Electronic games aren’t just fun, they’re an incubator of advanced software technologies as well as a big industry. The world market for game software alone should reach US $100 billion in 2019, according to the gaming industry research group DFC Intelligence. At the annual IEEE Conference on Computational Intelligence and Games (CIG), to be held from 31 August to 2 September in Tainan, Taiwan, presenters will highlight the latest developments in gaming, from edutainment to social dilemma games.
Sponsored by the IEEE Computational Intelligence Society, the conference’s main focus will be on computational intelligence, which emphasizes problem-solving approaches such as neural networks and fuzzy systems that are inspired by the natural world. Many CI algorithms rely on heuristics inspired by biological phenomena, such as evolution by natural selection.
Unbound by the limits of physical game boards, card decks, and the like, electronic games can react and learn the way human players do and can even evolve in response to player input and other game events. The advanced computational intelligence technologies covered at CIG can lead to a wider variety of plots, more challenging simulated opponents for single-player games, better-developed game characters, and games that generate their own levels of difficulty.
“Some game programs can react to the player’s emotions, sensing them via physiological signals such as changes in skin conductance [the skin momentarily becomes a better electrical conductor when external or internal stimuli are physiologically arousing],” points out Georgios N. Yannakakis, an associate editor of IEEE Transactions on Computational Intelligence and AI in Games.
“Computational intelligence can also be used in many other applications, such as robotics and e-learning,” says Chang Shing Lee, general cochair of this year’s conference. But games are particularly useful for the study of CI, providing competitive and dynamic environments that model many real-world problems, he explains. At the same time, CI methods help designers and developers devise new types of computer games.
The conference will also cover traditional artificial intelligence, alone and in combination with CI, as well as a number of other techniques. Among the most important ones for game developers today are procedural content generation (PCG) and Monte Carlo tree search, notes IEEE Member Julian Togelius, who sits on the conference’s advisory board and is an associate professor of computer science and engineering at New York University Polytechnic School of Engineering, in New York City. “PCG is a technique for automatically generating parts of the game. That’s really important because it saves work for programmers, reducing development costs.”
“Monte Carlo tree search is a weird and wonderful but very effective combination of Monte Carlo methods (which use stochastic sampling), with a classic tree-search technique called Minimax that’s used extensively in playing board games such as chess,” he continues. “It’s a real game changer that has revolutionized computer Go and also shown very good results in Ms. Pac-Man, among other games.”
Other topics at the conference include self-evolving algorithms for optimizing multiple factors without unduly degrading any of them, CI in the digital design process, building nonplayer game characters, and theoretical and empirical analysis of CI techniques for game development.
There will also be discussions on strategy games, card and board games, predator/prey evasion games, realistic games for simulation or training, and games involving random factors such as dice rolls, together with design issues specific to game consoles, video games, and mobile devices. One session will even cover scientific research on curling, a sport so dependent on strategy that it has been called “chess on ice.”
IEEE Fellow Xin Yao, president of the IEEE Computational Intelligence Society, will give a keynote address on coevolutionary learning in game playing. According to Togelius, in coevolution, multiple evolving entities put evolutionary pressure on each other, as in the real-world evolutionary race between, say, rabbits and foxes. This technique is widely used in automatic learning of game-playing strategies, evolving artificial neural networks, and automatic programming in software development. Yao will focus on cooperative coevolution, in which different coevolving algorithms cooperate to solve complex problems, and on how strategies learned this way can be generalized for use in new environments.
Competitions will also be held to highlight interesting CI and game research problems. Participants, mostly attendees, will design and code programs that simulate players for Geometry Friends, Starcraft, and other games.
For the game of Go, programs mostly submitted by conference goers will be pitted against expert human players. Computer-versus-human Go contests have a long history because of the game’s complexity.
“Go has a high branching factor, meaning that lots of different moves can be made at any given point, at least an order of magnitude more than in chess,” says Togelius, “Monte Carlo tree search, the method underlying most Go algorithms, can be used for all kinds of games such as arcade games, strategy games, role-playing games, board games, sports simulations, and even nongame applications that deal with planning in stochastic or very complex environments.” Program chair Shi-Jim Yen, a highly rated (sixth-dan) Go player, may participate—but as a Go program developer, not a player.
“Games are everywhere, used for training, education, and even for political argument,” says Togelius. “They’re useful in simulations of military situations, emergency services, ecological outcomes—any kind of system with multiple agents. And they help programmers test algorithms in complex ways that are faster than testing against contrived benchmarks, and more enjoyable, too.”