects aggregate expected utility over all the remaining future states. This is why irreversible choices seem so concerning.
One last thought here is that hingeyness should probably also include some measure of tractability. It could be one choice has a large effect on the future but we don’t have much of a capacity to affect that choice. For example, if we discovered an asteroid heading towards earth which we couldn’t stop. There’s no point in considering something the hinge of history if we can’t operate it! Currently, I don’t think that’s in the model but maybe you could add it by imposing costs on each choice? My guess is this model could become pretty mathematically rigorous and useful for thinking about hingeyness.
Fwiw “EA seems drawn to drama” is a take I’ve heard before and I feel like it’s kind of misleading. The truth is probably closer to “small communities are drawn to drama, EA is also drawn to drama and should (maybe) try to mitigate this”. It’s not super clear to me whether EA is worse or better than it’s reference class. Modelling the community as unusually bad is easy to say from the inside and could lead us to correct against drama in the wrong ways