Import AI 458: Reckoning with the Future and a Singularity Story

The 2026 Cosmos HAI Lab Lecture, delivered at the Human-Centered AI Lab (HAI Lab) within the Institute for Ethics in AI at the University of Oxford in collaboration with the Cosmos Institute, explores the implications of rapid AI progress. The speaker, a co-founder of Anthropic, argues that AI is not a normal technology and that society must choose between 'exploring the future' or 'retreating from the present.' The lecture references the Epoch Capabilities Index (ECI) to illustrate AI's accelerating benchmarks: AI passed the bar exam in March 2023, achieved silver in the International Math Olympiad (July 2024) and gold (July 2025), co-authored mathematical proofs (2025), and systems like Claude Mythos found novel software flaws. The speaker predicts that AI could achieve recursive self-improvement within two years, leading to autonomous companies, Nobel-level discoveries by human-AI teams (April 2027), and bipedal robots doing useful work (April 2028). At Anthropic, the release of Opus 4.6 in late 2025 led to most code being written by Claude, a shift toward verification layers, and a new 'trust economy.' Hiring now favors early-career LLM experts and experienced visionaries. The lecture includes a fictional story, 'As I Lay Dreaming,' about a family using life extension and AI to revive a suspended daughter during 'The Uplift.' The talk was given on Wednesday, May 20, 2026, at Oxford University, edited by Santi Ruiz.