The task of controlling a complex enterprise is routinely delegated to several agents to cope with the curse of dimensionality. The agents, having limited abilities and views of the enterprise, typically compete with one another only to reach suboptimal decisions. Whatever the enterprise, the work of its agents can be modeled by a dynamic game. This paper illustrates how altruistic agents can drive the decisions to optimal attractors for a family of games. Specifically, it develops an algorithm to find optimal altruistic responses to improve convergence to, and location of, attractors in games arising from the optimization of quadratic functions. Further, the paper provides evidence that the agents can learn altruistic responses from past experience.
