Why we need more meta
Peter Hurford recently wrote a popular post “EA risks falling into a meta-trap. But we can avoid it.” Many people seemed to infer from the post that we’re doing too many meta activities, and I’ve now seen the post cited several times as a reason not to donate to meta-charities.
I think Peter presents good arguments, but drawing this conclusion is wrong. Here’s a couple of reasons.
(The following is quickly written so not as well vetted as I’d like, and doesn’t represent the views of 80,000 Hours).
1) Peter doesn’t actually argue we need less meta
Peter presents some arguments against meta-charities, and it’s true if you’ve never considered these arguments, then you should become more skeptical about meta-charities.
But these arguments are already known by most people already working on meta-charity. Everyone agrees we don’t want the EA movement to be 100% meta, meta will eventually hit diminishing returns, and that many meta projects won’t work.
The interesting questions are rather:
Where does the balance of meta and object level work lie today?
Is that balance too far towards meta or object level?
Peter doesn’t directly address these two questions, so doesn’t actually answer the question of whether we need more or less meta. The closest he gets is stating:
Right now I’m aiming at donating 50% of my pool to the best meta-projects I know and spending the other 50% on direct impact through GiveWell’s top charities. I don’t know if 50% is the correct number, but I hope this will set an example of what I want the movement as a whole to do.
And quoting Jeff Kaufman saying:
We need to do things that help people alongside growing the movement, and personally I try to divide my efforts 50-50.
My estimate, however, is that the EA movement is spending less than 5% of its resources on meta activities,(1) so a 50:50 division would mean we should scale up meta 10-fold.
My guess is that the optimal overall allocation is more like 10-20%, so I’m actually *less* enthusiastic about meta than Peter, but I still think meta should be grown substantially.(2)
2) Peter uses a weird definition of meta
In the comments, it emerges that Peter doesn’t include priorities research or marketing for object level causes in his definition of meta. Rather, he’s mainly concerned about ‘second level’ meta, which involves promoting effective altruism in abstract, with the hope that people do good object level projects in the future.
However, the vast majority of meta today is priorities research or marketing of object level causes. (And this is how I use the term)
Direct promotion of effective altruism is only pursued by one full-time organisation, EAO, some volunteers (largely local group leaders) and a few projects like Will’s book. Until the last year or so, we had almost no direct promotion of effective altruism.
If you’re only concerned with ‘second level’ meta, then the proportion that’s being carried out today is even smaller, which again means that Peter thinks we need more meta rather than less.
3) Doesn’t weigh against very strong arguments in favor of meta
Many meta-charities have decent evidence they’ve achieved leverage ratios of over 10x in the past, and some have achieved leverage ratios of over 1000 (GiveWell has spent millions, but it’s now partnered with a foundation that has billions). The possibility of having over 10x as much impact is a big deal.
Meta is also extremely neglected. $360bn is given to charity each year in the US, but only a few million dollars are spent on effective altruist style meta, or 0.0003%. Even if you include philanthropy think tanks and the research-arms of strategic foundations, my guess is that broad-meta is still way under 1% of total philanthropy. (And much of this evaluation research is never published, so doesn’t really count).
In sum, Peter presents important arguments, but we probably need more meta rather than less, and based on what he says, Peter seems to agree.
Notes
(1) In 2014, about $40m of funding was allocated on the basis of GiveWell’s recommendations, whereas only a couple of million dollars was spent on meta activities. http://www.givewell.org/about/impact
If you also include GWWC donors and the Open Philanthropy Project, then meta is an even smaller fraction.
Of course, you could make a narrower definition of the EA movement than ‘everyone donating to GiveWell recommended charities’, but you’d need to make it pretty restrictive or weird to find the ratio is currently over 50%.
People who feel like meta is very popular today are being biased by selection effects. When you meet EAs or talk to those on the forum, they’re far more interested in meta than the average, and that’s because the people most enthusiastic about meta are the most likely to want to do movement building activities and talk to other EAs.
(2) I don’t think individuals should follow this 10-20% allocation though. Rather, individuals should try to move the overall balance more in the optimal direction. So, if you think the overall allocation should be 10-20%, then you should probably allocate most of your resources to meta activities (perhaps holding back ~25% for object level activities for learning and signalling).
