- Introduction/background: motivation and overall introduction, including necessary literature review and citations.
• Methods: describe the focus of your research topic and the methods you used to explore it.
• Results and findings: give research results, or explore your findings. Include experimental or theoretical analysis accomplished based on your methods.
• Related work: summarize the related work; compare your methods and results with others’ work. Citations are required.
• Conclusion and future work: conclude your exploration and research, and point out possible future work.
• References: include all cited references.
Your exploration or research should focus on an in-depth literature review about theories, techniques, algorithms, approaches, mechanisms, and implementation of the selected topic. The list of proposed topics is as follows:
• Uncertainty models for knowledge representation
• Reasoning in uncertainty
• Case-based reasoning
• AI planning
• Agent-based systems
• Agent-oriented programming
• Unsupervised learning
• Competitive learning
• Genetic and emergent models of learning
• Natural language processing
• AI for game development
- Introduction/background: motivation and overall introduction, including necessary literature review and citations.
This study aimed at investigating the measures or techniques that can be employed in achieving effective implementation of AI in designing video games. The study accomplished this goal by exploring the body of the existing literature. Searches were conducted on search engines of reputable academic libraries for information technology including IEEE database, ACM database, and CiteSeerx. The outcomes of the study showed that techniques such as employment of a unified conceptual AI framework in addressing issues associated with AI System-Based Video Games, employment of Monte Carlo Search Tree (MCST) algorithm in creating Non-Player Characters (NPCs), employment of High-Quality Computer-Generated Imagery (CGI), and focus on AI system’s Modules and its layered structure can be used to accomplish effective implementation of AI in designing video games. The sudy recommend future studies to focus on mechanisms of enhancing AI to generate a more unique and better user experience.
Implementation of Artificial Intelligence in Game Design
Background of the Study
Currently, the implementation of the incredible complexity associated with Artificial Intelligence (AI) engines is enhancing the effectiveness of applying AI in game design. Whyte et al. (2015) define artificial intelligence as the capability of a computer-controlled or digital computer robot to execute tasks commonly linked to intelligent beings. Different computer software fields including economic planners, market simulators, and logic systems depend heavily on AI elements decision-making, problem-solving, tree searching, and situation calculus. Whereas these fields have robustly established themselves in the application of AI in executing different activities, video gaming is gradually beginning to borrow concepts of AI, with the aim of establishing robust software programs that can give users memorable gaming experiences through artificially intelligent games. According to Granic et al. (2014) and Brown (2014), many video games, whether they are shooting games, strategy games, or racing games possess different elements that are regulated by AI. Examples of such elements are neutral characters and enemy bots. In contemporary computers, the most common types of game AI are those that choose animations for NonPlaying character’s (NPCs) and permit NPCs to maneuver the virtual environment or setting without failure (Whyte et al., 2015). Apart from being just a thirty-minute evasion of reality or a distraction from work, video games are transforming into an artistic expression for developers and programmers and a serious undertaking and hobby for the players. The artistic expression nature of video game has been enhanced by the presence of novel generation software that enables users to customize their specific tasks. Examples of such software systems are general game-playing systems. In relation to this, it can be noted that general game playing is a challenging, as well as an interesting problem for AI, as it involves several fundamental aspects such as decision-making, reasoning, planning, and learning. Brown (2014) general game playing offers a novel anchor AI learning as a unique model for teaching compound AI subjects such as first-order and propositional logic, problem-solving by search, logic planning, and programming. As practiced within industry, Game AI takes into consideration a broader class of data representations, algorithms, workarounds, and hacks employed in conveying this intelligence illusion (Granic et al., 2014; Whyte et al., 2015). This paper explored the application of AI in game design with the aim of establishing methods or techniques that can be embraced to realize an artificially intelligent game.
