What EA can learn from startups

What is the most “EA” activity? The example that comes to mind is Givewell doing cost-benefit analyses on which interventions produce the best return on investment, in terms of lives saved per dollar spent. When this was first conceived, it made a lot of sense: there were all kinds of charities running out trying random interventions, and what was missing was an attempt to quantify how these stacked up against one another.

But this analytical origin has permeated through the EA culture, such that the default go-to methodology for tackling a question is “think real hard, then write ten thousand words”. It feels like most EA work happens through long, philosophical, deeply researched tracts with copious footnotes posted on the EA Forum, rather than through trying things out and seeing what happens. Many topics discussed have little experimental grounding, and there’s little appetite for experimentally validating things.

Contrast this to building a software product. At many different levels, you have iterative feedback loops telling you whether the thing you did was good:

These feedback loops rapidly ground a developer’s understanding of what actually happens in the world. What are the feedback loops inside the EA ecosystem?

Another common startup adage is that “ideas are worthless, execution is everything”. But in the EA world, most of what gets evaluated are ideas. Karma, prestige, attention and funding accrue to those who speak prettily.

An analogy: there are two ways to aim a gun. You can sit there with a scope, chart of bullet velocities, wind map, and calculator, trying to estimate exactly where the bullet will land before taking the shot. Or: you can fire a tracer bullet, leaving a streak behind the flight path of the bullet, and then see how different angles

There are times when you want to do the first! If you’re firing a sniper rifle, and only have one chance at getting things right, then you do want to step back and run your calculations carefully; to measure twice before firing once. But it’d be a very costly mistake to use that same process to eg shoot down enemy aircraft; there, you should be firing hundreds of bullets a minute, using tracer rounds to identify where your shots are actually going.

Ben Kuhn describes the virtue of impatience:

With few exceptions, I have a hard time associating “impatient” with EA. The most impatient fastest-moving altruists I have heard of like Tara MacAulay and Sam Bankman Fried took leadership roles in CEA… then left to make a bunch of money instead. I speculate about an evaporative cooling effect, where capital EA “Effective Altruism” orgs lose out on fast movers because fast movers are allergic to staying in slow orgs.

Example: There’s a huge difference in payment speed from fundraising in Silicon Valley, vs receiving grants from FTX/SFF/LTFF. When Manifold went to raise our seed round, angel investors would usually sign after a single video call; median time between getting into contact with the funder to having the check hit our bank account was about two weeks. In contrast, with charitable funding from EA orgs, I can’t recall a single application process that got back to us in under a month; wiring the funds would usually take another 2+ weeks.

Some of this is due to the difficulty of funding through 501c3 charity systems vs for-profit investment schemes. But as another point: incorporating Manifold as a C-Corp was fairly streamlined, taking a couple weeks to figure out through Stripe Atlas. There’s no Stripe Atlas for 501c3 incorporation — getting an entity up-and-running to accept grant funding for Manifold looks like it will take 4-8 weeks.

It feels like EA prizes deliberation over moving fast and breaking things; consensus-seeking over unilateral action, asking permission over begging forgiveness, avoiding bad PR over acquiring good PR.

In general, it feels like EA asks you to speak in certain shibboleths (eg “be aligned”) before it is willing to direct funding and support your way. And it will enthusiastically fund people who pass the threshold, and basically ignore all others.

One example: the Carrick Flynn campaign drew out dozens of interested students and donors, who were especially excited about the prospect of “one of us” making it to Congress. It was cool to see the amount of resources coalescing around a single cause - but it was also in noted contrast to the relative apathy of EA folks to politics at large.

The contrast might be to a Catholic hospital (disclaimer, I’m Catholic); the emphasis is NOT on providing service only to other Catholic people. Because what would that prove? Every ingroup is happy to help ingroup members. This willingness to serve others was crucial in the spread of early Christianity — if a missionary goes to your village and preaches about the virtues of their faith, that’s not especially convincing. But if that same missionary comes in and spends long hours nursing you to health that one time you caught typhoid, then when you recover you’re much more receptive to the teaching behind the actions. Universal, unconditional love is far more inspirational than watertight logical arguments that seem to further your own aims.

Just looking at the name, “Effective Altruism” has two components: being effective, and being altruistic. Conditionally providing support based on ingroup cohesion can be very effective, but not especially altruistic. Sure, it’s an altruism that cashes out to helping people in the distant future — but how is this distinguishable from any other political or religious movement which claims to be helping others?

Lsusr on Ineffective Altruism:

When was the last time any EA group made “service to others” an explicit action that they did? Sure, there are a lot of service orgs out there, and it’s not in EA’s comparative advantage not to be feeding the homeless in soup kitchens.

What is the marginal impact of doubling down on in-funnel EA members, vs people on the border?

EA Global asks “How EA are you” before deciding to admit you.

My smart friend, ex-Google Brain and now VP of robotics at Halodi, asked: “For all the money and brainpower spent, what has AI Safety accomplished in the last 5 years?” And… I had a really hard time answering this question!

Jesus says in Matthew 7:15-20:

Essentially: the only way you can evaluate whether a person or organization is doing good, is by the good they have produced.

When I say “legible” here, I don’t just mean “other EAs are impressed by this” — I’m actually thrown by how low the bar in EA is for people to be impressed. Rather, it means “the rest of the world robustly thinks this is cool”.

Trevor Klee:

Carrick again: it definitely was an impressive amount of money spent and ads shown.

https://trevorklee.substack.com/p/how-to-win-an-election-in-10-easy?s=r

My sense is that EA places a lot of emphasis on individual talent. Research and writing (which forms the bulk of full-time EA output) is a fairly solo-directed activity. EAs focus on recruiting the smartest, drawing their talent pool from elite universities.

Some of this is due to the quick growth of the EA movement; if your median member has only been around for a year, they won’t have had time to form the connections to get really good at management. But a lot of it is also in the kinds of programs that EAs support: they recruit early-career folks quite heavily, with ample support for university and even high-school students. But on the flip side, few concessions are made to bringing on mid- and late-career folks, who have organization expertise to coordinate between the work done by different roles.

In contrast, the work that gets done in Silicon Valley companies is highly collaborative, with room for individuals to specialize in a variety of different roles. On one eng team building a single product, you might have:

Seems to run counter to “too many ICs” — but this is about cause area, that’s about operational efficacy.

EA feels like a recreational sports team or social club, instead of being a professional sports team or a cult

Would rather have 1 proto-Givewell than 10 University Groups

Examples of zero-sum

Funding

Job applications

Research (competes for attention)

Positive sum

Process improvements

Trade, specialization

Others: