AI Comes Up With Bizarre Physics Experiments. But They Work. | Quanta Magazine

Abstraction: AI designs counterintuitive physics experiments outperforming human-designed setups

Key points:

• Rana Adhikari (Caltech) used Mario Krenn's PyTheus AI software to redesign LIGO gravitational wave detectors; AI proposed a 3km ring to recirculate light, exploiting obscure Russian theoretical principles, potentially improving LIGO sensitivity by 10-15%
• PyTheus represents experiments as mathematical graphs and optimizes them toward desired quantum states; Krenn's student used it to find a simpler entanglement-swapping design than Zeilinger's 1993 classic — confirmed experimentally in 2024 by a Chinese team
• AI found Lorentz symmetries in LHC data without any prior physics knowledge (Rose Yu, UC San Diego)
• Kyle Cranmer's ML system derived a dark matter density formula that better fits data than human-made equations, though it lacks physical interpretation
• AI has not yet invented new physics concepts but is finding non-trivial patterns; researchers see LLMs as potentially automating hypothesis generation next

Connections: Ligo · Caltech · AI Scientific Discovery · Quantum Entanglement · Machine Learning

Source: https://www.quantamagazine.org/ai-comes-up-with-bizarre-physics-experiments-but-they-work-20250721/