User:User319697/sandbox

Views

Many experts complain that the "AI" in the term "game AI" overstates its worth, as game AI is not about intelligence, and shares few of the objectives of the academic field of AI. Whereas "real AI" addresses fields of machine learning, decision making based on arbitrary data input, and even the ultimate goal of strong AI that can reason, "game AI" often consists of a half-dozen rules of thumb, or heuristics, that are just enough to give a good gameplay experience. Historically, academic game-AI projects have been relatively separate from commercial products because the academic approaches tended to be simple and non-scalable. Commercial game AI has developed its own set of tools, which have been sufficient to give good performance in many cases.

Game developers' increasing awareness of academic AI and a growing interest in computer games by the academic community is causing the definition of what counts as AI in a game to become less idiosyncratic. Nevertheless, significant differences between different application domains of AI mean that game AI can still be viewed as a distinct subfield of AI. In particular, the ability to legitimately solve some AI problems in games by cheating creates an important distinction. For example, inferring the position of an unseen object from past observations can be a difficult problem when AI is applied to robotics, but in a computer game a NPC can simply look up the position in the game's scene graph. Such cheating can lead to unrealistic behavior and so is not always desirable. But its possibility serves to distinguish game AI and leads to new problems to solve, such as when and how to use cheating.

The major limitation to strong AI is the inherent depth of thinking and the extreme complexity of the decision making process. This means that although it would be then theoretically possible to make "smart" AI the problem would take considerable processing power.

In the past, due to the uncertainty of AI algorithms, game developers were not willing to apply it to the design of NPC and other game features. With development of AI, algorithms have become more reliable in past few years. Those algorithms are mainly focus on strategic games. Ai algorithms are providing a better solution to the design of game features.

Examples

• Creatures (1996)

Creatures is an artificial life program where the user "hatches" small furry animals and teaches them how to behave. These "Norns" can talk, feed themselves, and protect themselves against vicious creatures. It was the first popular application of machine learning in an interactive simulation. Neural networks are used by the creatures to learn what to do. The game is regarded as a breakthrough in artificial life research, which aims to model the behavior of creatures interacting with their environment.

• Sid Meier's Alpha Centauri (1999)
• Halo: Combat Evolved (2001)

A first-person shooter where the player assumes the role of the Master Chief, battling various aliens on foot or in vehicles. Enemies use cover very wisely, and employ suppressing fire and grenades. The squad situation affects the individuals, so certain enemies flee when their leader dies. A lot of attention is paid to the little details, with enemies notably throwing back grenades or team-members responding to you bothering them. The underlying "behavior tree" technology has become very popular in the games industry (especially since Halo 2).

• F.E.A.R. (2005)

A psychological horror first-person shooter with a gloomy and fascinating storyline. There the player fighting against battalion of clone super-soldiers, robots and paranormal creatures. The AI uses a planner to generate context-sensitive behaviors, the first time in a mainstream game. This technology is still used as a reference for many studios. The enemies are capable of using the environment very cleverly, finding cover behind tables, tipping bookshelves, opening doors, crashing through windows, and so on. Squad tactics are used to great effect. The enemies perform flanking maneuvers, use suppressing fire, etc. As mentioned earlier, what sets it apart is the opponent AI which is extremely good and acts on even minute details. For example, in the game if a player takes cover behind a barricade, the enemy AI's quickly processes and throws a grenade to eliminate the player. Another aspect is the game AI's communication in the game. The enemies talk and detail their movements according to that of the player. In the game has achieved what AI could not achieved in the earlier years, namely, to give a person a similar experience. The remarkable enemy AI's still impresses to this day. The game is one of the greatest shooters with game AI's. That's not to mention the weapons physics and movement animations in this game, where they were groundbreaking for their time. This game won GameSpot's "2005 Best AI Award".

