The Perfect AI
You round the corner, you had just managed to break the guards neck before he shouted. The noise as the body hit the floor may have alerted the guards downstairs. You turn the corner, you look at the stairs leading into the well-lit corridor. You step gingerly out on to the landing, and make your way down the steps. You get about 5 steps from the bottom and suddenly your point of view goes haywire. Then, a shot is heard. Game Over. In the "Death Replay", you saw the thud had indeed alerted the guards downstairs. One hid beneath the stairs, and tripped you up on the way down, the other was just around the corner. He nailed you in the back of the head with the .44 Magnum. You select "Replay Mission". "Gotta kill that guard quieter this time!" you curse to yourself.

Current AI in Gaming
Currently, gaming AI is slowly moving in the right direction. Previously, only a tiny percentage of the game cycle was dedicated to AI, now it is becoming a major part of the gaming cycle. Current AI is showing definite improvements in all genres of gaming, but it will be sometime before our silicon opponents will be able to put up a fight.

1st Person
Games like Quake2 and Unreal are showing some interesting advances in AI. Quake2 enemies were a world beyond those in Quake1 (not exactly a major achievement, but an improvement nevertheless), with ducking routines and better chasing, search, firing algorithms. Quake2 uses a state-based AI, enemies have various states that will change according to what happens around them. Here is a standard state chart:

Improvements we would like to see in first person shooter AI includes better cooperative tactics by the enemies. For example, they will attack in groups, retreat to get reinforcements, etc. Ambush tactics, dodging, jumping, following, strafing and generally more human like behaviour. Quake2 'bots' are a definite step in the right direction. I made a few changes to the Quake2 AI myself, check out the AdvAI mod here.

Flight Simulators
Flight simulation is also taking a step in the direction. Wingmen are coming more and more intelligent and cooperative, allowing the user to issue group tactics to take on incoming enemies. When engaging in combat, the wingmen actively fight aggressively or defensively, using the strengths of their aircraft and the weaknesses of their opponents to their advantages. Probably the best example of such AI is in the US Navy Fighters/ATF series! The game was released a long time ago, nevertheless I've never better AI in a game! The aircraft use group tactics, attack in groups, lead you one, bug you, and take advantage of their aircraft. Most simulators (both flight simulators like USNF, and other games like Mechwarrior) uses goal-based AI techniques. Like entities in the game world are told to protect the homebase, and do what is necessary to complete that goal. This adds a lot more complexity to the game AI system, due to the fact that goals then require sub-goals (attack intruder, return to base). These sub-goals all require programming. Also, such programs often require cooperative programming -- allowing entities to help each other, or act together to achieve a goal. Additionally, games such as flight simulators require AI for planes, boats, vehicles and troops. Either very flexible, or a massive amount of AI coding is needed to support this level of AI for this number and variety of entities in the game world.

Board Games
When it comes to most board games, there's only one word needed to understand the AI - iterations! Most board game AI simply uses the computer power to its advantage. PenteAI uses massive loops that check all the places on the board, and assigns values to each of the spaces according to how good that place is for the certain loop (there is blocking loops, defensive loops, winning loops etc). 3DTTT also uses similar programming techniques to achieve its AI. Some board games would require a little more thinking - the most obvious one being chess. Other games include most of the East Asian board games, like Go.

Conclusion.
AI in gaming is dramatically improving from the "attack-until-dead" or "fixed-flying-pattern" AI of a few years back. The AI still has a way to go, as developers are still focussing more on graphical appeal than AI. They will learn soon that the consumer is just as interested in having decent opponents, as they are in a beautiful game world to blow them up in!