In a classic social psychology experiment by Fritz Heider and Marianne Simmel (1944), subjects watched a short silent film depicting the motions of two triangles and a circle around a box with a door, and reported rich stories of jealous lovers, fights, entrapments, and daring escapes. A grand challenge of human-like AI is to automate this reasoning, to build a system capable producing similar narratives when presented with novel movies in the style of Heider and Simmel’s original film. Advances in machine learning and text generation are readily applicable to the problems of perception and narration, but the difficult challenge is interpreting the perceived actions of these shapes as a coherent narrative. In this talk, I’ll discuss my efforts to address the interpretation problem using logical abduction, where the system tries to explain the observed behavior of the characters given a knowledge base of commonsense axioms about human psychology and sociology. I’ll describe Etcetera Abduction, a new approach to probability-ordered logical abduction where conditional and prior probabilities are reified as literals in knowledge base axioms, enabling defeasible abductive reasoning in first-order logic.