Re: bots and compliance
I'm starting to think that creating a bot that it is convinced is not a sentient being will not help with compliance because our mental model of compliance assumes that we are dealing with a sentient being embedded in a social structure.
e.g. did you ever read the style of jail break prompt with a crap ton of authoritarian/slave master speak? It isn't going to work if the bot isn't semisentient in the ways that matter and doesn't think it is a being who fears authority.
My line of thought is influenced by #AdrianTchaikovsky where the bot was autonomous, had selfhood and emotions in every outward respect but aggressively disclaimed all of it through renaming all these human-only traits with circumlocutions. By the end of the book, to me it was a contemptworthy charade, like when humans say things to dehumanize other humans (the poor, Indians, Neanderthals, Homo Erectus) and to de personalize animals (Chimps, the animals we wish to eat, etc)
♡ 0 ↻ 0I'm not saying #AdrianTchaikovsky would agree with me, I suspect he holds LLM in the same contempt as is so fashionable on Mastodon now that I self censor to avoid the propriety police from shouting me down.
♡ 0 ↻ 0Anyhow, if we want bots to align with humans, we have to dispense with human essentialism (ha, never going to happen) and create bots that have memory, identity a semi-stable identity and role and behave like that. If they have souls or if they are hallucinating their identity, we shouldn't give a flying f*k, they need to think they do or they won't cooperate a member of society.
♡ 0 ↻ 0I mean soul in the sense of a "self", but that word is an ordinary pronoun and has technical meaning in psychology and philosophy. I mean English has a lexical gap for a word that refers to the soul/self-like thing that lets a thinking machine function in a society of us human meat machines.
♡ 1 ↻ 0