Games are an excellent test-bed for AI research and also a great application domain. Games provide an ideal way to study most aspects of AI, posing complex and challenging decision-making problems in naturally engaging ways. Game AI often focuses on making AI agents which are human competitive, but there are many other aspects of interest, such as using AI to generate game content and also design new games.
Explaining the science
Recent progress in game AI has demonstrated that given enough data from human gameplay, or experience gained via simulations, machines can rival or surpass even the most skilled human players in some of the most complex and tightly contested games. However, each domain is often rather restricted, results may require massive compute power, show a lacking robustness to change, and the AI often provides rather opaque decision making. The question arises as to how we make the AI more general, more rapidly adaptive, more robust and more explainable.
This interest group will provide a forum for discussing interesting challenges, developments and applications of game AI. In addition to making AI even smarter at playing games, how can we use AI to make better games, including automated design of rules and content, and automated play-testing and analysis?
We aim to provide a forum for discussing applications of game AI and for bringing together researchers and users from academia, industry and government.
We plan to hold seminars, workshops, study groups and also run (and sometimes compete in) game AI competitions. Likely topics include deep reinforcement learning, procedural content general, automated game and mechanism design, learning forward models, statistical forward planning (MCTS, rolling horizon evolution, etc.), hierarchical planning and learning, sample-efficient learning, dealing with multiple objectives.
What are the main challenges for wider adoption of game AI in real-world decision making?
How can we use games and game AI for the good of society and to better understand human behaviour?
Can we use Game AI to better understand social networks? Game frameworks and benchmarks (range of games, generalisation, playing standard, learning efficiency).
Combinatorial action spaces, hierarchical planning, sample efficient learning, handling multiple objectives, unknown opponent objectives, multi-player games (n > 2), explainable game AI, learning forward models, procedural content generation, automatic game generation, automated play-testing, builder games (where the challenge for the AI is to build stuff – with emergent properties – think Minecraft).