The Evolutionary Origin of Nervous Systems and Implications for Neural Computation

If neurones are the answer, then what was the question? Nervous systems are remarkably conserved throughout the animal kingdom. This conservation indicates that their core design and function features were established very early in their evolutionary history. It also indicates that all nervous systems solve the same fundamental task in essentially the same way, and have been doing so since their inception. Understanding when and why animals evolved nervous systems could help us identify that fundamental task. It could also provide context to help us determine how nervous systems solve that task so well. This thesis studies the evolutionary origins of nervous systems and the ecological conditions under which they evolved. It outlines fossil, ecological, and molecular evidence to argue that animals evolved nervous systems soon after they started eating each other, 550 million years ago. When animals started eating each other, they must have experienced a selection pressure to react to threats and opportunities in a precisely-timed manner. Nervous systems could have been one response to this selection pressure. This thesis then provides a quantitative ecological argument that, before animals started eating each other, they were forbidden from evolving neurones, but after, they were forced to either evolve neurones or be driven to extinction by a competitor that did. If predation was the question, then why are neurones the answer? This thesis argues that the first afferent sensory neurones were threshold detectors that produced spikes to alert animals of the proximity of other animals. Those spikes prompted a reaction, such as striking or fleeing, the instant that a state of the world became critical. Ancient animals would have found utility in devices that signalled as soon as the probability that a world state was critical exceeded some threshold. Extant neurones are well-adapted to solve these tasks; they produce an all-or-nothing spike when their membrane potential exceeds a threshold. Given that their core function features have hardly changed since their inception, they may still be functioning in an analogous manner. When animals evolved devices that spiked conditional on world states, it became possible for them to infer those states, given the spiking output of the devices. This thesis shows this principle explicitly with idealised models of predator-prey interactions. If an animal could implement Bayesian inference given sensory spikes, then that animal would extract all possible information about the world from those spikes; no ‘neural coding’ strategy can contain more information about the world. In situations requiring fast decision-making and reaction in accordance with potentially-fatal states of the world, an animal that uses Bayes’ rule to make inferences about the world from sensory spikes would outperform all competitors. Extant nervous systems could implement Bayesian inference by functioning as a biological analogue of a particle filter