Talk (Work in Progress): David Colaço (MCMP)

Title:

Metabolicconsiderations in cognitive modeling

Abstract:

The human brain makes up just 2% of body mass but consumes closer to 20% of the body’s energy. Nonetheless, it is significantly more energy-efficient than most modern computers. Although these facts are well-known, theoretical models of cognition rarely account for metabolic factors, despite recognition that cognition is bounded by biological resources.

We argue that metabolic considerations should be integrated into models of cognitive activities. We distinguish two uses of metabolic considerations in modeling. First, metabolic considerations can be used to evaluate the adequacy of models. Evaluative uses function like explanatory constraints. Metabolism limits which types of computation are possible in biological brains, as it is relatively fixed because of evolutionary optimization of energy use. Further, it structures and guides the flow of information in neural systems. We illustrate evaluative uses by a comparison of linear and cyclical models of synaptic memory. Second, metabolic considerations can be used to generate models. In this sense, they fulfill functions of exploratory modeling. They provide: (1) a starting point for inquiry into the relation between brain structure and information processing, (2) a proof-of-concept that metabolic knowledge is relevant to cognitive modeling, and (3) a potential explanation of how a particular type of computation is implemented. An example is the use of aerobic glycolysis to generate a model of prediction error transmission in thin, informationally efficient axons.

Through presenting these uses of metabolic considerations in cognitive modeling, we argue that cognitive models should be consistent with the brain's metabolic limits, and modelers should assess how their models fit within these bounds. Further, consideration of metabolism alsooffers insights into debates over the ideas of multiple realization and medium independence, speaking to the question of what physical details should be considered when theorizing about cognitive systems. These insights also show how metabolism can inform how we differentiate the activities of biological and artificial computational systems, speaking to the role of metabolism inaccounting for the “cognitive” activities of AI.

Jointwork with Philipp Haueis (Institute of Philosophy at Leibniz University Hannover)