Artificial Intelligence In Games

Artificial Intelligence in Computer Games

Submitted By Kiran Subramaniam S III Year BE/CSE AC07UCS033

Artificial Intelligence in Computer Games

Introduction What is game AI? In computer games, Artificial Intelligence generally means creating computer players that can think rationally and also can act humanly. First problems of game AI were solved by making challenging computer players that play the best move. But as the games involved more imagination, new problems emerged such as designing humanly behaving and responding characters. Games are traditionally played by a group of players. Few examples are chess, hideand-seek, football. In contrast, many computer games are single-player. So, there is a problem of interactivity in computer games. If the player perceives the game to be a deterministic machine, giving predictable outcomes, it probably will no longer feel like a game. To solve this problem, AI programmers create rational agents in the game to give the illusion of human players. If the player is faced by the challenge to win against intelligent rational opponents, the game feels more like a game. First computer games were digitized versions of regular board games in which intelligent opponents were adequate for the simulation of the real game. But the development of computer graphics, sounds and input technologies opened the way to new possibilities. Games began to have stories, plots, scenes and characters. Unlike board games, which were relatively simple abstractions, computer games began to create a constantly complexifying realm of fantasy worlds. But of course the deepness of these worlds not only depended on the imaginations of game and art designers, but also the artificial hearts and minds of the characters in the game. Since then, game AI has been required to model more and more intelligent and realistic behaviour.

2

Artificial Intelligence in Computer Games

Applications of AI in games There are many different applications of AI in games. Many game genres, each having different set of AI problems, create a wide spectrum of game AI. Despite its many application possibilities, AI algorithms and technologies advance very slowly, compared to the technologies in other aspects of games, such as graphics or physics. Application of game AI involves giving life to artificially intelligent agents in games. This task is easier to directly immitate human behaviour, because the agents are only expected to be intelligent in the context of the game. There are some algorithms that proved to be useful and are widely used in certain applications of game AI. These algorithms solved some of the AI problems in games so well that some of these problems are not considered in the AI area anymore. Problems like collision detection, pathfinding, visibility detection in fact do require artificially intelligent algorithms, but they are generally given different names such as physics or tree searching. AI is mostly related with higher order functions of the human brain, although it contains every level of human intelligence from perception and reasoning to deciding and behaving. Many game genres exist, and there are several game roles that require AI techniques. Enemies, allies of the player, other characters or units, any rational agent in the game uses artificial intelligence.

Game roles The game, with its set of rules and limited world, is like a playground for testing different kinds of artificial intelligence. These different AI problems are practically presented as game roles. These game roles define certain areas of problems in game AI. Well-known AI roles are tactical enemies, partners, support characters, strategic opponents and units (1). Tactical enemy Tactical enemies are most frequently used in first-person (FPS) or third-person (TPS) shooter games. The first well-known FPS is Wolfenstein, and the most influential one is Doom, both by id software. In first FPS games, tactical enemies simply waited standing until they see the player. When they did, they started firing their guns at the player. They

3

Artificial Intelligence in Computer Games

sometimes stopped firing and walked around. Generally called “monsters”, these creatures neither had a memory, nor the motive to flee to survive. They did not have an internal representation of the map, they simply belonged to the room they were in. Maybe their mind can be compared to that of a mosquito, having a gun instead of a procosbis, continuously coming and disturbing the player. As years passed, FPS games with better graphics were developed, but the “monster AI” seemed the same. But there was another effort to simulate computer players, in other words “bots”, in multiplayer FPS games. As the CPU power increased, and the players started giving more importance to intelligent enemies, these bots' intelligence advanced further. They became real tactical enemies, hitting, running away, hiding and sniping. This is a result of new algoritms applied. New bots have an internal map of the area. They behave according to their evaluations of certain points in the map in terms of visibility, safety and other abstract criteria (2). Partner Modeling partner characters require coordination with the player. Generally, partner characters are simple. They only follow the player as he goes, and shoot nearby enemies. This implementation only requires intelligence for pathfinding and detecting enemies. A more complex modeling can be allowing player to issue orders to his partners. Partnership can also be applied to enemies. To generalize, it is squad / team behaviour. Team behaviour is implemented in two ways. In the first solution, each character behaves independently, but they also look around and if they see a teammate, they behave accordingly. Team behaviour is a result of interactions between the individual members (3). In the second implementation, AI thinks not like a member, but the leader. Knowledge of members are centralized and a global planning is made, which allows more organized tactics (4). Support character Support characters in a game are similar to the minor roles in a movie. In the adventure and role-playing games, the player generally interacts with the support characters to follow the storyline of the game flow. These characters have preprogrammed responses to the questions player is allowed to ask. They do not use very sophisticated AI, but better support characters can be modelled with more advanced techniques. For example, non-

4

Artificial Intelligence in Computer Games

player characters (NPC) in a game may know about some events and tell them to each other, they may have some attitutes towards the player or other characters in the game (5). Strategic opponent Strategic opponents are the most obvious and first use of game AI. The player is challenged by an opponent, which is in fact an algorithm. In board games like chess, planning and predicting can easily be used. In probabilistic games and games that the player has more freedom, it becomes very difficult to look ahead and act accordingly. The game being real-time, puts a time constraint and makes the task even more difficult. Strategic decisions are deeper than most other types of decisions. In games with perfect information like chess, direct reasoning is possible. But uncertainty and lack of information makes it harder to implement an intelligently deciding agent. In many strategy games, the programmers pick the easy way and "cheat", meaning that game AI is given more information than what is allowed in the game. This really makes more challenging AI but it can also annoy the player. In a better game, the player must be able to "deceive" his opponent. This is only possible if reasoning under uncertainty is implemented. Computer player may "believe" in some game information, but not know for sure. Bayesian Networks (6) or Dempster-Schafer Theory (7) can be used for this kind of reasoning. Unit Units in first strategy games did not require AI. They just did what you told them to do. But as strategy games began to involve more complex issues such as ambushing or resource gathering, units became simple finite state machines. Also the real-time animations made it possible to create more lively characters. In some "god" games, units are not player's units at all, they are instead independent agents. For example, in Dungeon Keeper, you did not say your creatures what to do. They just hung around, attacked if they saw an enemy. You could change the placement of a creature, or you could get into one and control him seeing through his eyes. In Black & White, the player even cannot control the people, he can only change placement of a person or give a profession to a person. Other than that, the computer people decide what to do. There is also the player's creature, which has more freedom, except that the player may control it using a leash. The Sims is another example where artificial people have "free will" in a limited scale.

5

Artificial Intelligence in Computer Games

Conclusion Computer games and AI research are like two sides of one medal. Fruits of AI research gives life to computer games, and computer games provide new challenges for future AI researches. But of course there is much different aspects between these two. For example, games are industry-oriented whereas AI research is not. AI research mainly focuses on the optimum and best solutions, whereas games require more practical solutions. In the future, as game characters get more and more intelligent, games will get more interesting and more realistic. New AI tecniques also may create new and more "intelligent" genres of games. Everyday human relations and emotions may be used more in gameplay. Games may be able to learn from the player, and behave accordingly. Games may enlarge themselves by automatically generating maps, monsters and other game elements. Game AI is like a young tree, with many fresh branches open to development. Leaves of the tree are wide open to the light of AI research.

6