It is sometimes joked that the qualification needed for doing AI safety work is dropping out of a PhD program, which three people here have done (not that we would exactly recommend doing this!). Aside from those three, almost everyone else is doing or has completed a PhD.
Huh, I wonder if the sample was unrepresentatively high in its ~100% rate of PhD backgrounds? Here are the fractions of research staff at a few AI safety orgs that have started (and potentially completed) PhDs, based mostly on the organizations’ websites and listed staff’s LinkedIns:
1⁄8 of research staff at Redwood Research (although this doesn’t reflect recent hires)
(These orgs might be unrepresentatively low in PhD backgrounds, but the above numbers at least show that PhD backgrounds are far from universal among AI safety researchers.)
Relatedly, I’ve heard mixed opinions by safety researchers about whether PhDs are worth it. Some seem to worry that, compared to going directly into safety organizations (when that’s an option), PhD programs often (a) have very high time costs, (b) offer very little feedback about what is actually helpful for alignment, and (c) incentivize less relevant/useful work.
So overall I think people shouldn’t have the takeaway that there’s consensus on the value of PhD programs for this work.
Thanks for this!
Huh, I wonder if the sample was unrepresentatively high in its ~100% rate of PhD backgrounds? Here are the fractions of research staff at a few AI safety orgs that have started (and potentially completed) PhDs, based mostly on the organizations’ websites and listed staff’s LinkedIns:
1⁄8 of research staff at Redwood Research (although this doesn’t reflect recent hires)
5⁄11 of research staff at MIRI
1⁄2 of research staff at ARC
(These orgs might be unrepresentatively low in PhD backgrounds, but the above numbers at least show that PhD backgrounds are far from universal among AI safety researchers.)
Relatedly, I’ve heard mixed opinions by safety researchers about whether PhDs are worth it. Some seem to worry that, compared to going directly into safety organizations (when that’s an option), PhD programs often (a) have very high time costs, (b) offer very little feedback about what is actually helpful for alignment, and (c) incentivize less relevant/useful work.
So overall I think people shouldn’t have the takeaway that there’s consensus on the value of PhD programs for this work.