Abby Hoskin comments on Social scientists interested in AI safety should consider doing direct technical AI safety research, (possibly meta-research), or governance, support roles, or community building instead