If AI does the science, can people still explain the result?
Hoda Heidari and Aydin Mohseni propose a test for handing research tasks to AI: can the scientific community explain and contest a result, and can people still choose the questions and judge the meaning? Their paper does not show that AI already outperforms scientists or that the test improves results.
What the source reports
Hoda Heidari and Aydin Mohseni ask what human scientists should keep doing if AI takes on more research tasks. They argue that a scientific community needs the ability to explain, defend and challenge findings. People should also retain the power to set research questions and interpret results. Their proposed test is about which tasks to hand off and which decisions must stay with people. It does not say people must perform every calculation. This is a framework for thinking about science, not a trial of learning or research outcomes. Nor does it prove that AI has already become better than human scientists.
Original source
- Title
- Why and How to Do Science When AI Is Better at It: A Framework Grounded in Human Understanding and Direction
- Author
- Hoda Heidari, Aydin Mohseni
- Publication
- arXiv cs.AI
- Date
- Wednesday, October 28, 2026