Why GiveWell can’t recommend MIRI or anything like it

There’s an old joke about a man, head down, slowly walk­ing in cir­cles un­der the light of a street lamp. It is dark and he has lost his wallet.

A passerby offers his as­sis­tance and asks what they’re look­ing for. A sec­ond does the same.

Fi­nally, this sec­ond helper asks “Is this where you lost it?”

“No,” comes the re­ply.

“Then why are you look­ing over here?”

“Be­cause this is where the light is!”

The ten­dency to look for an­swers where they can be mea­sured or found may also be pre­sent in psy­cholog­i­cal re­search on rats. We don’t re­ally look at rats for psy­cholog­i­cal in­sight be­cause we think that’s where the psy­cholog­i­cal in­sights are, that’s just the only place we can look! (Note, I know look­ing at rats is bet­ter than noth­ing, and we don’t only look at rats).

Like­wise, GiveWell. They’ve re­leased their new list of seven char­i­ties they recom­mend donat­ing to. Six are efforts to in­crease health in a cheap way, and the last is di­rect money trans­fers to help peo­ple break out of poverty traps. In the­ory, these are the most cost-effi­cient pro­duc­ers of good in the world.

Ex­cept, not re­ally. Tech­nolog­i­cal re­search, es­pe­cially AI, or per­haps effec­tive ed­u­ca­tional re­form, or im­prov­ing the sci­en­tific com­mu­nity’s norms might very well be vastly more fruit­ful fields.

I don’t think these are all miss­ing from GiveWell’s list only be­cause they don’t mea­sure up, but be­cause, by GiveWell’s met­rics, they can’t be mea­sured at all! GiveWell has pro­vided, per­haps, the best of the char­i­ties that can be eas­ily mea­sured.

What if the best char­i­ties aren’t eas­ily mea­surable? Well, then they won’t just not be on the list, they can’t be on the list.