Algorithmic Information Theory and Novelty Generation
Simon McGregor · 2007
This paper discusses some of the possible contributions of algorithmic information theory, and in particular the central notion of data compression, to a theoretical exposition of computational creativity and novelty generation. I note that the formalised concepts of pattern and randomness due to algorithmic information theory are relevant to computer creativity, briefly discuss the role of compression in machine learning theory and present a general model for generative algorithms which turns out to be instantiated by decompression in a lossy compression scheme. I also investigate the concept of novelty using informationtheoretic tools and show that a purely “impersonal” formal notion of novelty is inadequate; novelty must be defined by reference to an observer with particular perceptual abilities.