Arvind Narayanan: The question shifts from 'What can AI do?' to 'What should AI do?'
soumitrashukla9 · x · 2026-10-19
Princeton professor Arvind Narayanan argues we need a mindset shift: capabilities are no longer the deployment bottleneck, so the question moves from "What can AI do?" to "What should we actually delegate to AI?" He uses his own workflow as evidence — it kept changing for years as AI automated more grunt work, but has been stable for six months since he hit his stride with coding agents. He could have AI do his research and essays end to end; quality isn't the main concern — the value of his essays lies in the voice and perspective he and his collaborators bring, which readers can't get from AI-generated text they could produce themselves.
More from AGI Musings
- Cortes Island project mines LLM world knowledge to make a place legible — and questions the cost — anselm · 2026-10-19
- Hot take: AI will crack quantum gravity before scaled general-purpose robotics arrives — mattbeane · 2026-10-19
- Pedro Domingos: The less you understand intelligence, the easier it is to think you've solved it — pmddomingos · 2026-10-19
- Scoble: 'What if robots never touch anything?' talk helped store lift sales 258% — Scobleizer · 2026-10-19
- Investor: AI's radical speedup of knowledge discovery could yield cancer cures — matt_slotnick · 2026-10-19
- Is AGI alignment even solvable in principle? Researcher questions the premise — danfaggella · 2026-10-19