I wrote this to help think through considerations. Now I decided to publish it. habryka left some comments on the google doc, some of which I copied here.
One way to do philanthropy:
Foundation style. Start/pick a foundation or donation-advising-org. Give it money and high-level direction but defer to it on details. Maintain some insight into its grants and reasoning, so that if you have conviction that it's messing up, you can tell it to start/stop doing certain things.
Another way to do philanthropy:
S-process style. Pick a bunch of recommenders. Have each recommender describe what they would do with every possible budget, then compete to convince you. Allocate money to the most compelling recommenders' recommendations. Repeat regularly.
I don't know which is better. I want to think through considerations. Here's some potential advantages of foundation style (some of which could be attained by sophisticated versions of s-process style — you can kinda read this as hard goals for s-process style philanthropy to achieve):[1]
Flexibility. You can move fast. You can also make weird pledges rather than just making grants.
Ability to do private/sensitive stuff. Fund stuff in private; reason in private.
Steering. You can steer grantees much better: you can offer them $X conditional on them promising to do or not do certain things. (S-process recommenders don't have good tools for steering, at most just "purpose-restricted funding" or "conveying that they wish for the org to do certain things."
Division of labor. For each org, someone is responsible for keeping track of it and maybe advising it or explaining how the funder is orienting to it. Each recommender doesn't have to evaluate each org they might like.
(Maybe other upsides downstream of bureaucracy. Also many downsides.)
Institutional knowledge. Like, SFF has no institutional knowledge, and maybe you can do slightly better but not as well as CG.
habryka says "I think you can do much better than CG at institutional knowledge via this process, though it's pretty unproven."
Oh also various important things related to being permanent (vs e.g. SFF is reinstantiated with a new composition once per year) or representing the institution or having some power that doesn't depend on the funder still liking your recommendations next time. Being able to make pledges or share predictions about your future behavior, or maybe just being predictable to grantees.
Grantmakers get to focus more on funding good things, rather than (a) wasting effort competing with other recommenders and making recommendations legible to the funder and (b) optimizing for recommendations that the funder will like when they inspect the s-process.
On (a), habryka says bureaucratic institutions require all kinds of bullshit to make a grant. Makes sense.
(Maybe kinda: avoiding the unilateralist's curse. An s-process can take steps to avoid adverse selection, but it generally seems good to fund the best things according to each of several different perspectives, and an amazing version of "foundation style" would do this but I buy that bureaucracies will tend toward being conservative and one-viewpoint-y.)
(Not necessarily: professional grantmakers. I think including professional grantmakers is an upside, and SFF makes worse decisions because very few of the recommenders have grantmaking expertise/context [and SFF doesn't facilitate access to such expertise/information], but an s-process could/should include some grantmakers or donation-advising-orgs among its recommenders.)
One other thing: habryka suggests that the s-process is supposed to help avoid grifters and confidence games. I don't get it? In the s-process the recommenders will optimize for (a) saying stuff that the funders want to hear and (b) recommending stuff that the funders will feel happy to fund, just like in the CG setup except probably more because the funders are more hands-on and it's very salient to the recommenders that they're competing with each other to look good to the funders. Update: habryka says (excerpt):
One of the key things you get is critique. You have evaluators who have different worldviews who critique the evaluations by other people.
The standard issue with foundations is that everyone at the foundation is trying to show a unified front to the funder, because you all share a reputation, and it's easy for you to coordinate on a mutual reputation protection alliance.
I'm thinking about this because habryka is creating a new s-process system, not because of my experience with SFF (but I also hope to write about soon). SFF's problems are mostly due to bad execution, not fundamental problems with s-process philanthropy. I'm tentatively excited about habryka's new s-process.
I wrote this to help think through considerations. Now I decided to publish it. habryka left some comments on the google doc, some of which I copied here.
One way to do philanthropy:
Another way to do philanthropy:
I don't know which is better. I want to think through considerations. Here's some potential advantages of foundation style (some of which could be attained by sophisticated versions of s-process style — you can kinda read this as hard goals for s-process style philanthropy to achieve):[1]
One other thing: habryka suggests that the s-process is supposed to help avoid grifters and confidence games. I don't get it? In the s-process the recommenders will optimize for (a) saying stuff that the funders want to hear and (b) recommending stuff that the funders will feel happy to fund, just like in the CG setup except probably more because the funders are more hands-on and it's very salient to the recommenders that they're competing with each other to look good to the funders. Update: habryka says (excerpt):
I'm thinking about this because habryka is creating a new s-process system, not because of my experience with SFF (but I also hope to write about soon). SFF's problems are mostly due to bad execution, not fundamental problems with s-process philanthropy. I'm tentatively excited about habryka's new s-process.