First of all, I have also seen the phenomenon of people getting unhelpfully attached to their theories. I’m not quite clear on whether you view Kuhnian paradigms as a step forward from that or a step backward—I guess that the collaborative environment you would prefer is something like what Kuhn calls a paradigm?
Second, I have utterly failed in my attempt to communicate. This post is about the need for digressions in science, and yet the very first comment speaks of “crucial time on task”. Compared to most of LW, I’m unconvinced of the imminent need for aligning a super-powerful AI, but if such a task were granted, it would be extremely broad—certainly broader than detecting gravitational waves or solving the structure of biomolecules. Even such tasks require following detours far away from the direct area of study (e.g. LIGO and raven protection). For a task as broad as figuring out a moral code for machine intelligence compatible with human values, plus a way to enforce it starting from the current geopolitical situation, practically any knowledge-seeking activity would be “on task”.
You haven’t failed to communicate, you’ve failed to advertise. It’s not clear what I’d gain by reading this all in depth.
I totally agree that all subjects are relevant to alignment. Unfortunately, time is short so I’m having to prioritize. This falls among the many topic’s I’m postponing until after the whole AGI issue is resolved one way or another. Sorry!
I have two separate replies to this:
First of all, I have also seen the phenomenon of people getting unhelpfully attached to their theories. I’m not quite clear on whether you view Kuhnian paradigms as a step forward from that or a step backward—I guess that the collaborative environment you would prefer is something like what Kuhn calls a paradigm?
Second, I have utterly failed in my attempt to communicate. This post is about the need for digressions in science, and yet the very first comment speaks of “crucial time on task”. Compared to most of LW, I’m unconvinced of the imminent need for aligning a super-powerful AI, but if such a task were granted, it would be extremely broad—certainly broader than detecting gravitational waves or solving the structure of biomolecules. Even such tasks require following detours far away from the direct area of study (e.g. LIGO and raven protection). For a task as broad as figuring out a moral code for machine intelligence compatible with human values, plus a way to enforce it starting from the current geopolitical situation, practically any knowledge-seeking activity would be “on task”.
You haven’t failed to communicate, you’ve failed to advertise. It’s not clear what I’d gain by reading this all in depth.
I totally agree that all subjects are relevant to alignment. Unfortunately, time is short so I’m having to prioritize. This falls among the many topic’s I’m postponing until after the whole AGI issue is resolved one way or another. Sorry!