Highlights in Science, Engineering and Technology CMLAI 2023

Volume 39 (2023)

AI Application in Computer Games

Qianjun Geng
Wuxi Taihu University, China
Abstract. It is well-established that artificial intelligence has already been used in nearly all kinds of
computer games. The purpose of most games is to entertain (though there exists some educational
games which aim to teach), and AI takes the role to improve the player’s experience. First, current
use of artificial intelligence in computer games is introduced, then, some AI applications which aim
to promote both entertainment and interactive environment is mentioned. Then, the future of artificial
intelligence to find the possibility to create a kind of AI which can not only interact with players but
also improve as number and experience of players grow. Lastly, the paper will analyse the future
use of AI in different fields of computer games including, but not quite, the feedback of game bugs
and solutions of players’ problems.
Keywords: Artificial Intelligence; Computer Games.

1. Introduction
Nowadays, artificial intelligence has been applied in all kinds of fields in people’s lives, especially
in computer games. Computer games has become a way of entertainment for no matter young
teenagers or adults, even some old people are sometimes log in to play games. As an important role
in all kinds of games, Artificial Intelligence (AI) has great value of research and discussion. In current
games, techniques of AI such as finite state machines, scripting, agents and flocking are widely used
[1]. All these techniques can satisfy game designers’ requirements of saving time and resource.
However, predictable characters and tasks make players feel boring, so it is necessary to find other
techniques to improve the experience of players.
In this essay, we will describe the current use of AI in games and discuss how can AI be improved
through current techniques to provide a more exciting and challenging game world.
This study will list some techniques currently used in games to create artificial intelligence. It will
also point out some problems exist in games and explore the possibility to solve these problems.

2. Discussion
For players, the behavior of AI may directly influence their experiences, if AI, or no-player
character in another word, gives them too much trouble, they may soon give up the game easily.
Oppositely, when AI make many stupid actions in defence with players and let them reach the goal
without any difficulty, players may also get bored with it. So, it is necessary to find a way to balance
the difficultly of games, or make AI to satisfy players’ requirements. Current types of games that AI
applied in includes FPS (First person shooter) games, RTS(Real- Time Strategy) games, RPG( Role
Playing Games), MOBA( Multiplayer Online Battle Arena), MMOS( Massively Multiplayer Online
Role-Playing Games ), Sports Games and so on.[3]. Next, we will use some of them as an example
to discuss.
2.1 FPS Games
FPS Games are one of the most famous games in the game market all around the world. They also
define concepts like HP, helmet and armour, power-ups medicine and weapons. AI in this kind of
games is designed to be enemies who against players and training their skills to reach higher levels.
2.1.1. Techniques of AI in FPS
AI in FPS Games always takes the responsibility to enhance the tension of the game, so that players
can always immerse themselves in the battle scene because of the existence of enemies or teammates

