Researchers at Caltech and MIT have developed a theorem, dubbed IT-Ï€, that identifies the most informative variables for predicting outcomes in AI and physics models while removing measurement units.