Lifetime learning of division of labor through social sanctioning
For the full paper, please see: Yaman A*, Leibo JZ*, Iacca G, Lee SW. The emergence of division of labor through decentralized social sanctioning. 2023 (* equal contribution)
(Accepted, Proceedings of the Royal Society B)
Division of labor is crucial in specialization of individuals in different roles and working together to achieve greater welfare as a group than would be possible alone. However, such beneficial arrangements may be difficult for groups of lifetime-learning individuals to discover when they are driven only by self-interest. This is because some critical roles are less rewarding than other roles, so all individuals prefer someone else to take them on. However, many computational models of lifetime learning formulate this process as a maximization of individual payoffs. Such a formulation cannot, on its own, account for the learning division of labor. When individuals are driven by self-interest they cannot learn to perform roles that do not pay as well as other roles, yet such roles are often necessary for the group to function and achieve a high overall welfare. Consequently, a fundamental question arises:
How can division of labor emerge in groups of self-interested lifetime-learning individuals?
Decentralized social sanctioning. Here, we hypothesized that social norms, which we take to be patterns of social sanctioning, are sufficient to incentivize individuals in groups to select prosocial role choices, thereby enabling group-level division of labor to emerge from self-interested lifetime-learning. To test this hypothesis, we propose a model where lifetime learning is shaped by social sanctioning. Such social norms work by redistributing rewards within the population to disincentivize antisocial roles while incentivizing prosocial roles that do not intrinsically pay as well as others.
Promoting convergent evolution through complexity regularization. Introducing complexity regularization leads to the convergent evolution of simpler social norms. This allowed the emergence of similar and simpler norms in multiple independent evolutionary processes.
Division of labor in spatial games. We designed two spatially structured games to investigate lifetime-learning of division of labor. These games are designed in a way to require a group of individuals to learn a distribution of roles. However, although some roles payoff sub-optimally, they are required to be performed by some individuals to improve the overall welfare of the group. These games introduce excellent scenarios for testing lifetime-learning processes of division of labor.
Impact and implications
Our work can stimulate novel research directions in multiple research fields, including computational biology/neuroscience, computer science and artificial intelligence.
Computational biology/neuroscience. Our model of social sanctions in conjunction with lifetime-learning can account for the emergence of cooperation and division of labor in groups of purely self-interested lifetime-learning individuals. Furthermore, the design of the games for studying the emergence of division of labor can inspire future research in real world experiments to confirm the hypotheses of the use of social sanctions.
Computer science/artificial intelligence. From the game-theoretical point of view, the use of social sanctions can transform the payoffs into a new effective game with different Nash equilibria. Thus, when social norms are regarded as equilibrium selection devices, they can allow navigation of the society into a good equilibrium while avoiding the bad ones. Our findings can inspire future research in the use of social sanctions and similar mechanisms to transform games in similar ways to settle into an equilibrium that could be beneficial for the group.