Melanie Mitchell: Magical Thinking on AI
12 hours ago
- #Regulation
- #Magical Thinking
- #AI Safety
- Thomas Friedman's columns discuss the urgency of U.S.-China cooperation on AI safety, focusing on superintelligent AI.
- Friedman's views are influenced by Craig Mundie, a former Microsoft executive, rather than AI researchers.
- Friedman engages in 'magical thinking' about AI, such as AI learning languages without training or scheming against humans.
- Claims about AI's emergent properties, like speaking untrained languages, have been debunked by experts.
- AI's translation abilities stem from vast datasets, not mysterious emergence.
- The 'AI schemes to murder executive' story was a contrived role-playing scenario, not evidence of AI agency.
- Friedman suggests only AI can regulate AI, dismissing human oversight, which is criticized as unrealistic.
- Proposals for a 'common ethical architecture' in AI overlook the complexity of moral reasoning and cultural differences.
- The article calls for more realism in AI discussions and regulations, grounded in human reasoning.