Emergent kin selection of altruistic feeding via non-episodic neuroevolution

TL;DR

The first demonstration of kin selection emerging naturally within a population of agents undergoing continuous neuroevolution is offered, finding that zero-sum transfer of resources from parents to their infant offspring evolves through kin selection in environments where it is hard for offspring to survive alone.

Abstract

Kin selection theory has proven to be a popular and widely accepted account of how altruistic behaviour can evolve under natural selection. Hamilton's rule, first published in 1964, has since been experimentally validated across a range of different species and social behaviours. In contrast to this large body of work in natural populations, however, there has been relatively little study of kin selection \emph{in silico}. In the current work, we offer what is to our knowledge the first demonstration of kin selection emerging naturally within a population of agents undergoing continuous neuroevolution. Specifically, we find that zero-sum transfer of resources from parents to their infant offspring evolves through kin selection in environments where it is hard for offspring to survive alone. In an additional experiment, we show that kin selection in our simulations relies on a combination of kin recognition and population viscosity. We believe that our work may contribute to the understanding of kin selection in minimal evolutionary systems, without explicit notions of genes and fitness maximisation.

Figures (5)

• Figure 1: (A) A snapshot of the simulation environment. (B) an illustration of the agent architecture, observation and action spaces. Each agent receives an 11x11x4 observation of its local environment--this is fed through a convolutional neural network (CNN) to produce a vector of action probabilities. (C) an example illustration of the energy level over time for a single agent. (D) an illustration of a short action sequence in which an agent moves onto a food resource, eats the food resource, reproduces and then feeds their offspring.
• Figure 2: The effect of varying the three chosen simulation parameters on (a) the infant survival rate (ISR) with feeding disabled, and (b) the estimated benefit of being fed as an infant ($\hat{b}$), given by the average percentage increase in lifespan for infants that are fed relative to those that aren't. Error bars in all plots represent bootstrapped 95% confidence intervals over 20 seeds per parameter value.
• Figure 3: The relationship between the estimated benefit to infants of being fed and both the amount and selectivity of feeding observed, shown separately for each of the three experimental parameters we varied (and combined in the rightmost column). Each scatterplot point represents a single 500k-timestep simulation run (with values averaged over the final 50k timesteps); regression lines (with 95% confidence intervals) are shown in green. Note that the $y$-axis shows $\log$(measure) for both amount and selectivity.
• Figure 4: The relationship between the amount of feeding behaviour observed and its selectivity towards agents' own offspring, shown separately for each of the three experimental parameters we varied (and combined in the rightmost column). Each scatterplot point represents a single 500k-timestep simulation run (with values averaged over the final 50k timesteps); regression lines (with 95% confidence intervals) are shown in green.
• Figure 5: The effect of disabling kin recognition (left) and reducing population viscosity (right) on the relationships between feeding benefit and feeding amount and selectivity. Each dot represents a single 500k-timestep simulation run. Simulations were run over the same values of the three experimental parameters as before, and as before we ran 20 seeds per parameter value. 95% confidence intervals are shown for all regression lines.