FROM THE FRONTIER
AI is designing scientific experiments that leave experts stumped
Made with Midjourney
LIGO’s gravitational wave detector is one of the world’s most sensitive scientific instruments. It can measure changes smaller than the width of a proton. But after spotting the first gravitational wave in 2015, physicists had wanted to push sensitivity even further to detect different types of cosmic events.
Caltech physicist Rana Adhikari decided to turn to AI. His team fed AI systems the building blocks for complex physics experiments. Unlike humans, the AI had no preconceptions about what “good” experimental design should look like, allowing it to explore configurations that would seem ridiculous to human scientists.
The results looked like “alien things” (at first). The machine learning system created such bizarre, asymmetrical designs for improving LIGO’s gravitational wave detectors that researchers initially dismissed them as incomprehensible. One researcher said his students would have been rejected for proposing anything similar.
But the AI was onto something. After months of analysis, the team realized the system had rediscovered obscure decades-old theoretical principles from Russian physicists — ideas that had never been tested experimentally. The AI’s counterintuitive design could make LIGO 10-15% more sensitive — improvement for an instrument that could already sense the distance to Alpha Centauri down to the width of a human hair.
Similar breakthroughs are happening across the sciences. Just last week, scientists created AI-generated antibiotics that can kill gonorrhoea and MRSA superbugs. At Nanjing University, researchers used AI to create quantum entanglement through a method no human had conceived. AI systems are also churning out new equations for dark matter and discovering symmetries in particle data. The pattern is clear: AI is thinking outside human assumptions to solve bottlenecks that have persisted for decades.