EA is vetting-constrained

I'm very interested in hearing from grantmakers about their take on this problem (especially those at or associated with CEA, which it seems like has been involved in most of the biggest initiatives to scale out EA's vetting, through EA Grants and EA Funds).

• What % of grant applicants are in the "definitely good enough" vs "definitely (or reasonably confidently) not good enough" vs "uncertain + not enough time/expertise to evaluate" buckets?
• (Are these the right buckets to be looking at?)
• What do you feel your biggest constraints are to improving the impact of your grants? Funding, application quality, vetting capacity, something else?
• Do you have any upcoming plans to address them?

Note also that the EA Meta and Long-Term Future Funds seem to have gone slightly in the direction of "less established" organizations since their management transition, and it seems like their previous conventionality might have been mostly a reflection of one specific person (Nick Beckstead) not having enough bandwidth.

(Just saw this via Rob's post on Facebook) :)

Thanks for writing this up, I think you make some useful points here.

Based on my experience doing some EA grantmaking at Open Phil, my impression is that the bottleneck isn't in vetting precisely, though that's somewhat directionally correct. It's more like there's a distribution of projects, and we've picked some of the low-hanging fruit, and on the current margin, grantmaking in this space requires more effort per grant to feel comfortable with, either to vet (e.g. because the case is confusing, we don't know the people involved), to advise (e.g. the team is inexperienced), to refocus (e.g. we think they aren't focusing on interventions that would meet our goals, and so we need to work on sharing models until one of us is moved), or to find. 

Often I feel like it's an inchoate combination of something like "a person has a vague idea they need help sharpening, they need some advice about structuring the project, they need help finding a team, the case is hard to understand and think about". 

Importantly, I suspect it'd be bad for the world if we lowered our bar, though unfortunately I don't think I want to or easily can articulate why I think that now. 

Overall, I think generating more experienced grantmakers/mentors for new projects is a priority for the movement.

I'm intermittently working on a project to provide more scaleable and higher quality feedback for project proposals for several months. First alpha-stage test should start in a time-horizon of weeks and I'll likely post the draft of the proposal soon.

Very rough reply ... the bottleneck is a combination of both of the factors you mention, but the most constrained part of the system is actually something like the time of senior people with domain expertise and good judgement (as far as we are discussing projects oriented on long-term, meta, AI alignment, and similar). Adding people to the funding organisations would help a bit, but less than you would expect: the problem is, for evaluating e.g. somewhat meta- oriented startup trying to do also something about AI alignment, as a grantmaker, you often do not have the domain experience, and need to ask domain experts, and sometimes macrostrategy experts. (If the proposal is sufficiently ambitious or complex or both, even junior domain experts would be hesitant to endorse it.) Unfortunately, the number of people with final authority is small, their time precious, and they are often very busy with other work.

edit: To gesture toward the solution ... the main thing the proposed system will try to do is "amplify" the precious experts. For some ideas how you can do it see Ozzie's posts, other ideas can be ported from academic peer review, other from anything-via-debate.

[meta: I'm curious, why was this posted anonymously?]

This seems like a good point, and I was surprised this hadn't been addressed much before. Digging through the forum archives, there are some relevant discussions from the past year:

• A post by RandomEA suggesting an EA crowdfunding platform (Raemon in the comments suggests having a 'common app' for the various funders, which seems like a good idea)
• benjamin-pence's announcement of the EA Angel Group (current status unclear)
• Brendon_Wong's post on three ideas for improving EA funding: a 'kickstarter' for projects, a platform for distributed grantmakers to share expertise and grant opportunities, and improved centralized grantmaking. Lots of interesting discussion in the comments.
• CEA's EA Meta Fund grants post mentioned a $10K grant for Let's Fund, an org that "helps people to discover and crowdfund breakthrough research, policy and advocacy projects" via performing "in-depth research into fledgling projects" and sharing their recommendations. So far, Let's Fund has a couple of object-level posts about specific areas (improving scientific norms and doing climate change research), and is running a crowdfunding campaign for $75,000.

