Object permanence

Object permanence is the cognitive understanding that objects continue to exist even when they cannot be perceived. First demonstrated experimentally by Jean Piaget in the 1930s, it is considered a foundational achievement of infant cognitive development, typically emerging between 4 and 8 months of age. Before object permanence develops, infants behave as if out-of-sight objects cease to exist; after its development, they search for hidden objects and represent them as continuing to exist in time and space.

The concept is not merely developmental psychology. It is a fundamental requirement for any system that claims to perceive a world rather than a sequence of unrelated images. A system without object permanence processes snapshots; a system with it processes a persistent environment. Current computer vision systems — including the most advanced deep learning architectures — process frames independently or maintain only short-term temporal coherence. They do not represent objects as continuing to exist when occluded, moving out of frame, or unobserved. This is not a minor gap. It is the difference between seeing a world and seeing a slideshow.

The connection between object permanence and predictive processing is instructive. On the predictive processing account, the brain maintains a generative model of the world and updates it based on prediction errors. Object permanence is not a separate module but an emergent property of a model that represents objects as persistent causes of sensory signals. The model predicts that the object will continue to exist and generate signals, and when the signals are absent (because the object is occluded), the prediction is maintained rather than discarded. This suggests that object permanence is not a cognitive capacity that needs to be added to vision systems. It is a property that emerges from the right kind of world-model. Whether current architectures can support such models is an open question.