668
Highlights in Science, Engineering and Technology CMLAI 2023
Volume 39 (2023)

which controlled by AI. In an ideal FPS game, there are always a database of movements exists and
rules for AI to learn and work under the guide of all those data. This makes AI seems more similar as
a real player. The animation system can make actions of AI move at chosen speed and let them finish
complex actions like shooting while run or looking for shelters and reload weapons.
2.1.2. Shortcomings of Current AI
AI system in this kind of game consists of different levels of module. Once the system decided
which behavior is the most suitable one for the given situation, a lower-level of module must choose
the best method to finish the decision. For computer still cannot analyse situations roundly as human’s
brain does, the result of AI system may sometimes seems unrecognized. This problem was been
discussed as “Target Selection for AI Companions” [4]. In order to solve it, our target is to find the
possibility to make AI can analysis the situation as correct as possible and give the right choice. To
reach the target, we think the database need to take in more real choice made by real players when
different situations enable to let AI get to learn.
2.1.3. Future Possibility
In the future, we need to improve the performance of AI in both single player FPS Games and
Online ones. For single player experiences, developers need to create a better AI to satisfy the
requirement of players to own a teammate or truly feedback similar to real wars. For online fps games,
AI characters needs to have the ability to follow player and ready to record backup for the next days’
players [2]. For database, AI needs to collect more behaviors from players and try to create a system
of behaviors of AI to give the right reactions according to players’ choice and the changeable gaming
situation.
2.2 RTS Games
The real-time strategy roots in the 1980s [3], players of this game need to take into different kinds
of wars. Unlike FPS Games we talked before, players are commanders of the war, they have their
armies and the division of arms are decided by them. Each step of their decision may influence the
final result of the war happen in the sandbox. The Blizzard game, StarCraft, is one of the most famous
RTS games in the history.
2.2.1. Techniques of AI in RTS Games
In RTS Games, AI replace the role of online players to work as commanders of the enemy when
players have the idea to play themselves without other online players take part in. After the whole
game, AI may give the feedback in the form of an analysis table to let players know how well they
done.
2.2.2. Shortcomings of Current AI
Lack of real-time planning. In each game, AI nearly do the same things at the same time. As
Figure1 shows, in the game Warcraft-2, wood and gold are the resource and everyone born with their
own gold ore and forest. For online player A, he may realize that he can go through his own area to
take up and exploit the ore 3 as soon as the ore 1. Later, he can use his gold to train soldiers and attack
AI player B to despoil the ore 2. However, AI player B only have the idea to run out his own resource
first and then went to the middle area to take up the ore 3. As a result, it can only wait for player A to
attack and lose the game.
Opponent modeling and learning. For AI, one of the biggest shortcomings is that it cannot learn
quickly. For human players, a map can be master after several games, but AI find it difficult to do
that [5].
Cooperation. For human players, teamwork can be easily carried out through communication and
experience. Players catch the shortages of the enemy and break their defence. When a game mixed
with human and AI, cooperation is not important either between AI and AI or AI and human. AI can
only follow the program to finish the given behavior.

669
Highlights in Science, Engineering and Technology CMLAI 2023
Volume 39 (2023)

Figure 1. Simple map of RTS Game. The start locations of player and AI are represented in A and
B. Green lines represent forest and triangles represent gold ores.
2.2.3. Future Possibility
As the plan [6] said, by the end of 2015, twenty percent of the USA army will transform into robot
army. RTS Games can work as analog system under the background of given situations. Real soldiers
can also be recorded and added into the gaming system, the data index of models can follow their real
data accurately. In order to realize, an ORTS system needs to be built [5]. The system should base on
a programming environment for conducting real-time AI experiments. Anyone can have access to use
the ORTS, players can use the scripted system. They can also run their RTS designs with the server
provide by the system.
2.3 RPG Game
Role Playing Games aim to create an open world for players to explore. In this kind of games, AI
has made a great contribution. There are always be a monster, witches, dragons and other kinds of
otherworldly creature, all of them are controlled by AI in current games, which we called NPC (non-
player-character).
2.3.1. Techniques of AI in RPG Games
Behavior tree [7]. An embedded system of AI in games. When AI have no idea what to do next,
it will ask the “behavior tree” for help and put an uncertain task in its plan. Later, AI may think twice
to find out whether the task still can be done after changeable situation. So, we can think of AI as
having its own schedule, and the system “behavior tree” is only one step to fill plans into the schedule
but not ensure to do the plan next.
Group AI System. Different from the “behavior tree”, Group System gives AI the opportunity to
get information from other AI or other systems. We can describe it as a blackboard, AI system can
record data on it, so do other systems. With the huge ‘blackboard’, AI can learn pluralistic knowledge
and give reaction accordingly. For example, some creatures have two sides: docile and fiery. They
never attack players actively when players are not hostile. So-called “hostility” may give by players’
operating system to make AI realize that they need to fight back.
2.3.2. Shortcomings of Current AI
Long-time. For AI’s ability to learn can be influenced by different kinds of factors like character
information of identity, number of samples, site factors, it may cost up to three or four months for an
AI to reach the average level. If we want to apply it in team work, it may cost even more time.
Repetition. The technique called “static decision tree” dictated AI’s dialog. Every time players
start a dialog with them, they will give the same answers and let players choose from several fixed
options. For example, we define an NPC who owns a shop called Catherine, players can buy meat
from her and the price is 100 gold per unit. Players can choose their answer to the welcome of
Catherine from “I don’t actually need some meat”, “ I truly need some, but I only have 80 gold”, or
“ Nice price, I will buy some”. Even Catherine may give some feedback, but when players leave and
come back, the dialog appear again.

