Learning from multiagent emergent behaviors in a simulated environment

Traditionally, determining the most efficient designs and practices—whether for determining how store merchandise should be arranged or where people and machines should be laid out in a factory floor—has required vast amounts of data and human assessment. These efficient designs can be the difference between a thriving company and a struggling one. Recent advancements in multiagent reinforcement learning within virtual environments, such as DeepMind’s Capture the Flag or Open AI’s Learning to Compete and Cooperate, have led to a novel approach for tackling efficient design and practices.

Danny Lange (Unity Technologies) explains how observing emergent behaviors of multiple AI agents in a simulated virtual environment can lead to the most optimal designs and real-world practices, all without introducing human bias or the need for vast amounts of data.

This session was recorded at the 2019 O'Reilly Artificial Intelligence Conference in New York.