I found a few more discussions that seemed relevant, but not that many.* One reason this might not be very discussed is major donors are often pretty well plugged-into the community, so their social networks might do a good enough job of bringing valuable opportunities to their attention. (And that goes double for big grantmakers.) Still, it seems to me like a hub of useful centralized information could benefit everyone, if it can establish itself as a Schelling point and not as another competing standard. And improving information flow to small donors alone is obviously still valuable, though I'd worry a bit about duplicated work and low-quality analyses resulting from a norm of distributed vetting.

*It's interesting that most of the relevant posts are from the past year. Maybe a result of the 'talent-constrained' discourse getting people interested in what value small donors can provide beyond more funding for big projects?

"If you look at where, for example, EA Funds spends their money, it seems like most of the funds are just going to safe bets"

I notice that this link is to the August 2018 disbursement, which was indeed all to established orgs. The two disbursements since then have included at least some grants to less established projects (November 2018, March 2019).

I think this is basically accurate. As I mentioned in another thread, the issue is that the scaling-up-of-vetting is still generally network constrained.

But, this framing (I like this framing) suggests to me that the thing to do is a somewhat different take on Earning to Give.

I had previously believed that Earning-to-Give people should focus on networking their way into hubs where they can detect early stage organizations, vett them, and fund them. And that this was the main mechanism by which their marginal dollars could be complementary to larger funders.

But, the Vetting-Constrained lens suggests that Earners-to-Give should be doing that even harder, not because of the marginal value of their dollars, but because this could allow them to self-fund their own career capital as a future potential grantmaker.

And moreover, this means that whereas before I'd have said it's only especially worth it to Earn-to-Give if you make a lot of money, now I'd recommend harder for marginal EAs to join donor lotteries. If a hundred people each put in $10k into 10 different donor lotteries, now you have 10 people with $100k, enough to seed fund an org for a year. And this is valuable because it gives them experience think about whether organizations are good.

There could be some systemization to this to optimize how much experience a person gets and how the org(s) they funded turned out to fare. (Maybe with some prediction markets thrown in)

This topic seems even more relevant today compared to 2019 when I wrote it. At EAG London I saw an explosion of initiatives and there is even more money that isn't being spent. I've also seen an increase in attention that EA is giving to this problem, both from the leadership and on the forum. 

Increase fidelity for better delegation

In 2021 I still like to frame this as a principal-agent problem.

First of all there's the risk of goodharting. One prominent grantmaker recounted to me that back when one prominent org was giving out grants, people would just frame what they were doing as EA, and then they would keep doing what they were doing anyway.

This is not actually an unsolved problem if you look elsewhere in the world. Just look at your average company. Surely employees like to sugarcoat their work a bit, but we don't often see a total departure from what their boss wants from them. Why not?

Well I recently applied for funding to the EA meta fund. The project was a bit wacky, so we gave it a 20% chance of being approved. The rejection e-mail contained a whopping ~0.3 bits of information: "No". It's like that popular meme where a guy asks his girlfriend what she wants to eat, makes a lot of guesses, and she just keeps saying "no" without giving him any hints. 

So how are we going to find out what grantmakers want from us, if not by the official route? Perhaps this is why it seems so common for people close to the grantmaker to get funded: they do get to have high-fidelity communication.

If this reads as cynicism, I'm sorry. For all I know, they've got perfect reasons for keeping me guessing. Perhaps they want me to generate a good model by myself, as a proof of competence? There's always a high-trust interpretation and despite everything I insist on mistake theory.

The subscription model

My current boss talks to me for about an hour, about once a month. This is where I tell him how my work is going. If I'm off the rails somehow, this is where he would tell me. If my work was to become a bad investment for him, this is where he would fire me. 

