When it comes to AI, trust used to mean one thing — accuracy. Does the model predict correctly? Then we started asking harder ...