Social Segregation Dynamics with Reinforcement Learning and Agent Based Modeling.

Is it possible to solve social segregation? Our models show that social integration can be promoted by creating interdependencies among people. Relationships cannot be simply dictated but an environment with proper social rewards is possible to promote.
Social Segregation Dynamics with Reinforcement Learning and Agent Based Modeling.

In this paper we study the dynamics of social segregation and explore ways to promote integration by combining mathematical population models with artificial intelligence [1]. We find that social integration can be promoted by creating interdependencies among people’s actions. Either economic or social, interdependencies promote the human interactions and communication required for building trust, which is the basis of endurable social relationships. Our model reveals that young individuals are perhaps the most receptive population for this type of interventions and the most probable ones for embracing change.

The analysis relies on agent-based models acquired from the science of complex systems. In particular, the Schelling model of social segregation [2]. This model shows that properties as complex as urban segregation can emerge from autonomous individual decisions without the need of centralized enforcement (Figure 1). Segregation is a systematic problem that arises from our own choices. While these models inform about the origins of the social complexity, they are limited for exploring solutions to problems. For this reason, the combination of complexity models with artificial intelligence tools and algorithms enables a further discovery of the hidden space of possible actions [3].

The motivation behind the paper dates back to our childhood, upbringing and life history. Being originally from Venezuela (Alfredo) and Turkey (Egemen), we have witnessed first-hand what it means for societies to get increasingly polarized and segregated. In the case of Venezuela, the population got so fragmented and polarized that it became unable to recognize itself any further, leading to the emergence of multiple conflicts and hostile environments. In previous papers [4,5] we show how urban segregation is inexorably tied to a separation of the space of people’s beliefs and ideas. People learn by imitation [6] and the structure of social networks determines the flows of information. The basis of all polarization lies on a network of segregated interactions [7]. Therefore, creating the space for social interactions is crucial in order to promote social integration.

However, solving a real problem is not as easy as creating a mathematical model. People need to be motivated to establish links and interdependencies among them spontaneously [8]. Behaviors cannot be dictated top-down [9] since unwanted side effects can occur with these types of strategies due to the complexity of human behavior. Solution should emerge bottom-up in order to be robust and effective in time [10].  The complexity of the problem at local scales is too heterogeneous for one-size to be able to fit them all. Governments could use their leadership capacity to create the environmental conditions for social communication to happen and relationships be naturally created. 

By understanding the complex dynamics of societies, we can intervene intelligently. Otherwise, by not handling the relevant information of the system behavior or its members, we will continue to create tomorrow's problems with today’s solutions. In the era of pandemics and social protests due to racial discrimination, it is imperative to understand how human nature works in order to take effective actions towards healthier social environments. Together we can build a world in which purpose is shared, trust in each other is strong, and growth is mutual.


Figure 1. Agents collective behavior for multiple values of segregation parameter (rows) at multiple times (columns). Rows represent outcomes associated to different values of segregation parameter (alpha). Columns show the state of the system at different points of the simulation. In Panel (a) colors indicate the concentration of both types of agents (blue and red). White indicates the average pattern.

Written by:

Alfredo J. Morales and Egemen Sert.


 [1] Sert, E., Bar-Yam, Y. and Morales A. J. "Segregation Dynamics with Reinforcement Learning and Agent Based Modeling." Scientific Reports 10, 11771 (2020)

[2] Schelling, T. C. "Dynamic models of segregation." J. Math. Sociol.1, 143–186 (1971).

[3] Mnih, V. et al. "Human-level control through deep reinforcement learning." Nature 518, 529 (2015).

[4] Morales, A.J., Dong, X., Bar-Yam, Y., Pentland, A. "Segregation and Polarization in Urban Areas." R. Soc. Open Sci. 6, 190573 2019.

[5] Morales, A.J., Borondo, J., Losada, J.C. and Benito, R.M. "Measuring political polarization: Twitter shows the two sides of Venezuela." Chaos Interdiscip. J. Nonlinear Sci.25, 033114 (2015).

[6] Axelrod, R. "The dissemination of culture: a model with local convergence and global polarization." J. Conflict Resol.41, 203–226 (1997).

[7] Granovetter, M. "The strength of weak ties: A network theory revisited." Sociological theory. 201-233. (1983)

[8] Maslow, A. H. "A Dynamic Theory of Human Motivation." (1958).

[9] Taleb, N. N. "Black swans and the domains of statistics." Am. Stat.61, 198–200 (2007).

[10] Ashby, W. R. "Requisite variety and its implications for the control of complex systems." In Facets of systems science, 405–417 (Springer, 1991).

Please sign in or register for FREE

If you are a registered user on Social Sciences Community, please sign in