Backup mandate Research Council: The impact of emergent dynamics in realistic decentralized multi- agent systems with a heterogeneous population of artificial agents

Project Details


With the current rate of progress in artificial intelligence, the number of complex multi-agent systems with advanced artificial agents will only grow over time. Since interactions in these multi-agent systems can lead to highly complex and nondeterministic dynamics, it is not easy to predict whether the impact of the resulting multi-agent behavior will be positive or negative for the agents or the system. Yet, it becomes critical that we understand such dynamics, especially
when considering systems where the resulting dynamics could cause significant societal harm.

In this proposal, I aim to study the impact of the emergent dynamics in realistic decentralized systems where agents have mixed incentives. I will measure the conditions under which the resulting dynamics are beneficial or harmful to either a collection of agents or the entire environment. Unlike previous work, I will run my experiments with a heterogeneous population of agents with varying capabilities, including communication and coalition forming, to more closely resemble real life scenarios.

I will ground my theoretical contributions in the application of automatic energy trading in smart grids, where smart energy agents act autonomously on behalf of their human owners. Due to the general nature of my experiments, the results will easily carry over to many other domains where complex decentralized systems are prevalent, like autonomous vehicle fleets, smart shipping, logistics or water and heat management.
Effective start/end date1/11/2131/10/22


  • multi agent reinforcement learning
  • emergent dynamics
  • partially observable stochastic games with mixed incentives

Flemish discipline codes

  • Adaptive agents and intelligent robotics
  • Machine learning and decision making
  • Neural, evolutionary and fuzzy computation