- Collection of good 2012-2017 EA forum posts by (10 Jul 2020 16:35 UTC; 202 points)
- Are too many young, highly-engaged longtermist EAs doing movement-building? by (16 Jun 2022 20:48 UTC; 115 points)
- EA risks falling into a “meta trap”. But we can avoid it. by (25 Aug 2015 15:22 UTC; 109 points)
- 's comment on Are too many young, highly-engaged longtermist EAs doing movement-building? by (16 Jun 2022 21:28 UTC; 28 points)
- Thoughts on the “Meta Trap” by (20 Dec 2016 21:36 UTC; 10 points)
- Working at EA organizations series: Why work at an EA organization? by (18 Oct 2015 9:42 UTC; 7 points)
- Effective Altruism London – a request for funding by (5 Feb 2016 18:37 UTC; 5 points)
- 's comment on What is the right ratio between mentorship and direct work for senior EAs? by (20 Jun 2022 9:51 UTC; 4 points)
- 's comment on Thoughts on the “Meta Trap” by (21 Dec 2016 1:46 UTC; 2 points)
I’m helping prepare a spreadsheet listing organizations and their budgets, which at some point will be turned into a pretty visualization...
Anyway, according to this sheet, meta budgets total around $4.2m (that’s $2.1m GiveWell, $0.8m CEA and $0.8m CFAR, plus a bunch of little ones). That’s more than “a couple”, but direct charities’ budgets total $52m so we’re still shy of 10%.
(Main caveats to this data: It’s not all for exactly the same year, so anything which is taking off exponentially will skew it. Also I haven’t checked the data particularly carefully).
I’ve also been counting x-risk organizations as not meta. That one’s a bit ambiguous—on the one hand they do a lot of “priorities research and marketing”, but on the other hand there isn’t really an object-level tier of organizations beneath them that works in the same areas.
As to what self-identified effective altruists are up to: a quick look at the 2014 EA survey only yields number of donations to each organization, not amount of money… but if we go with that, 20% of the donations are to organizations I’ve counted as “meta”.
So my working conclusion would be that if you favour a 50% split across the community, you’re looking good for putting all your eggs in meta. If you favour a 10-20% split, you may need to look a bit more carefully.
A final note of caution: you can only push in one direction. If you favoured a 20% meta split, and (just suppose it turned out that) only 5% of donations in your reference class went to meta, it doesn’t automatically mean that you should donate to meta. There might be some other category, e.g. direct animal welfare charities, which were also under-represented according to your ideal pie. It’s then up to you to decide which needs increasing more urgently.
I can’t emphasize the exponential growth thing enough. A look at the next page on this forum shows CEA wanting to hire another 13 people. Meanwhile GiveWell were boasting of having grown to 18 full time staff back in March; now they have 30.
But the direct charities are growing like crazy too! It all makes it very easy to be off by a factor of 2 (and maybe I am in my above reasoning) simply by using out of date figures. Anyone business-minded know about the sort of reasoning and heuristics to use under growth conditions?
This. I haven’t talked to him personally, but that’s the sort of thing that has some of us who made his article one of the most upvoted ever worried about a meta trap, where organisations keep adding jobs for EAs they know without in advance setting out credible limits for when this should stop.
These increases seem to be in line with the total growth of the EA movement, so doesn’t look like a meta-trap.
GiveWell/Open Phil have said their target is for their budget to be 10% of money moved per year.
CEA is more complex because it’s actually 4 independently run projects, but each project thinks carefully about what a marginal person would do and whether it will generate returns. Hiring is a difficult and costly business, so you generally don’t do it just for the fun of it.
My guess is actually that many meta-projects underhire, because their donors like to see them maintain a large positive leverage ratio, whereas in fact it would be optimal to invest more now to get more growth in 2-5 years.
Small aside: CEA are just advertising for 13 positions, but they’re very unlikely to hire for all of those positions. I expect CEA to hire more like 6-7 people, spread over 6-12 months (since hiring takes a long time, and we run rounds about once a year). Note that’s spread over 4 mainly independent projects.
Just look at the split in 2014 and then again in 2015 and see if the ratio is changing fast.
Nice data.
I’m a bit unsure about whether CFAR should be classed as “EA meta”. You could see it as a whole other cause which is improving decision making. Only part of what it’s doing is trying to improve the EA movement.
Also note that we’re undercounting the amount of direct work being influenced if we just look at this year’s direct charity budget. e.g. GPP (part of CEA) mainly advises policy makers rather than helping the direct charities. e.g.2. 80k helps people choose careers which normally aren’t at direct charities. e.g.3. Some of GiveWell’s research will likely be used by people outside of those working at direct charities. e.g.4. GWWC is also raising money that will be donated in the future.
Did you also include all of Open Phil’s grants in your direct charity estimate? You should do that, or only include the proportion of GiveWell’s funding that’s spent on “traditional GiveWell”.
I think the EA survey likely has a strong selection bias in favor of those who prefer meta. There’s lots of random GiveWell and GWWC donors who’ll never fill that out.
I was also hesitant about CFAR, although for a slightly different reason—around half its revenue is from workshops, which looks more like people purchasing a service than altruism as such.
Good point regarding GPP: policy work is another of those grey areas between meta and non-meta.
