I think the cultural slide will include self-censorship, e.g., having had this experience (of being banned out of the blue), in the future I’ll probably subconsciously be constantly thinking “am I annoying this author too much with my comments” and disengage early or change what I say before I get banned, and this will largely be out of my conscious control.
(I don’t want to start a fight and hopefully I’ll write a post explaining the behavior I’m talking about, but I’ll say abstractly, my hope in general is for people (me, you, anyone) to try as much as feasible to make fairly precise updates, like “this specific behavior pattern is bad / unhelpful / unwelcome in this context” rather than “I should be vaguely more worried about being vaguely annoying”.)
I think when a human gets a negative reward signal, probably all the circuits that contributed to the “episode trajectory” gets downweighted, and antagonistic circuits get upweighted, similar to AI being trained with RL. I can override my subconscious circuits with conscious willpower but I only have so much conscious processing and will power to go around. For example I’m currently feeling a pretty large aversion towards talking with you, but am overriding it because I think it’s worth the effort to get this message out, but I can’t keep the “override” active forever.
Of course I can consciously learn more precise things, if you were to write about them, but that seems unlikely to change the subconscious learning that happened already.
I think that’s fairly limited evidence, would want to see more data than that before claiming anything is vindicated.
Yeah, I would take a bet with you about eg if you’ll be banned by another author in the next 3 years. I think at least 60% on “no”.
I think the cultural slide will include self-censorship, e.g., having had this experience (of being banned out of the blue), in the future I’ll probably subconsciously be constantly thinking “am I annoying this author too much with my comments” and disengage early or change what I say before I get banned, and this will largely be out of my conscious control.
(I don’t want to start a fight and hopefully I’ll write a post explaining the behavior I’m talking about, but I’ll say abstractly, my hope in general is for people (me, you, anyone) to try as much as feasible to make fairly precise updates, like “this specific behavior pattern is bad / unhelpful / unwelcome in this context” rather than “I should be vaguely more worried about being vaguely annoying”.)
I think when a human gets a negative reward signal, probably all the circuits that contributed to the “episode trajectory” gets downweighted, and antagonistic circuits get upweighted, similar to AI being trained with RL. I can override my subconscious circuits with conscious willpower but I only have so much conscious processing and will power to go around. For example I’m currently feeling a pretty large aversion towards talking with you, but am overriding it because I think it’s worth the effort to get this message out, but I can’t keep the “override” active forever.
Of course I can consciously learn more precise things, if you were to write about them, but that seems unlikely to change the subconscious learning that happened already.