Gary Marcus called for a pause on generative AI in Oct. 5 posts about the New York City Council's AI hearing. His commentary combined agreement with warnings about insufficient scrutiny of AI-written code with a disagreement over whether every kind of AI has a control problem.
Marcus reported Jacob Coxon criticizing the review of AI-generated code, while noting that his account might be slightly paraphrased from memory. Marcus agreed with the concern. In a subsequent post, he said Kokotajlo echoed Coxon's warning that AI-written code was not being vetted carefully enough.
These posts establish Marcus's account of the discussion. They do not provide a verified transcript or measurements of how much code AI writes or how thoroughly developers check it.
Marcus draws a distinction between AI systems
Marcus also reported Coxon arguing that no AI system could yet be controlled. He disputed that broad claim, pointing to AlphaGo and GPS routing as counterexamples in his post.
His own position was narrower: he argued that control of generative AI remained unresolved and advocated pausing that technology until it could be controlled. That is Marcus's assessment and policy preference, rather than a finding established by the hearing.
The Council's Sept. 28 announcement scheduled the Oct. 5 Committee of the Whole hearing to examine AI risks and potential safeguards. It named proposals for whistleblower incentives, a private right of action for people harmed by AI agents and independent third-party validation. The notice describes proposed measures; it does not announce their enactment or a generative AI pause.
Comments