Engrams reconsidered: three windows on memory traces

Abstract

At the heart of this thesis is a simple problem-we remember, and yet no memory has ever been found. Does this problem derive from the experimental limitations of neuroscience, or is there a problem with our thinking? In other words, is the problem difficult because it is a hard scientific problem, or because we have assumed too much of the ideas we take for granted? This thesis seeks to improve how we think about the concepts of "memory" and "information processing and storage" in biological systems, through pooling together established literature and emerging research in neuroscience, philosophy, and machine learning. Through a combination of criticism, cross-fertilisation and elaboration of new ideas, it synthesises and presents more critical ways to ask questions about memory for a variety of audiences (neuroscience, philosophy, computer science/machine learning, and the cognitive sciences generally) with differing interests and objectives, and which each field can develop further. Chapter 1 provides a critical assessment and taxonomy of memory trace concepts and underscores the need for neuroscience to expand beyond characterising putative memory traces vehicles. Chapter 2 showcases what engagement with memory neuroscience has to offer neuroscience-inspired artificial intelligence, such as a shift away from spike-centric modelling. Finally, Chapter 3 outlines the shortcomings of philosophical accounts of representation for focusing only on spiking activity, and offers a novel defence of why existing concepts of computation and information processing cannot be applied to living systems.