An Open Letter to Steven Pinker on AI
3 hours ago
- Scott Alexander challenges Steven Pinker to a public debate on AI, despite Pinker's reservations about confrontational formats.
- He refutes Pinker's four arguments against superintelligence: 1) intelligence is a valid, scalable concept; 2) AI can develop self-preservation and dominance drives through convergent goals, misgeneralization, and reward hacking; 3) multi-goal AIs still pose existential risks; 4) humans will inevitably grant AI physical infrastructure access.
- Alexander lists 11 grievances against Pinker, including mischaracterizing the AI risk community as fatalistic, accusing them of believing in omniscient AI, cherry-picking experts, misrepresenting studies, using pseudo-arguments, and distracting from near-term harms.
- He defends probabilistic predictions (p(doom)) as legitimate, citing superforecasting methodology and experts' track records.
- Alexander criticizes Pinker for endorsing low-quality voices (e.g., Claire Lehmann) who attack the AI safety community unfairly.
- He urges Pinker to adopt 'Pinkerism'—a rational, evidence-based approach to AI risk—and warns that denial may discredit techno-optimism, drawing parallels to climate change denial.
- Alexander insists on debating Pinker specifically, rejecting alternatives, and jokingly suggests a duel if debate is refused.