A Machine That Daydreams

Some aspects of human cognitive processing across experiential cases in episodic memory seem quite different from conventional artificial reasoning by logical rules, such as that seen in CBR systems. One difference is that in humans, linkages between particular experiences can apparently be made in a number of qualitatively different ways, forming recollective chains of memories along different dimensions. Data-driven, creative, free-association from one memory to the next does not appear to be economically described by rules. Efforts to enable computers to deal with cultural content such as narratives in new ways could benefit from sequential indexing of this kind, provided that the conceptual representations are rich enough, and that a way can be found of modeling the emotional impact each elicits. A conceptual-graphbased FGP (Fetch, Generalise, Project) machine using a knowledgebase of archetypical narratives enhanced with affect is described, which demonstrates how such emotive “memorywalks” can be computed.