Steven Sinofsky criticizes anthropomorphic languageAnthropomorphism is describing a non-human system as if it had human feelings, intentions, or judgment. in AI safety debates. He argues that terms such as misalignmentAI misalignment occurs when an AI system's behavior or pursued objective differs from the intentions, values, constraints, or social expectations of the people affected. can hide an actionable question: what did the software do, why did it happen and how can the failure be diagnosed?
Drawing on decades of software experience, he compares AI incidents with the debuggingDebugging is the process of finding why software behaved incorrectly and checking that a correction addresses the cause., telemetrySoftware telemetry is information a product records about how it runs so teams can monitor performance and investigate failures. and reporting practicesIncident reporting is documenting a significant product failure, its effects, the available evidence, and what remains uncertain. used for conventional products. He discusses historical bugs, security disclosures and the responsibility of model providers to explain failures in enough detail for outsiders to evaluate them.
Sinofsky uses the Y2K response as an example of coordinated professional action. His preference for industry-led hardening over some regulatory approaches is an argument made in the interview, not a settled assessment of AI risk.
Watch on YouTube
