Considering metabolism in cognitive models

“The human brain is a very hungry organ,” says David Colaçofrom the Munich Center for Mathematical Philosophy (MCMP) at LMU. “Although it makes up just two percent of body mass, it consumes around a fifth of the body’s energy.” At the same time, it is “more energy-efficient than most modern computers.”

In a study with Dr. Philipp Haueis from Bielefeld University, Colaço investigates how such metabolic findings can help improve mathematical and theoretical models of cognitive capacities. “Although it is known that the metabolism of the brain plays a decisive role in processes like memory, perception, and attention, models rarely address metabolic factors,” says Colaço, who researches at the interface of philosophy and neuroscience at the Chair of Philosophy of Science at MCMP.

The two researchers systematically compiled and analyzed existing knowledge about brain metabolism for the first time. This included empirical findings about energy consumption, neuronal scaling – the relationship between brain size, number of neurons, and energy requirements – and the energy expenditure of information processing in the brain.

How thinking arises at the neurobiological level

Their results, which are based on a literature review and philosophical argumentation, have now been published in the journal Behavioral and Brain Sciences. In their paper, Colaço and Haueis show how insights into brain metabolism can be used in the future: “Firstly, it can help evaluate whether existing models of cognition and behavior are biologically plausible – and consistent with the energy limits of the brain,” explains Colaço. “After all, natural energy consumption puts a limit on which computational processes in the human brain are possible in the first place – and how much information it can process.”

For example, memory models that are based on stable biochemical equilibrium states are unrealistic from a metabolic point of view – because such systems would quickly succumb to thermal death. Secondly, knowledge about metabolic processes can be used to generate new, more biologically plausible models. “If, say, we want to computationally investigate the relation between brain structure and information processing, these metabolic considerations can serve as a starting point,” says Colaço.

At the neural level, for example, the study revealed energetic trade-offs: Thus, thin axons save energy per bit of information transmitted, but they transmit information more slowly; whereas thick axons are faster, but transmit less information per unit of energy. Such considerations could be factored into computational models of mental processes in the future. As the authors point out, few researchers in philosophy and neuroscience have treated the relationship between metabolism and cognition.

With their work, Colaço and Haueis want to provide fresh impetus – for research into energy turnover when people are engaged in mental effort; for the question as to how thinking arises at the neurobiological level; and for comparisons between biological and artificial intelligence.

Publication

Philipp Haueis & David J. Colaço: Metabolic considerations for cognitive modeling, 2025