Enhancing gameplay for non-ideal player groups through al: insights from 'trash pandamonium'

Multiplayer party games are a popular genre designed to foster social interaction between players in a fast-paced environment. However, a significant challenge arises when groups of players do not have the ideal number of participants to fully enjoy the game. The current solution that game developers employ is using AI players to fill the gaps, but these AI players often fall short in providing a satisfying experience. This thesis aims is to investigate the AI player-based approach and determine best practices for designing AI players. The investigation consists of analyzing the design choices made during the development of the Capstone game "Trash Pandamonium." Key findings indicate that through the advancement of AI tools, and using frameworks for AI development, the creation of AI players that can provide a fulfilling gaming experience can be a simple process. Importantly, these AI players do not need overly complex behaviors to create a complete gaming experience when human players are not available. These insights indicate that AI players can effectively complete the game experience for non-ideal group sizes without requiring extensive time to design intricate behaviors, allowing more developers to look at including AI players as an effective option for their own games without having to worry about hiring more staff or devoting tens to hundreds of hours of development time.

ABSTRACT Multiplayer party games are a popular genre designed to foster social interaction between players in a fast-paced environment. However, a significant challenge arises when groups of players do not have the ideal number of participants to fully enjoy the game. The current solution that game developers employ is using AI players to fill the gaps, but these AI players often fall short in providing a satisfying experience. This thesis aims is to investigate the AI player-based approach and determine best practices for designing AI players. The investigation consists of analyzing the design choices made during the development of the Capstone game “Trash Pandamonium.” Key findings indicate that through the advancement of AI tools, and using frameworks for AI development, the creation of AI players that can provide a fulfilling gaming experience can be a simple process. Importantly, these AI players do not need overly complex behaviors to create a complete gaming experience when human players are not available. These insights indicate that AI players can effectively complete the game experience for non-ideal group sizes without requiring extensive time to design intricate behaviors, allowing more developers to look at including AI players as an effective option for their own games without having to worry about hiring more staff or devoting tens to hundreds of hours of development time. ii TABLE OF CONTENTS ABSTRACT ii INTRODUCTION 1 TERMINOLOGY 5 TRASH PANDAMONIUM 10 AI IN TRASH PANDAMONIUM 13 CONCLUSION AND FUTURE WORK 18 REFERENCES 20 iii 1 INTRODUCTION In the ever-shifting market of modern game development, developers must answer numerous questions and make important decisions to create a more enjoyable experience for the player. One of the most central problems game developers face in this regard is how to attract and retain players in their game. Given the saturation of the market with countless titles and the demand for innovative gameplay and mechanics, solving this problem can be quite the challenge (Goh, 2023). When players can pick from hundreds of different games, all within the same genre, how does a developer stand out and convince the consumer to purchase their game? How does the developer keep the player engaged with their game for extended periods of time? In response to these questions, developers are often driven to explore novel avenues, experiment with cutting-edge technologies, and redefine traditional gaming paradigms in their quest to craft experiences that are not just entertaining but truly immersive and unforgettable. One possible answer to these questions is the implementation Artificial Intelligence (AI) to control parts of the game, whether that be mechanics, enemies, or even other players. The history of AI in games begins with some of the first video games ever produced, such as “Pong” (Pong, 1972) and “Pac-Man” (Pac-Man, 1980). Pong had very simple AI where the opponent moved to where it predicted the ball would end up, with a small error added at the end. Pac-man, on the other hand, had different behaviors for each of the ghost, such as chasing Pac-man directly or heading to where Pac-man will be in two steps. Having multiple simple AI behaviors creates more realistic feeling opponents without having to be truly complex. As the technology surrounding video 2 games advanced, so did the complexity of the AI in these games. The game “The Chessmaster 2000” (The Chessmaster 2000, 1986) won its category at the US Open Computer Chess Championship, showcasing the fact that a game could integrate AI capable of beating research-grade computers, while still being a game (Oxner, 1986). When the game “Half-Life” was released by Valve in 1998, it marked a turning point in video game AI. In the game, non-player characters (NPCs) reacted to changes in the environment caused by the player or other NPCs (Half-Life, 1998). This included things such as scientists in the game being afraid of the dark and running to light sources if the player turns off the lights in a room, and NPCs reacting to smells coming from corpses. Though those examples are both of AI for single-player experiences, AI is being leveraged more and more to enhance multiplayer games across many different genres as well. In many Battle Royale games, such as “PlayerUnknown's Battlegrounds” (PUBG, 2017), AI bots are placed into games to make sure that matches are filled without players having to wait a long time to queue into a match. In “Counter Strike 2” (Counter Strike 2, 2023), AI bots will replace players that leave during a game, allowing those still in the game