SwRI IEEE AESS Speaker Event Talk “Mitigating Unintended Consequences in Multi-Agent Reinforcement Learning”

Mastering effective collaboration within a team poses a formidable challenge for both human and artificial intelligence entities alike. Striking a harmonious balance between individual objectives and collective goals becomes a pivotal factor in determining the success of all stakeholders involved. This presentation delves into the realm of collaborative teamwork through the lens of multiagent reinforcement learning within the intricate framework of the StarCraft II environment. The presentation will encompass strategies for mitigating unintended consequences, commonly referred to as specification gaming, and elucidate how teams of AI agents can leverage insights from other artificial intelligence entities to enhance their overall performance. The presentation will further provide an in-depth exploration of the QMIX algorithm, offering a comprehensive understanding of its application in the context of the StarCraft II environment. Co-sponsored by: WIE Speaker(s): Garrett, Bldg: 51, Archives, Southwest Research Institute, 229 Avenue D, San Antonio, Texas, United States, 78238