KI-Zeitschrift - Korrekturabzug - nur zum internen Gebrauch!

In cognitive systems, one often distinguishes between phylogeny and ontogeny. Phylogeny refers to the initial configuration of the system and its evolution from generation to generation. On the other hand, ontogeny refers to the learning and development of a given system during its lifetime. The issue that arises in this context is the requirements for the minimal architecture for a cognitive vision system. There are two perspectives on this, depending on whether one takes a cognitivist stance or an emergent stance. In the cognitivist stance, the issue comes down to the balance between required ‘pre-knowledge’ and acquirable knowledge. Or, put another way, how much does one need to know and to be able to do in order to be capable of learning new things, such as concepts or actions? That is, we need a clear cut set of conditions under which certain learning can take place. In the emergent stance, there is a trade-off between phylogenic configuration and ontogenic development. Phylogeny determines the visuo-motor capability that a system is configured with at the outset and which facilitates the system’s innate behaviours. Ontogenic development gives rise to the cognitive capabilities that we seek. Since we don’t have the luxury of having evolutionary timescales to allow phylogenic emergence of a cognitive system we can’t wait around to evolve a cognitive system from nothing we must somehow identify a minimal phylogenic state of the system. In practice, this means that we must identify and effect visuomotor capabilities for the minimal reflex behaviours that ontogenic development will subsequently build on to achieve cognitive behaviour. Put simply: we need to decide what visual processing capabilities are needed for a minimal emergent cognitive vision system.