As someone who runs one of EAs advocacy contingents, I think the overall idea of more criticism is probably a good idea (though I suspect I’ll find it personally unpleasant when applied to things I work on), but I’d suggest a few nuances I think exist here:
EA is not unitary, and different EAs and EA factions will have different and at times opposing policy goals. For example, many of the people who work at OpenAI/Anthropic are EAs (or EA adjacent), but many EAs think working at OpenAI/Anthropic leads to AI acceleration in a harmful way (EAs also have differing views of the relevant merits of those two firms).
Which views are considered EA can change the composition of who identifies as EA, EA-adjacent, unopposed, and EA-hostile—e.g. my perception of Sam Altman would be as EA-adjacent, but the perception that EAs have been critical of OpenAI, along with other events, likely pushed him further away from EA than he’d otherwise be; Elon Musk and Peter Thiel may also be related examples.
Advocacy is inherently information-lossy, since it involves translating information from one context into a format that will be persuasive in some sort of politically useful way. Usually this involves simplification (because a popular or decision-maker audience has less bandwidth than an expert audience) and may also involve differentiation (since the message will probably tend to be optimized to fit something like the existing views of its audience). This is a hard challenge to manage.
One type of simplification I’ve noticed is from an internal EA-organizing perspective—where the experts/leaders at the center tend to have nuanced, reasonable views, but those views, when being transmitted to organizers who again transmit to less experienced people interested in EA, can become translated into a dogma that is simplistic and rigid.
Two case studies of EA (or EA-adjacent) advocacy—monetary/macroeconomic policy and criminal justice reform—have had interestingly different trajectories. With monetary policy in the U.S., EA-funded groups tended to foreground technical policy-understanding and (in my opinion) did a good job transitioning their recommendations as macroeconomic conditions changed (am thinking mainly of Employ America). The criminal justice reform movement (where I founded a volunteer advocacy organization, the Rikers Debate Project) has in my opinion been mostly unable to reorient its recommendations and thinking in response to changing conditions. In my opinion, the macroeconomic policy work had more of a technocratic theory of change than more identity-oriented criminal justice reform efforts funded by EA though there were elements of technocracy and identitarianism in both fields. (Rikers Debate, which was not funded by EA groups, has historically been more identitarian in focus).
As someone who runs one of EAs advocacy contingents, I think the overall idea of more criticism is probably a good idea (though I suspect I’ll find it personally unpleasant when applied to things I work on), but I’d suggest a few nuances I think exist here:
EA is not unitary, and different EAs and EA factions will have different and at times opposing policy goals. For example, many of the people who work at OpenAI/Anthropic are EAs (or EA adjacent), but many EAs think working at OpenAI/Anthropic leads to AI acceleration in a harmful way (EAs also have differing views of the relevant merits of those two firms).
Which views are considered EA can change the composition of who identifies as EA, EA-adjacent, unopposed, and EA-hostile—e.g. my perception of Sam Altman would be as EA-adjacent, but the perception that EAs have been critical of OpenAI, along with other events, likely pushed him further away from EA than he’d otherwise be; Elon Musk and Peter Thiel may also be related examples.
Advocacy is inherently information-lossy, since it involves translating information from one context into a format that will be persuasive in some sort of politically useful way. Usually this involves simplification (because a popular or decision-maker audience has less bandwidth than an expert audience) and may also involve differentiation (since the message will probably tend to be optimized to fit something like the existing views of its audience). This is a hard challenge to manage.
One type of simplification I’ve noticed is from an internal EA-organizing perspective—where the experts/leaders at the center tend to have nuanced, reasonable views, but those views, when being transmitted to organizers who again transmit to less experienced people interested in EA, can become translated into a dogma that is simplistic and rigid.
Two case studies of EA (or EA-adjacent) advocacy—monetary/macroeconomic policy and criminal justice reform—have had interestingly different trajectories. With monetary policy in the U.S., EA-funded groups tended to foreground technical policy-understanding and (in my opinion) did a good job transitioning their recommendations as macroeconomic conditions changed (am thinking mainly of Employ America). The criminal justice reform movement (where I founded a volunteer advocacy organization, the Rikers Debate Project) has in my opinion been mostly unable to reorient its recommendations and thinking in response to changing conditions. In my opinion, the macroeconomic policy work had more of a technocratic theory of change than more identity-oriented criminal justice reform efforts funded by EA though there were elements of technocracy and identitarianism in both fields. (Rikers Debate, which was not funded by EA groups, has historically been more identitarian in focus).