Let your ideas die so you don’t have to. Navigating parochialism. Fond of fallibilism.
I’m currently working as founder and research lead at Formation Research on technical interventions for lock-in risks.
As a researcher, I am interested in AI-enabled totalitarianism, authoritarianism, coups, and power concentration, and governance-informed solutions for these problems leveraging technical methods of verification and cooperation.
Good call. I wrote about this in a previous, more high-level threat models post but didn’t include it explicitly here, and I agree it’s an important category of risk.
Agreed, and this is not explicit in the post which maybe it should be. It’s explicit in my past reasoning (the idea that in talking about these outcomes, we are thinking about the most pressing, e.g., the most widespread, harmful, oppressive, etc). In saying ‘long-term AI-driven power concentration’, the outcomes I care about preventing are the most pressing.
I agree power concentration isn’t inherently bad, but saying what good power concentration looks like is thorny and I have not gone to much effort to try to conceptualise it. I am focusing on preventing the worst-case bad outcomes, which I think are at least a bit easier to identify.
I also agree that diffused power can also lead to bad outcomes, however the positioning of lock-in risk research focuses on these concentrated or stable states because they share specific properties (as outlined in my other writing) at which we can target interventions.