Proceedings of the Annual Meeting of the Cognitive Science Society

People use knowledge about physical objects to predict and plan their actions, but this knowledge about objects can bedirectly perceived or simply inferred. In this experiment, participants chose the direction to shoot computerized cannonsto hit targets. These cannons differed in how fast they shot the cannonball, but participants could learn this informationeither from observing the full trajectory of a prior shot, or just observing the outcome. While the cannonballs initial speedcan be determined from the end state alone, additional information in the full trajectory might improve these estimates.We find that performance is only worse in the end-state trials if these trials were tried first; if participants judged the fulltrajectory trials first, their performance did not decline on the end-state trials. We explore this order effect using a modelof noisy physical inference that assumes learning from prior trial blocks.