Claims

For current AI systems it is very unlikely they are conscious; consciousness may be confused with intelligence, and the assumption that computation is the basis of consciousness is shaky, even if the brain is a biological machine producing consciousness—it may be a different kind of machine that silicon cannot emulate.

We have already developed language models that are very close to passing the traditional Turing test (indistinguishable from humans), and some personas generated by GPT-4 have already passed 5-minute Turing tests; in the past this would have been evidence of consciousness, though now we may resist drawing that conclusion.

Language models have semantics and understanding, as evidenced by their ability to solve problems requiring understanding of concepts like paint fading and temporal reasoning (the room-painting problem), though they lack grounding in physical reality; multimodal systems trained with robotic perception can be grounded