Visual theory of behavior. An actor robot (black circle) is programmed to move towards the nearest food (green circle) that it can see, and consume it. Sometimes (A), the nearest green circle is directly visible to the actor, but sometimes (B) the nearest green circle is occluded by an obstacle. When occluded, the actor will move towards the closest visible circle, if any. After watching the actor act in various situations, an observer-AI learns to envision what the actor robot will do in a new, unseen situation (C). The observer’s prediction is delivered as a visualization of the actor robot’s “trajectory smear” (D). This entire reasoning process is done visually, sidestepping the need for symbols, logic, or semantic reasoning.