15.12.19 - 20.12.19, Seminar 19511

Artificial and Computational Intelligence in Games: Revolutions in Computational Game AI

The following text appeared on our web pages prior to the seminar, and was included as part of the invitation.

Press Room

Motivation

The 2016 success of Google DeepMind's AlphaGo, which defeated the Go world champion, and its follow-up program AlphaZero, has sparked a renewed interest of the general public in computational game playing. Moreover, game AI researchers build upon these results to construct stronger game AI implementations. While there is high enthusiasm for the rapid advances to the state-of-the-art in game AI, most researchers realize that they do not suffice to solve many of the challenges in game AI which have been recognized for decades. This Dagstuhl Seminar aims at getting a clear view on the unsolved problems in game AI, determining which problems remain outside the reach of the state-of-the-art, and coming up with novel approaches to game AI construction to deal with these unsolved problems.

Among the challenges that the seminar will tackle are the following:


  • Determining the limitations of MCTS and deep learning for computational game playing
  • Defining more appropriate game complexity measures, which accurately reflect the difficulty of creating an AI that achieves superhuman play
  • Learning game playing under adverse conditions, such as imperfect information, continuous action spaces, and deceptive rewards
  • Implementing computational game AI for games with 3 or more players
  • Implementing AI for games with open-ended action spaces, such as interactive fiction and table-top roleplaying games
  • Going beyond MCTS in General game playing
  • General video game playing
  • Computation for human-like play

The seminar will consist partly of discussion groups, and partly of practical work on implementing AI. Apart from a daily round-up meeting, plenary talks will be kept to a minimum. The seminar welcomes both academic researchers and industry practitioners.

License
Creative Commons BY 3.0 Unported license
Jialin Liu, Tom Schaul, Pieter Spronck, and Julian Togelius