• S.T.A.L.K.E.R. series (2007-)

A first-person shooter survival horror game where the player must face man-made experiments, military soldiers, and mercenaries known as Stalkers. The various encountered enemies (if the difficulty level is set to its highest) use combat tactics and behaviors such as healing wounded allies, giving orders, out-flanking the player or using weapons with pinpoint accuracy.

• StarCraft II (2010)

A real-time strategy game where a player takes control of one of three factions in a 1v1, 2v2, or 3v3 battle arena. The player must defeat their opponents by destroying all their units and bases. This is accomplished by creating units that are effective at countering your opponents' units. Players can play against multiple different levels of AI difficulty ranging from very easy to Cheater 3 (insane). The AI is able to cheat at the difficulty Cheater 1 (vision), where it can see units and bases when a player in the same situation could not. Cheater 2 gives the AI extra resources, while Cheater 3 gives an extensive advantage over its opponent.

• Hearthstone (2014)

An online digital collectible card game where 2 players against each other using a deck of 30 cards. The player is winning by defeating the hero of the opponent. Every card has different damage, blood and effect, which makes the AI control difficult since the consequences of a single card can be hard to measure. In recent years, many AI developers have been building algorithms to teach computers to play Hearthstone. Besides being a test bed for the developers, the players can play against the computer in the game "Adventure" mode. The difficulty can be adjusted by selecting "normal" and "difficult" modes.

One of the AI algorithms implemented is machine learning method. Researchers are trying to predict the predict the opponent's deck using machine learning supervised and unsupervised learning method. Knowing what cards are in the opponent's deck boost the winning chance significantly. Even though the possible card combination is very large, the players tend to use certain combination instead of all of them. This makes it possible for supervised learning algorithms to predict the composition of deck. In addition, examining the relationship between the deck composition and the winning chance will also give players advices on building their own card deck. The machine learning algorithm can compare the different cards between those in the winning decks and those in the losing decks. The comparing result will indicate which specific cards can boost the winning chance and which cannot. More importantly, the AI algorithm can be used to assist the player to choose the best move along a game. By generating the state space and the winning probability, the player can select the best move from the action space. This network training process can be done with the reinforcement learning method.

• Honor of Kings (2015)

Honor of Kings is a is a multiplayer online battle arena game on iOS and Android mobile platform. Every player will control a hero to fight in a battle arena. The traditional mode is the 5v5 battle arena but there are also 1v1 and 3v3 modes. The team is winning by defeat the base of the opponent. The AI control of the real-time strategy game is more complicated compared to the traditional board games. To win the game, the AI agent not only need to plan, attack and defend but also need to perform detailed operation like making combos and deceiving opponents. According to the Tencent researcher, there are totally 10^600 possible states and 10^18000 possible actions.

Tencent's AI architecture consists of four modules: Reinforcement Learning (RL) Learner, Artificial Intelligence (AI) Server, Dispatch Module, and Memory Pool. It uses special techniques to reduce the state space and the action space. In 2018, Honor of King AI "Wukong AI" defeat one of the best professional team of human players.

Challenges

Game industry is one of the fields where Artificial Intelligence develops the fastest. The wide variety of problems it offers and the possibilities it contains make video games a perfect test bed for AI. However, there are still many obstacles researchers need to overcome in the future.

• Emotion design

Unlike the traditional AI agents do, in video games, AI algorithms are not trying to design the most powerful characters. The design of games must satisfies the players and build emotional connection with players. There are some good examples of non-playable characters (NPCs) that are remembered by the players because their rich feelings aroused the same feeling of players. These NPCs are programmed to simulate human feelings by control their moves, voices, expressions, etc. However, Artificial Intelligence is not good at dealing with human emotion yet. The action space is also too large to be fully predicted by AI. Therefore, emotion design of NPCs are a big challenge for AI gaming.

• General game playing (GGP)

General game playing is the design of AI agents to play more than one game without special training. It requires AI programs to mimic human in learning rules and making decisions. GGP is not only a challenge in the game industry but the society as well.