Brain energetics and the connectionist concept in cognitive neuroscience

Abstract

One of the central paradigms of modern neuroscience is the connectionist concept suggesting that the brain's cognitive functions are carried out at the level of neural networks through complex interactions among neurons. This concept considers neurons as simple network elements whose function is limited to generating electrical potentials and transmitting signals to other neurons. Here, I focus on the neuroenergetic aspect of cognitive functions and argue that many findings from this field challenge the concept that cognitive functions are performed exclusively at the level of neural circuits. Two of these findings are particularly foretelling. First, activation of the cerebral cortex in humans (sensory stimulation or solving cognitive problems) is not associated with a significant increase in energy demand. Second, the energetic cost of the brain per unit mass in primates, including Homo sapiens, is approximately proportional to the number of cerebral neurons, but not to the number of synapses, complexity of neural networks or the level of brain's intellectual abilities. These findings contradict the predictions of the connectionist concept. Rather, they suggest that cognitive functions are generated by intra-neuronal mechanisms that do not require much energy. In this context, inter-actions among neurons would serve to coordinate activities of neurons performing elementary cognitive functions. This function of the network mechanisms also does not require much energy.

The controllosphere: The neural origin of cognitive effort.

“…In both of these cases, seemingly complex neural representations can be described with more elementary mathematical algorithms. That said, the fact that the modern incarnation of connectionist networks, deep learning models, provides increasingly rich explanations of neurocognitive phenomena (Churchland & Sejnowski, 2016; Cichy & Kaiser, 2019; Doerig et al, 2023; Kriegeskorte, 2015; Peters & Kriegeskorte, 2021; Saxe et al, 2021; Yamins & DiCarlo, 2016) attests that the correspondence between the neural and algorithmic levels may be more than just an analogy (notwithstanding various objections: Arshavsky, 2023; Bowers et al, 2022; Pang et al, 2023). It thus remains to be determined how deep into the brain the controllosphere metaphor can be pressed.…”