Representing As Compressing. Bridging the Biological and Artificial Perceptual Systems

Psychology, statistics, computer science, machine learning, philosophy and many other disciplines routinely use information-talk in regard to a whole array of both biological and artificial structures. Thus, it is generally uncontroversial to hold that regardless of the differences and peculiarities of given cognitive systems, they are all information-processors. However, surprisingly, the notion itself (especially in the neural context) for long has escaped thorough philosophical exploration, as the other core concepts, such as representation (e.g. Chemero 2009, Ramsey 2007, Cummins 1996) or computation (Piccinini 2017, Fresco 2014) are frequently subject to. In order to do that I rely ideas from both information theory and philosophy, such as Daniel Wilkenfelds recent ‘understanding-as-compression’ theory that frames compression-likes processes as maximizing scaffolding-to-output ratio. My aim is to show that the highly-structured nature of the informational makeup of the environment in which perception is performed (as demonstrated by the Natural Scence Statistics research program), it makes intuitive sense to think of perceptual processes as compression-like. Compression constitutes canonical, well-understood and elegantly formally defined information-processing (Shannon 1948, Chaitin 1977) which could help explain how the perceptual system generalizes and generates information. Later on I assess the potential objections about the inability of information-based explanations to account for representations. To dismiss them, I turn to the recent theories of the naturalization of informational content (Isaac 2019, Skyrms 2010). I argue that it opens up the possibility of comparing artificial and biological signalling systems in the virtue of representational similarity. I mobilize a host of empirical evidence from both experimental neuroscience and artificial neural systems in order to assess how well the latter might model or explain the former. I also investigate how structural representational similarities between different cognitive agents can be traced due to the relation of infomorphism.

Recommended citation: Mamak, W (draft). Representing As Compressing. Bridging the Biological and Artificial Perceptual Systems: A Case for Information-based Explanations in Cognitive Science.
Download Paper