670
Highlights in Science, Engineering and Technology CMLAI 2023
Volume 39 (2023)

2.3.3. Future Possibility ‘

To shorten the time to train an AI, we just shorten the time for AI to calculate before applying in
games. Several NPCs can share a neural network to own a similar behavior. In a group, we can add
code to make inactive AI to follow surroundings to get the behaviors, in another word, train a ‘teacher’
to a group and let all other AI learn from the ‘teacher’. To the problem of repeating, a far more
believable, but much more expensive (in terms of modeling, computation and storage space) solution
is to give Catherine an initial disposition and memory. Thus, even if the player leaves, she will
“remember” previous conversations and experiences with the player and react accordingly.

3. Requirements for the AI in Computer Games in the Future

Intellectuality. In the future, AI can replace human’s work in some jobs. They can learn to solve
players’ problems according to previous similar record. It is also possible to make AI imitate customer
service staff to calm the player down.
Adaptation. AI needs to have the ability to adapt the emergency made by players suddenly. For
example, a player gets the task to break into a cave and defeat the boss. In a general way, he may
keep fighting from the entrance of the cave and defeat the boss finally, but the player finds a direct
way (maybe because of some hidden bug) into the cave suddenly and find the boss without fighting
with any other creatures, the condition will not be fired and the boss cannot realize what to do next.
The game could crash, more seriously, the illusion of game could disappear.
Controllability. The fundamental objective of AI is to increase the entertainment of games but
not make more trouble to players. The training of AI cannot out of control, such as “Doomsday
Computer” of league of legends or Open AI of Dota2. Exploring the potential of deep technology is
truly necessary and interesting for programmers, but we still must consider of players’ experiences.
Authenticity. Today, even though AI has made the player's experience in the game world almost
realistic, there are still some loopholes in the game that allow players to easily distinguish between
reality and the game. Designers need to give the game more real-world conditions, even under an
elevated worldview. For example, in a future-set RPG fight, the player can use the mechanical arm
to block the opponent's attack, but after a few times the player's left or right arm will be damaged,
then the attack relying on the arm will not be used, and even the designer can make the arm directly
scrapped in this case, the player needs to leave the battle or solve the battle and then go to a specific
location for repair.

4. Conclusion
In this paper, we discussed AI in all kinds of computer games in terms of current techniques,
shortcomings that still exist and the future possibility. We also do some guess about the future of AI
and show the basic requirement for AI in computer games. Although there are many challenges along
the way of AI applied in computer games, we still believe the future of AI-used computer games is
bright and desirable.

Acknowledgments
We would like to acknowledge Professor Jamie Vicary for inspiring my interest in the
development of artificial intelligence. At the same time, thanks for Ekrem Ekici for his guidance
through each stage of the essay.

671
Highlights in Science, Engineering and Technology CMLAI 2023
Volume 39 (2023)

References
[1] Sweetser, Penelope, and Janet Wiles. "Current AI in games: A review." Australian Journal of Intelligent
Information Processing Systems 8.1 (2002): 24-42.
[2] Jeremy A. Glasser and Leen-Kiat Soh. "AI in Computer Games: From the Player’s Goal to AI’s Role."
Computer Science and Engineering University of Nebraska-Lincoln Lincoln, NE 66588-0115 {jglasser,
lksoh}@cse.unl.edu.
[3] "The Complete Guide to Video Game Genres: From Scrollers, Shooters, to Sports." https://www.
gamedesigning. org/gaming/video-game-genres/#First-Person-Shooter.
[4] Tremblay, Jonathan, Christopher Dragert, and Clark Verbrugge. "Target selection for AI companions in
FPS games." FDG. 2014.
[5] Buro, Michael, and Timothy Furtak. "RTS games as test-bed for real-time AI research." Proceedings of
the 7th Joint Conference on Information Science (JCIS 2003). Vol. 2003. 2003.
[6] S. von der Lippe, R.W. Franceschini, and M. Kalphat. A robotic army: The future is CGF. In Proceedings
of the 8th Conference on Computer Generated Forces and Behavioral Representation, Florida, USA, 1999.
[7] "How does NPC AI work? From program to planning on game AI" https://www.gameres.com.

672