An AI system has recovered several cornerstone laws of mechanics from experimental data, offering a test of whether machines can move from spotting patterns to building scientific explanations.
Researchers at Peking University describe AI-Newton in Physical Review Research, published November 4. The framework works across multiple experiments, proposing interpretable physical concepts and progressively extending the laws it constructs to broader settings.
Applied to a large, noisy mechanics dataset, it rediscovered Newton’s second law, conservation of energy and universal gravitation. The authors report that it operated without supervision or prior physical knowledge.
That distinction matters. Work on making AI obey physical laws typically supplies scientific constraints to guide predictions. AI-Newton instead investigates whether useful concepts and general relationships can emerge from observations.
The study also adds a concrete example to efforts to build AI tools for scientific discovery. Its appeal lies in explanations that researchers can inspect, rather than predictions alone.
Still, rediscovering established mechanics is a proof of concept. It does not demonstrate that the system can uncover unknown laws across scientific disciplines. The next challenge is establishing how reliably this approach works when the governing relationships remain uncertain.