The AI expectations gap

Why it matters

A survey conducted by Szymon Sacher, Tess Cotter, Charles I. Jones, Anton Korinek, and Peter McCrory compares the expectations of Anthropic staff with those of a nationally representative sample of US adults. The two groups largely agree on what AI can already do, but diverge sharply on future progress: Anthropic staff overwhelmingly expect AI to reach a Nobel-level discovery by the end of the decade, while only a minority of US adults do. Staff also anticipate far larger time savings on AI-suited tasks and expect AI to handle most of their own current work within that horizon, whereas US workers expect far less. On adoption, the general public sees human preferences, privacy concerns, reliability, and regulation as the main obstacles, rather than technology itself. The most notable finding is that both groups reason similarly about labor markets: people who expect AI to do more of their own work also forecast higher unemployment, and at equal levels of expected automation, both groups predict comparable unemployment rates. This suggests the disagreement is about capability timelines, not economic logic.