The Perception of Motion: Michotte Revisited

This paper extends the range of artificial-intelligence vision programs to the domain of moving objects, using the work of Michotte as a source of data for a computer program which purports to ‘see’ the action types he examined. Given the thesis that internal descriptions of a nonlinguistic type provide the basis for a constructive, knowledge-based theory of perception, the properties of a symbol-manipulating system capable of supporting this are explored as a challenge to the Michotte notion of “direct perception without interpretation”.