Comments on CAIS
LessWrong, January 12, 2019
Abstract
This article discusses Comprehensive AI Services (CAIS) – a model of AI development proposed by Eric Drexler – which posits that AI development will progress through the creation of specialized AI services rather than a singular general-purpose AI. The author, Richard Ngo, broadly agrees with the empirical claim that AI services will become superhuman in many ways before a single superhuman AI is developed. However, he also argues that CAIS is likely to underestimate the convergence of AI services into a system resembling a singular, goal-directed agent. Moreover, he contends that, even if CAIS accurately describes the first stage of AI development, it may be a less accurate model of the period when AI becomes truly dangerous. – AI-generated abstract.
