An artificial intelligence proposes experimental setups no human would think of – and outperforms established designs in the process. A team led by Mario Krenn (University of Tübingen) and TU Wien describes in the journal Nature how algorithms are reshaping the way physics experiments are planned.
An unimaginably vast search space
The software, called PyTheus, arranges optical components such as lasers, lenses, mirrors and detectors into new configurations. „It’s a massive optimization problem. There is an unimaginably vast space of possible experiments that can be set up with the available components,“ Krenn explains. The computer-discovered layouts often break with conventional design rules – yet still deliver more precise results or enable entirely new measurements.
From microscopy to gravitational-wave detectors
The authors, with first author Jonathan Klimesch, survey where such automatically designed experiments are already being tested. The fields of application reach far beyond quantum optics:
- higher-resolution electron microscopy
- more sensitive particle detectors
- gravitational-wave observatories
- setups for fusion research
Why some designs work remains unclear
One catch remains: the AI produces layouts that measurably work – but the physical principle behind them is often hard to explain. The researchers therefore see the tools less as a replacement than as an idea generator that complements human ingenuity. The review appeared on 2 September 2026 in Nature (Volume 657, pp. 47–58).
Sources: Nature · University of Tübingen (Tübingen AI Center)
