Representation Within a Single Neuron

I start with two puzzles about intelligence. First, intelligence does not depend on neuron number in the way one might intuitively expect. A bee, with about 1 million neurons, can perform cognitive tasks that severely test a human, with about 100 billion. A nematode worm, with 302 neurons, can do tasks such as avoidance learning, flexible decision-making and sensorimotor integration. Second, attempts to emulate even the simplest nervous systems with artificial neural networks have failed (so far) to live up to expectations. I then present a hypothesis that solves both puzzles: in smaller brains, individual neurons do more of the computational work. Computational tasks that in large brains require large networks of neurons can sometimes be accomplished in small brains within a single neuron. I motivate this hypothesis with an example from Caenorhabditis elegans. I close by reflecting on the potential consequences of this hypothesis, if true, for AI and for the evolution of intelligence, cognition and consciousness.