I had a similar experience back when I was doing RAISE. Near the end, there was one person from Berkeley who was funding us. About once a month, for about an hour, we would talk about whether it was a good idea to continue this funding. When he updated away from my project being a good investment, he discontinued it. This finally gave me the high-fidelity information I needed to decide to quit. If not for him, who knows for how much longer I would have continued.

So if I was going to attempt for a practical solution: train more grantmakers. Allow grantmakers to make exploratory grants unilaterally to speed things up. Fund applicants according to a subscription model. Be especially liberal with the first grant, but only fund them for a small period. Talk to them after every period. Discontinue funds as soon as you stop believing in their project. Give them a cooldown period between projects so they don't leech off of you.

Another data point suggesting a vetting bottleneck is Open Phil’s recent shift in how they’re funding EA meta/community organizations, including those who work on long-termist causes. This was motivated in part by “high uncertainty about how to set the right grant amounts for these organizations and our sense that we aren’t providing the level of accountability, oversight and vetting that we ideally would like to. We believe that individual donors (particularly to these organizations) sometimes seem to think our investigations into the organizations in question have been deeper than is actually the case.” (emphasis added/shifted)

In other words the funder with the most incentives, capabilities, and resources to vet these organizations (which I’d guess are abnormally hard to vet) doesn’t think it’s doing enough vetting, and is worried other donors are also under-vetting (based on erroneous assumptions). And it’s not just small projects that are under-vetted, the problem seems much broader.

Excellent post, although I think about it using a slightly different framing. How vetting-constrained granters are depends a lot on how high their standards are. In the limit of arbitrarily high standards, all the vetting in the world might not be enough. In the limit of arbitrarily low standards, no vetting is required.

If we find that there's not enough capability to vet, that suggests that either our standards are correct and we need more vetters, or that our standards are too high and we should lower them. I don't have much inside information, so this is mostly based on my overall worldview, but I broadly think it's more the latter: that standards are too high, and that worrying too much about protecting EA's reputation makes it harder for us to innovate.

I think it would be very valuable to have more granters publicly explaining how they make tradeoffs between potential risks, clear benefits, and low-probability extreme successes; if these explanations exist and I'm just not aware of them, I'd appreciate pointers.

Another startup contacted at least 4 grantmaking organisations. Three of them deferred to the fourth.

One "easy fix" would simply be to encourage grantmakers to defer to each other less. Imagine that only one venture capital fund was allowed in Silicon Valley. I claim that's one of the worst things you could do for entrepreneurship there.

grantmakers fall back on prestige because they don’t always have the resources to properly evaluate ideas

It seems like this recent post describes the opposite pattern, of someone with a highly prestigious resume spending a lot of resources getting evaluated, and getting rejected despite their resume. I wonder why the pattern would be different between hiring and grantmaking?

Anyway, one idea for helping address the bottleneck is to maintain a shared open-source grantmaking algorithm. The algorithm could include forecasting best practices, a list of ways projects can cause harm, etc. Every time a project fails despite our hopes, or succeeds despite our concerns, we could update the algorithm with our learnings. It could be shared between established EA grantmakers, donor lottery winners, independent angels, etc.

I don't think such an algorithm would eliminate the need for domain expertise. But it might make it less of a bottleneck. The ideal audience might be an EA who is EtG and thinking of donating to a friend's project. They can vouch for their friend and they have a limited amount of domain expertise in the area of their friend's project. They could do some fraction of the algorithm on their own, then maybe step 7 would be: "Find a domain expert in the EA community. Have them glance over everything you've done so far to evaluate this project and let you know what you're missing." (Arguably the biggest weakness of amateurs relative to experts is amateurs don't know what they don't know. Plausibly it's also valuable to involve at least one person who is not friends with the project leader to fight social desirability bias etc. Another way to help address the unknown unknowns problem is making a post to this forum and paying for critical feedback. OpenPhil has a relevant essay re: what they aim for in their writeups.)

I can't help but appreciate the irony that 5 hours after having been posted this is still awaiting moderator approval.

Thank you for writing this post!!