- Title: Towards Adaptive Game-playing Agents
- Speaker: Niels Justesen (modl.ai)
- Time and date: 1pm to 2pm, September 23rd, 2020 (Wednesday)
- Room: Virtual (Zoom)
The Game AI Research Group is glad to announce a (virtual) talk by Niels Jusesen on Wednesday Sept 23 at 13:00.
All welcome (especially students), no pre-booking required
Abstract
In the last few years, deep reinforcement learning has enabled computers to master Go, Atari, and Dota. Additionally, AlphaStar was able to win against a professional player in StarCraft, a game that has been considered the grand challenge in game AI for long. It may thus seem that AI has achieved all the long-standing goals that were set forth by the research community. This presentation takes a critical stand on this matter. First, a framework is presented for evaluating the fairness of human vs. AI competitions which highlights several shortcomings of AlphaStar (and game AI in general) that are concerned with adaptation. Then, a new AI challenge is presented that is based on the board game Blood Bowl that requires adaptation, tactical positioning, and long-term strategic planning. Finally, promising research directions are presented that could result in more adaptive AI agents.
Bio
Niels Justesen is an AI Researcher at modl.ai in Copenhagen where he works on game-playing bots and automated testing. He has a Ph.D. from the IT University of Copenhagen where he did research on adaptive game-playing bots with a focus on deep reinforcement learning and evolutionary algorithms. He is organizing the Bot Bowl competition where bots compete in Blood Bowl which is one of the hardest board games ever made. He has visited and worked with the NYU Game Innovation Lab and Jean-Baptiste Mouret at Inria in Nancy, France.