A central problem in the cognitive sciences is identifying the link between consciousness and neural computation. The key features of consciousness-including the emergence of representative information content and the initiation of volitional action-are correlated with neural activity in the cerebral cortex, but not computational processes in spinal reflex circuits or classical computing architecture. To take a new approach toward considering the problem of consciousness, it may be worth re-examining some outstanding puzzles in neuroscience, focusing on differences between the cerebral cortex and spinal reflex circuits. First, the mammalian cerebral cortex exhibits exascale computational power, a feature that is not strictly correlated with the number of binary computational units; second, individual computational units engage in noisy coding, allowing random electrical events to gate signaling outcomes; third, this noisy coding results in the synchronous firing of statistically random populations of cells across the neural network, at a range of nested frequencies; fourth, the system grows into a more ordered state over time, as it encodes the predictive value gained through observation; and finally, the cerebral cortex is extraordinarily energy efficient, with very little free energy lost to entropy during the work of information processing. Here, I argue that each of these five key features suggest the mammalian brain engages in probabilistic computation. Indeed, by modeling the physical mechanisms of probabilistic computation, we may find a better way to explain the unique emergent features arising from cortical neural networks.
© 2023 Cognitive Science Society LLC.