Can AI trained on historical data predict responses to never-before-seen policy changes and other shocks? Given enough parameters, can we dispense with the Lucas Critique? I genuinely don't know, but let me provide some food for thought from a working paper I'm excited about. /1