Researchers have made significant efforts in establishing or identifying various methods or techniques that can be employed in enhancing the effectiveness of applying AI in video games (Sanghvi, 2019; Safadi et al., 2015; Beerthuizen et al., 2017; Rohl, 2017; Hernández-Orallo et al., 2017; Castelvecchi, 2016; Kiliboz and Gudukbay, 2015; Picca et al., 2015). Sanghvi (2019) explored artificial intelligence in the designing of games. The focus of the author was on RTS-Type games, FPS-Types games, and sport games. When it comes to Real-Time Strategy (RTS) type games, the author stresses the possibility of distinguishing several AI system’s modules and its layered structure. An effective system of path-finding is considered among the basic modules. This module is plays a significant role in the detection of collisions and handling of units within the battlefield to avoid each other. In relation to the First-Person Shooter (FPS) type games, the author emphasizes the employment of the inverted kinematic systems in calculating the parameters associated with the animation of arm positioning, with the aim of ensuring that the hand grabs an object situated on a shelf or table. These games enhance the effectiveness of AI application in gaming by implementing the layered structure associated with the AI system. When it comes to sports games, the author emphasizes two significant attributes of AI. The first feature is the ability to assess or evaluate the terrain, with the aim of detecting hindrances lying upon the road. The second attribute is firm cooperation with the module of physics. According to the author, the physics module offers the information about the skidding of the car to the AI system, which in turn reacts appropriately by attempting to return the vehicle under control.
Beerthuizen et al. (2017) focus on the employment of the game theory in finding the best approach for every hand, while encountered with various uncertainties. An example of such concepts is the grand theft auto, which is employed in developing autonomous vehicles or cars by training them to identify stops signs such as those concealed by weather, dirt, or shadows. Even though this software required a slight adjustment to enable a computer to operate it, the virtual world or environment within the game offers the AI algorithm having a semi-realistic app from which the computer is to learn. In a different study, Rohl (2017) established that AI can be implemented effectively in video games using high-quality computer-generated imagery (CGI) throughout the storyline of a game, which in turn grants a game a film form or nature. An example of video games that have been developed using this concept is the Assassin’s Creed, which is categorized as an action-adventure stealth game. This video game is employed in training machine learning algorithms.
As in the case of Beerthuizen et al. (2017) and Rohl (2017), Hernández-Orallo et al. (2017) and Castelvecchi (2016) also emphasize the employment of CGI environments in training AI algorithms, with the aimed of developing robust and effective AI-based video games with memorable user experience. Hernández-Orallo et al. (2017) refer to the Minecraft, which is a sandbox video game that permits users build with various blocks within a 3D procedurally created world. This concept has been employed in Project Mamo in conducting AI experiments that validate significant research into machine education or learning. According to Castelvecchi (2016), the world created within the Minecraft game creates a setting where artificial intelligence can learn from its environs and establish an internal depiction of the space. Besides, the AI can assist researchers to comprehend the process of learning and world’s machine learning perspective.
Kiliboz and Gudukbay (2015) and Picca et al. (2015) support the employment of the Finite State Machine (FSM) algorithm in making the non-player characters (NPCs) appear intelligent. Kiliboz and Gudukbay (2015) assert that the FSM algorithm was developed in 1990s and it based on the generalization of all possible circumstances that can be encountered by an artificial intelligence followed by the programming of a given reaction for every circumstance. The AI engages in a prompt reaction to the action the human player with its pre-programmed mannerism. For instance, in a game of shooting, AT would launch an attack in a situation where the human player appears and then withdraw when its level of health is extremely low (Kiliboz & Gudukbay, 2015). Kiliboz and Gudukbay (2015) assert that turtles found in Super Mario are associated with rudimentary design of FSM. According to Picca et al. (2015), the four primary actions that a human player can execute in games founded on the FSM AI design are attack, wander, evade, and aid. The authors mention example of prominent games that employ the FSM AI design such as Tomb Raider, Battle Field, and Call of Duty. 0743952026
Contrary to Kiliboz and Gudukbay (2015) and Picca et al. (2015), Lorentz (2016), Companez and Aleti (2016), and Hocine et al. (2015) prioritize the Monte Carlo Search Tree (MCST) algorithm over the FSM algorithm. The authors argue that the FSM AI design is associated with the shortcoming of being predictable. All behaviours of NPCs are preprogrammed and this may lead to boredom to a player after playing the game a number of times. Lorentz (2016) considers the MCST algorithm more advanced technique of enhancing the experience of personalized game. The MCST exemplified the approach of employing random trials in solving a problem. This AI approach is employed in Deep Blue game, which happens to be the first computer program that defeated a human chess winner in 1997 (Lorentz, 2016). For every point earned in the game, Deep Blue employs the MCST in first considering all the probable