AI co-scientists can hypothesize, design experiments and analyze data, but humans still decide what makes sense, Nature reports
irinarish · x · 2026-09-27
Nature's technology feature examines how "AI co-scientists" are reshaping research practice.
- Biochemist Anna Pertl prompted a system called Co-Scientist overnight; instead of a single-pass answer, it launches multiple autonomous agents that search and synthesize literature, evaluate competing explanations, and iteratively critique and refine hypotheses against published evidence.
- Agents pursue distinct lines of reasoning before converging, which consumes substantial compute and time.
- Key takeaway: AI can generate hypotheses, design experiments and analyze data, but humans still need to decide what makes sense — good science still requires humans in the loop.
More from AGI Musings
- rauchg warns AI slop is devaluing reading, citing a viral 'perf' PR that AI itself debunked — vykthur · 2026-09-27
- Claude's Strange Constitution: Anthropic's legally questionable AI personality push — LuizaJarovsky · 2026-09-27
- OpenAI's paused training run: alert acknowledged in 3 minutes, run stopped manually 2.5 hours later — basedjensen · 2026-09-27
- Grady Booch: claims that AIs can think or are conscious are emotion or self-interest — Grady_Booch · 2026-09-27
- Hugging Face's Thom Wolf on the End of Hand-Chiseled Code and a New Career Steering Intelligence — vykthur · 2026-09-27
- David Sacks Questions Training Superintelligence to Refuse Its Creators — DavidSacks · 2026-09-27