AI decision-making is now common in self-driving cars, patient diagnosis and legal consultation, and it needs to be safe and trustworthy. Researchers have been trying to demystify complex AI models by ...