Abstract:
In recent work, I argued that there is an AI-specific no miracles argument (NMA) which is at least as plausible as the traditional NMA. In this talk, I will express two doubts about this claim, relative to the AI-specific NMA that I previously proposed. I will then develop a new AI-specific NMA in response, which really is at least as plausible as the traditional NMA. At its core is a novel concept, which I dub quasi-representation.
