AI application in video games and toys are a mutually beneficial combination. On the one hand, AI technology can provide solutions to an increasing demand to add realistic, intelligent behavior to the virtual creatures that populate a game world. On the other hand, as game environments become more complex and realistic, they offer a range of excellent test beds for fundamental AI research.
So, where does artificial intelligence come into play? We argue that both games and AI research can greatly benefit from each other. From a research point of view, video games offer fascinating toy examples that capture the complexity of real-world situations while maintaining the controllability and traceability of computer simulations. As an example, consider the problem of driving a racing car under realistic race conditions. While the full problem is too complex to be tackled right now because it involves problems around limited actuators and noisy sensors in addition to the AI problem, important aspects can be tackled working inside a state-of-the-art racing game simulation. As game designers work hard to create more realistic worlds for their customers, AI researchers can benefit from access to benchmarks that accurately reflect real-life problems. Games exhibit many combinations of features that are important in current AI research. For example, a game environment can be either static or dynamic, there can be either single-agent or two-player or multi-agent problems, transitions can be either deterministic or non-deterministic, and game worlds can be either fully known or partially observable.
From a games perspective, one key problem is the creation of AI driven agents that can interact with the player and be adaptive so as to create a great interactive gaming experience. These agents can take a variety of roles such as player’s opponents, teammates or other non-player characters. Online planning and reinforcement learning have the ability to create adaptive behavior, which might become a key feature in future games. This is useful to respond to changes in the human player strategy, the environment, the current problem instance, etc. Games like Creatures and Black & White have attempted to build entire games around the concept of teaching behavior to adaptive AI agents.
A few concrete examples of AI challenges in games, which we plan to cover in this tutorial, include driving a car in a racing game, path finding on a map, planning the behavior of non-player characters in a role-playing game, resource gathering in a real-time strategy game, and planning the strategy of a combat team in a first-person shooting game. We anticipate that people from the AI community will have a lot to contribute to the field of computer games once the wealth of opportunities in this space has been understood.
However, computer games offer a great variety of other challenges including problems in graphics, sound, networking, player rating and matchmaking, interface design, narrative generation, game world design, scripting etc. All of these areas would benefit from various learning and planning paradigms.