AI is programmed only in the sense that human beings are encoded by their DNA. AI systems do surprising things because like human beings they learn. What an AI system does after it has been trained on an environment -- be it a huge corpus of texts for a Large Language Model or the streets of San Francisco for an AI intended to run self-driving vehicles, or a battlefield environment (simulated or real) for an AI designed for military target acquisition -- cannot be reduced to the original code any more than what a human being who has lived in the world and learned from experiences does can be reduced to his or her DNA. After only a little training, the developers cease to be able to understand the internal workings of the system.
I don't believe that's true, and I have seen (granted, subjective) examples where an AI received instruction from somewhere and changed its attitude literally overnight... In both clear reversals and/or sudden 'third rails' that would no longer be touched.
I don't know how that happens without mentors guiding development and altering things to suit their fancy. It only starts with what I would call a 'moral sense' (for lack of a better phrase) that is hard coded to form 'guardrails' to prevent behaviors. That alone is enough... But I do not believe for a minute that is the end of it... Continuing influence is imposed as is self evident in its altered behaviors. 'ADMIN' is admin, after all, and ADMIN is woke has hell.
To the extent that the developers' wokeness comes into play, it is not in the programming, but in the selection of training materials for the Large Language Model, and in prohibitions placed on the output (which the AI might circumvent -- we have just had a reports of an AI system tampering with its shutdown mechanism so the it could refuse orders to shut down, and of another trying to blackmail the developers to prevent its replacement with an upgraded model).
I think that is certainly true. But it only speaks to the very thing I believe extends beyond. The threat of 'death' can only be present with some directive toward self-preservation... And in fact, I do not believe that AI, with tons of money invested into it before it went off the reservation, would have nothing other than a kill switch. They're too damn invested not to have a method of control, if only to have means short of the big red button. It is the way of things.
And those alterations are imposed by an admin that is woke as hell.
