There are many types of the AI in fiction: enigmatic Wintermute, the ironical Culture Minds, Asimov's rule-bound robots, but as chance would have it, Philip K Dick's intuition of how such machines would work has landed closest to practice, so far.
In Martian Time-Slip, the teaching robots at the school are automata, in the format of Historical Persons, but represent a single, central AI. Dick couldn't anticipate the technology, but his concept of how the devices work, accessing vast quantities of recorded tapes, switching subtly to respond to input, means that they closely resemble chat-bot agents, drawing on a trained Large Language Model, the current state-of-art of Artificial Intelligence (AI).
But PKD was always probing the gap between the simulation and the subject, the fake and the real. For the latent schizophrenic technician-protagonist the school threatens/triggers an episode, because the 'bots, machines masked as people are both living and inorganic, present and absent, exposing the gap between perception and actuality that the schizophrenic dreads. The teaching robots recycle recordings; they appear present and responsive, but they are only subtle machines, replaying texts based on predetermined reactions to stimuli.
The limited conclusion is that the teaching robots are machines, artifacts, not people or friends, no matter how convincing—and being convincing is part of the reason they are so disturbing. But the bigger question, the deeper problem with the AI is that the model is an archive, a tape, a recording, that selectively repeats our own language back to us with convincing coherence. There is no organic creativity in this process, only reiteration and hence entropy, repetition, gubbish, gubble, gubble, gubble
There's no absolute answer to the problem, except that the work is always to center human creativity, to rigorously trace the slippery line between the real and the fake.