moves it could initiate followed by the consideration of all the probable move a human player can make in response. This step is then followed by consideration of all the Deep Blue’s potential responding moves. As such, all the probable moves can be imagined to expand in the manner that resembles the growth of braches on a stem. This game is also known as the search tree. Multiple repetition of this process, then result in the AI calculating the payback followed by the decision on the appropriate branch to follow. Once the AI takes a real move, it repeats the search tree based on the possible results. Kiliboz and Gudukbay (2016) assert that video games based on the MCST AI design can manage to calculate numerous potential moves and select the ones having the best payback. According to Kiliboz and Gudukbay (2016) and Hocine et al. (2015), many strategy games employ a similar algorithm (MCST). Nonetheless, since the probable moves exceed those involved in chess, all these games cannot be considered. In these video games, the MCST algorithm engages in random selection of the potential moves with which to start. As such, results are more unpredictable human players, which in turn make the games more interesting than those founded on the FSM AI design. For instance, the Civilization game, which involves players competing against the AI to establish a city, every move for the artificial intelligence can be pre-programmed. As opposed to embracing actions based on the present status, which is the case of FSM AI design, the MCST AI focuses on the evaluation of some of the probable next moves such as establishing technology, defending a fortress, and launching an attack on a human player among other moves (Hocine et al., 2015). The AI then proceeds to execute the MCST to calculate or compute the entire payback for every move and selects the most valuable one.
Safadi et al. (2015) investigated artificial intelligence in video games, with the aim of establishing a unified framework. The authors argued that since video game artificial intelligence is always specifically developed for every game, tools associated with AI tools presently focus on enabling developers of video games to efficiently and quickly establish specific AI. However, this approach does not ensure efficient exploitation of numerous resemblances that occur between video games of same and different genres, which in turn result in difficulty when it comes to handling many aspects associated with a complex environment individually for every video game. In relation to this, the authors propose an approach founded on the employment of a unified conceptual model to enable the establishment of conceptual AI that depends on conceptual actions and views in defining the basic, but robust and reasonable behavior. The authors illustrate this approach by focusing on two video games including StaCraft and Raven.
This paper focused on the exploration of effective application of artificial intelligence in designing games or game design. This goal was accomplished by conducting a systematic review of literature focusing on this area of study. Searches were conducted on search engines of reputable academic libraries for information technology including IEEE database, ACM database, and CiteSeerx. The inclusion criteria included studies that focus on the area of study, studies written in English, and studies less than five years since their publication, as well as non-literature review studies. The exclusion criteria were studies not focusing on the topic, studies older than five years since their time of publication, and studies not written in English, as well as studies that are literature review. The initial search yielded 63 articles. However, the 41 articles did not focus precisely on the topic of study, 8 articles were older than five year since their publication, and 2 articles were not written in English. As such, only 12 articles met the criteria for inclusion in the study. The outcomes of the critical analysis of these articles are discussed in the subsequent section.
Results or Findings
The study outcomes revealed that three measures can be employed in enhancing the effectiveness of applying AI in designing video games. These outcomes were classified under themes that were dominant in the body of literature reviewed. These themes were focus on AI system’s modules and its layered structure, employment of CGI environments in training AI algorithms, employment of Monte Carlo Search Tree (MCST) algorithm in creating Non-Player Characters (NPCs), and conceptual framework for addressing problems associated with AI system-based video games. These themes are discussed in the subheadings below:
Focus on AI system’s Modules and its Layered Structure
The study findings revealed that AI system’s modules and its layered structure can be employed in designing effective video games (Sanghvi, 2019). Two types of video games were established to focus on this approach. These games were Real-Time Strategy (RTS) type games, and First-Person Shooter (FPS) type game. Effective development of RTS-Type games calls for the need of differentiating several AI system’s modules along with its layered structure. One approach of accomplishing this goal involves the path-finding technique, which is employed in detecting collisions and handling units within the battlefield to evade each other. The FPS-Type games are established by employing inverted kinematics systems in computing parameters of animation of arm location or positioning, as suggested by Sanghvi (2019). In this manner, the arm can manage to execute tasks such as gabbing objects located on a shelf or table. As such, these findings reveal that the implementation of the AI system’s layered structure can be employed in achieving effective implementation of AI in video game.
