Can AI produce scientific theory?
In brief
- Current AI models like ChatGPT do not produce scientific theories but manipulate data through induction.
- Major scientific theories emerged via hypothesis-driven approaches, distinct from machine learning methods.
- The author urges combining AI with human expertise to explore new theories, without concluding on AI's ability to think.
Axel Cypel, winner of the 2024 Roberval Prize for his essay *Voyage au bout de l'IA*, was asked whether artificial intelligence can generate scientific theories. He argues that current models like ChatGPT manipulate data without understanding it, relying solely on formal induction. Major scientific theories, from Galileo to Einstein, emerged through a hypothesis-driven approach, fundamentally different from machine learning.
The author highlights that scientific theories act as "ontological winches," capable of pulling new concepts or data into existence. He questions whether AI could replicate this process, given its reliance on existing data. Cypel also draws a parallel between scientific theorization and artistic creation, both requiring radical novelty.
For the author, the question transcends technical aspects: it challenges the very nature of creation and innovation. He concludes by advocating for combining AI's strengths with human expertise to explore new theories, without definitively answering whether machines can truly think.
This summary gives you the essentials. The full article, its figures and quotes are on ActuIA.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error