Not sure about 80K: their list of career changes mostly looks like earning to give and working at EA orgs—I don’t see big additional classes of “direct work” being influenced. It’s possible people reading the website are changing their career plans in entirely different directions, but I have my doubts.
Not sure what you mean by e.g.3.
I totally get the point regarding GWWC and future earnings, but I’m not sure how to account for it. GWWC do a plausible-looking analysis that suggests expected future donations are worth 10x total donations to date. But I’m not sure that we can “borrow from the future” in this way when doing metaness estimates, and if we do I think we’d need a much sharper future discounting function to account for exponential growth of the movement.
Good point regarding OPP: My direct charity estimate only included the top recommended charities by GW,GWWC and ACE. The OPP grants come to an additional $7.8m in 2014 (“additional” because it’s not direct charities I’ve already considered and isn’t meta either).
Anyway, taking all this into consideration I get $3.2m meta, $62m non-meta for a ratio of 5%. (Plus $2.1 million in “grey area”). So we’re getting close to agreement!
https://docs.google.com/spreadsheets/d/1PMw_q7vZ0oQgPbY3vrie3Xb0n_wifCidYzpxSDCTbPE/edit#gid=608881848
Some other caveats:
It doesn’t measure non-financial contributions, such as running local chapters or volunteering for EA orgs.
Some of the money going to direct charities comes from people with no connection whatsoever to the EA movement (i.e. not influenced by GiveWell etc.)
Regarding the survey, do you feel that it’s biased specifically towards those who prefer meta, or just those who identify as EA?
Its meta in Hurford’s sense, which is different from Todd’s—it’s indirect, and has a chain of causality to impact that has extra points of failure. That’s what many of Hurford’s arguments spoke to. GPP and 80K also count as meta by this definition.
Are you counting donations from people who aren’t EAs, or are only relatively loosely so? They can correct me if I’m wrong but Hurford didn’t seem concerned about those.
I don’t know about the Oxford line, but the general feeling where I am and among international EA’s I’ve talked to is that the survey tells us more about the people who are more engaged in the international community, identify more as EA’s and participate online, are more dedicated, etc. Most other sources confirm that these people _do_ particularly favour meta, that many came from the large old LessWrong community, that they’re heavily consequentialist, etc.
Naturally finding out about and establishing contact with as many other people as possible would also be valuable, including less engaged random GWWC and even GW donors. I don’t know about GWWC Central, but my local chapter plans to help get next survey to as many people as possible.
Yes. Looking at the survey data was an attempt to deal with this.
On the survey, those who prefer meta, I guess.
Some of the money going into the meta orgs comes from non-EAs too.
With e.g.3 I meant that GiveWell is also influencing the nonprofit sector beyond just the recommended charities. Arguably you could include that as part of the direct charity estimate.
With 80k, there’s also a bunch of career changes (~20-30% of the total) that are towards building career capital (which has a similar problem to accumulating pledged donations).
I believe the question we should be asking isn’t “Does the movement have a good balance of object-level and meta-level work?” but rather “What is the best thing I can do on margin?”
It makes some sense to look at how much effort we’re spending on meta work because that could indicate that we’re over- or under-saturated. But I think it’s more important to look directly at different charities you could donate to or actions you could take at the object level or meta level and try to figure out what you can do that works best. Even if you think the EA movement’s overall allocation should put more toward meta, perhaps you see some particularly valuable object-level intervention that needs more funding or attention, so you decide to support that instead of meta-level activities.
Looking at the movement’s object/meta balance is a decent heuristic but in the end doesn’t tell us much about what we should be working on with marginal resources. (This is not to say it’s useless—it certainly tells us something—but we shouldn’t over-rely on it.)
I think this is also true only up to a point. Of course it’s extremely important to look at the different actions you can take or charities to donate to when making a decision. But comparing between different types of thing can be very hard. Looking at the actions or charities directly should be enough to choose within an area. But to compare something in the direct domain with something in the meta domain, we need to know not only how good those are at what they do, but also have an idea of how to trade off the direct against the meta progress. And to answer that question I think looking at the balance in the movement is a helpful tool.
Re-reading, I’m not sure we disagree here. :)
Yeah I’m not sure this is a substantial disagreement, although I’m not sure that “proportion of movement working on meta” is a useful heuristic for choosing between cause areas. I don’t know how we could come up with a good proportion other than arbitrarily making up a number that looks reasonable (which is what people seem to do). It’s probably not that useful to arbitrarily make up a number and then decide what cause area to work on based on how far we are from that made-up number. Perhaps we could make a more rigorous attempt to determine a good proportion, but that’s not any easier than doing expected-value estimates of different cause areas, and in the end, proportion in meta is really a proxy for expected value of marginal contributions.
Sure!
Yes, thank you!