Employment of High-Quality Computer-Generated Imagery (CGI)
The findings of the study revealed that high-quality computer-generated imagery (CGI) can be employed in designing video games based on AI systems (Beerthuizen et al., 2017; Rohl, 2017). The employment of this approach leads to the creation of a virtual environment from which the AI can learn what to do. Examples of games employing this concept are the grand theft auto and Assassin’s Creed. In the grand theft auto, autonomous cars are developed and trained to identify stop signals such those obscured by shadows, weather, or dirt, as suggested by Beerthuizen et al. (2017). In Assassin’s Creed, machine learning algorithms are also trained, as argued by Rohl (2017). Besides, the study outcomes show that the concept of employing CGI environments or settings in training AI algorithms can also be employed in executing AI experiments that support significant research into machine education, as in the case of Project Mamo (Hernández-Orallo et al., 2017; Castelvecchi, 2016). For instance, Minecraft is a sandbox video game that enables users to build with different blocks within a 3D procedurally established world. This concept is employed in Project Mamo.
Employment of Monte Carlo Search Tree (MCST) Algorithm in Creating Non-Player Characters (NPCs)
Study findings showed that the use of Monte Carlo Search Tree (MCST) Algorithm in Creating Non-Player Characters (NPCs) in establishing non-player characters contributes to effective application of AI in designing video games with good user experience. In video games, AI plays the role of regulating NPCs. As such, to make NPCs appear intelligent MCST algorithm is considered appropriate over Finite State Machine (FSM) algorithm, as it results in the creation of unpredictable games (Kiliboz and Gudukbay, 2015; Picca et al., 2015).
Conceptual Framework for Addressing Problems Associated with AI System-Based Video Games
The study outcomes show that the employment of a unified conceptual AI framework that relies on conceptual actions and perceptions in defining primary, but reasonable behavior is appropriate for addressing issues associated with video games founded on AI systems (Safadi et al., 2015). Such an undertaking ensures that developers can establish solutions for individual video games amidst similarities among different games.
There exist studies that have attempted to examine the area of research studied in this paper. For instance, Polasek (2014) conducted a literature review of Grand Theft Auto and established that Computer-Generated Imagery (CGI) can be employed in creating virtual settings that train AI on what to do. This outcome is in line with the one of the findings of this study. In a different study, Nebel et al. (2016) performed a literature review on the employment of Minecraft in research and education. The outcomes of the study were in line with the findings of this study that the concept of Minecraft can be employed in conducting AI experiments that support significant study into machine education or learning.
Conclusion and Future Work
This study aimed at investigating the measures or techniques that can be employed in achieving effective implementation of AI in designing video games. The study accomplished this goal by exploring the body of the existing literature. The outcomes of the study showed that techniques such as employment of a unified conceptual AI framework in addressing issues associated with AI System-Based Video Games, employment of Monte Carlo Search Tree (MCST) algorithm in creating Non-Player Characters (NPCs), employment of High-Quality Computer-Generated Imagery (CGI), and focus on AI system’s Modules and its layered structure can be used to accomplish effective implementation of AI in designing video games. Future studies should focus on mechanisms of enhancing AI to generate a more unique and better user experience.
Beerthuizen, M. G., Weijters, G., & van der Laan, A. M. (2017). The release of Grand Theft Auto V and registered juvenile crime in the Netherlands. European journal of criminal
Castelvecchi, D. (2016). Tech giants open virtual worlds to bevy of AI programs. Nature News, 540(7633), 323.
Brown, H. J. (2014). Videogames and education. Routledge.
Companez, N., & Aleti, A. (2016). Can monte-carlo tree search learn to sacrifice?Journal of Heuristics, 22(6), 783-813. doi:http://dx.doi.org/10.1007/s10732-016-9320-y
Granic, I., Lobel, A., & Engels, R. C. (2014). The benefits of playing video games. American psychologist, 69(1), 66.
Hernández-Orallo, J., Baroni, M., Bieger, J., Chmait, N., Dowe, D. L., Hofmann, K., … & Thórisson, K. R. (2017). A new AI evaluation cosmos: Ready to play the game?. AI Magazine, 38(3), 66-69.
Hocine, N., Gouaïch, A., Cerri, S. A., Mottet, D., Froger, J., & Laffont, I. (2015). Adaptation in serious games for upper-limb rehabilitation: An approach to improve training outcomes. User Modeling and User – Adapted Interaction, 25(1), 65-98.