With these high expectations, there are two game AI challenges new to this next generation. The first is for the subtle visible behavior of agents to keep pace with the incredibly detailed, high-polygon models. The second is to create agents which provide more interesting and novel gameplay experiences for the player.

The first challenge is relatively straightforward to understand, but difficult to overcome in practice. Increasingly realistic agent models must be complemented with equally adept and detailed behavior. It is critical for agent behavior consisting of navigation, movements, gestures, blinking, gaze, mannerisms, dialogue, and facial expressions to match the visual quality of the agent. Carefully crafted intelligence is required to direct animation, attention, and intention in a seamless and convincing manner.

As we venture towards extremely realistic looking characters, we perhaps run the risk of falling into the Uncanny Valley. This concept was proposed by Masahiro Mori in 1970 to explain an uneasiness which humans feel towards robots as they approach humanness. Mori claimed as robots start to resemble humans, we feel more empathetic toward them. However, as they approach humanness, the little differences which aren't quite right become extremely disturbing and unnerving-making robots seem more like undead zombies than real people. For example, film critic Roger Ebert proposed the movie Final Fantasy: The Spirits Within might have been rejected by audiences because its slightly imperfect computer animated humans fell into the Uncanny Valley. While the existence of the Uncanny Valley is debatable and hasn't been proven, it is nevertheless prudent for game developers to be aware of the challenge. If you are attempting realism, you must nail it, lest it become distracting to the overall experience.

Unfortunately, matching the visual quality of agents with competent and realistic behavior is unlikely to result in games which are substantially more fun to play. While we must maintain the steady climb in realism, it won't result in making games noticeably more enjoyable-which is what players demand with this next generation of consoles.

The second challenge is where I throw down the gauntlet and challenge game AI to save the day. If increasing realism doesn't give next-gen games the requisite new feel, then something else must help achieve it. One answer is for game design and AI to work hand-in-hand toward creating completely new gameplay experiences. This is a huge challenge because it requires the game designer to understand what is possible with AI and to closely work with the AI programmer. Because this type of relationship is rare in game development, it's an area which has huge potential for many game genres.

If high-end graphics pull the player in visually and accurate physics make the player feel like the world is real, then AI has the power to engage the player mentally. We don't want the AI to necessarily outwit the player (which is relatively easy), rather we want the player to rationalize and internalize the intelligence of the AI and reason about how to overcome it. It's not a question of how to beat the player but a question of how we can design a game in which intelligent agents can be creatively manipulated and exploited by a crafty player.

Given these two challenges for next-gen games, there is much work to do in the future. We need to become better at simulating realistic human behavior and we need to creatively use AI to bring new experiences to players. Meeting both of these game AI challenges will help differentiate this new generation of games from the last.