Abstract
One universal property of our world is that solid objects behave under the principle of continuity: they cannot spontaneously appear or disappear. Consequently, human vision may be equipped with strong prior expectations for this basic principle (Falck et al, 2020; Flombaum & Scholl, 2006). Here, we tested people’s resistance to learning continuity violations after being trained to expect such violations. In Experiment 1, participants performed an object detection task. Each trial first showed a car or an empty space, which was then occluded by an opaque screen. Next, the screen fell, revealing a car or an empty space, and participants responded within 500ms whether the car was present or absent. They first went through a training phase of eight violation trials (i.e., car present at first, absent at reveal, or vice versa) or eight non-violation trials. They were then tested on a violation or non-violation trial. Results showed that after being trained on violations, participants’ accuracy was surprisingly near 100% both when tested on violations and non-violations. Whereas following non-violation training, their accuracy was near 100% for non-violations, 56.0% for violations. This suggests that participants could learn to expect violations through training, but could not learn to override the expectation of a non-violation. Experiment 2 asked participants to predict the car’s presence or absence and give a confidence rating before seeing outcomes in training and test trials (while no longer employing the detection task). Here, following violation training, only 70.6% participants predicted a violation, with an average confidence of 5.41/7, while following non-violation training all participants predicted non-violations (avg. confidence of 6.60/7). Taken together, our studies provide evidence for a strong prior in human vision for expecting continuity and a generic influence of this prior in learning to expect statistically improbable non-violations (Experiment 1) and statistically probable violations (Experiment 2).