Framing safety as accidents, not superintelligence — shares its discarded foil with → Formal verification and control theory recur as safety building blocks

Asimov's first law is the one concept framing-accidents-vs-speculative-risk and formal-methods-imported both claim, and the paper faults it on two independent axes depending on which register picks it up. As a way of talking about safety, it fails for being monolithic: "robots must not harm humans" offers no decomposition, where accidents-in-ml immediately fractures harm into five separately tractable technical causes. As a candidate route to a machine-checkable safety property, it fails for being informal: Weld and Etzioni's 1994 attempt to formalize it within classical planning is set against the complete formal verification of the federal aircraft collision-avoidance system, real engineering rigor achieved for cyber-physical systems that the same section admits has "not focused much on modern machine learning systems" (concrete-problems, §"Cyber-Physical Systems Community:", p. 20). One fictional sentence anchors both of the paper's self-boundary exercises, undercut on granularity in one and on rigor in the other.