Metazoans are capable of gathering information from their environments and respond in predictable ways. These computational tasks are achieved by means of more or less complex networks of neurons. Task performance must be reliable over an individual's lifetime and must deal robustly with the finite lifespan of cells or with connection failure - rendering aging a relevant feature in this context. How do computations degrade over an organism's lifespan? How reliable can computations remain

more »

... out? In order to answer these questions, here we approach the problem under a multiobjective (Pareto) optimization approach. We consider a population of digital organisms equipped with a neural network that must solve a given computational task reliably. We demand that they remain functional (as reliable as possible) for an extended lifespan. Neural connections are costly (as an associated metabolism in living beings) and degrade over time. They can also be regenerated at some expense. We investigate the simultaneous minimization of the metabolic burden (due to connections and regeneration costs) and the computational error and the tradeoffs emerging thereof. We show that Pareto optimal designs display a broad range of potential solutions: from small networks with high regeneration rate, to larger and redundant circuits that regenerate slowly. The organism's lifespan and the external damage rates are found to act as an evolutionary pressure that improve the exploration of the space of solutions and poses tighter optimality conditions. We also find that large damage rates to the circuits can constrain the space of the possible and pose organisms to commit to unique strategies for neural systems maintenance.