to avoid playing with fewer players on their team. In “Tom Clancy’s The Division” (The Division, 2016) an AI director controls enemy factions that react to player strategies and movements, employing flanking maneuvers and taking cover during firefights. These built-in AI opponents illustrate how AI can directly enhance multiplayer experiences by maintaining game balance and providing dynamic challenges for players. While these implementations of AI are intriguing, this thesis focuses on AI specifically within the domain of multiplayer party games. For this thesis, multiplayer party games will be considered as social games, designed for multiple players, with 3 simple rules and quick gameplay. By incorporating AI into multiplayer party games, developers aim to provide dynamic and adaptable characters for players to engage with, elevating the overall gaming experience. While these characters often take the form of opponents that face off against the human players in the game, the AI can also be used to assist the player in team-based game modes. This versatility in how the AI players can be viewed by the human players is what truly elevates the playing experience for everyone involved (McGee, 2010). Research suggests that incorporating AI companions into multiplayer games can significantly enhance player enjoyment (Merrill, 2022). Enjoyment is an extremely important metric for developers, as players will often stop playing a game if they do not find it enjoyable. Moreover, AI players can effectively mitigate issues stemming from player skill discrepancies by being a middle ground between players of different skill levels. This can help to foster a more balanced and enjoyable gaming environment for everyone playing (Larche, 2020). This goal can be achieved by offering human players options of AI players with different skill levels, placing the choice on the player as to how difficult they want the game to be. As such, the integration of AI into multiplayer party games represents a promising avenue for developers seeking to create immersive and enjoyable gaming experiences that cater to diverse player preferences and skill levels. The integration of AI players into multiplayer party games introduces a new dimension of social interaction and gameplay dynamics as well. By simulating humanlike behaviors, AI players can contribute to more engaging gameplay experiences (Yannakakis, 2017). Additionally, AI-driven characters can adapt their strategies and 4 behaviors based on the player's actions and preferences, creating a sense of dynamic challenge and unpredictability. This adaptability fosters a deeper sense of immersion and investment in the gaming experience, as players engage with AI companions that evolve and grow alongside them (Yannakakis, 2017). Incorporating AI players into multiplayer party games also holds the potential to expand the accessibility and inclusivity of gaming experiences. By offering customizable difficulty levels and adaptive AI behaviors, developers can cater to a broader range of player skill levels and preferences (Drachen, 2009). This inclusivity promotes a more welcoming and accommodating gaming environment, where players of all backgrounds and abilities can participate and enjoy the experience together. This collaborative aspect enhances the social dynamics of multiplayer party games, fostering camaraderie and teamwork among players as they navigate challenges and conquer obstacles together. AI characters are undeniably a valuable asset for game developers attempting to balance a multiplayer game. This thesis aims to show how differing complexities of AI characters in a game can all help achieve a more balanced and enjoyable game for players. This thesis is presented within the context of the development of a game titled "Trash Pandamonium” (TP), a game intended for four human players that implemented AI players to fill in for any missing human players. The rest of this thesis is structured as follows. The Terminology section introduces important terminology used throughout the thesis regarding the implementation of AI players. The Trash Pandamonium section will then provide a basic overview of the game's development. The AI in Trash Pandamonium section describes the specifics of AI player implemetation in Trash Pandamonium. The Conclusion and Main Takeaways 5 section summarizes the main findings of this thesis and thoughts on the future of AI in multiplayer games. TERMINOLOGY Game development is a multifaceted field that requires an understanding of various technical, artistic, and theoretical concepts. Key among these are game design principles, which encompass the rules and mechanics that define player interaction and the overall gameplay experience (Adams, 2014). Knowledge of programming languages and software development practices is also crucial, as they form the backbone of game creation, enabling developers to bring their designs to life. An understanding of user experience and user interface design ensures that games are not only enjoyable but also accessible and intuitive for players (D.P., 2017). Finally, project management skills are essential for coordinating the efforts of diverse teams and ensuring that projects are completed on time and within budget. These topics collectively contribute to a comprehensive understanding of game development, forming the foundation upon which this thesis is built. Given the breadth of the field, a few key terms for this thesis are defined in this section. Unreal Engine 5 (UE5) is a game engine developed by Epic Games. Game engines like UE5 are almost essential in large-team modern game development, as they provide comprehensive tools for designing, rendering, and simulating interactive worlds (Epic Games Unreal, 2024). This streamlines the development process and enables developers to focus on creativity rather than technical constraints. The AI capabilities in UE5 allow AI characters to make contextually appropriate decisions, which results in enhanced realism and interactivity in games (Csepregi, 2023). The integration of these AI 6 tools ensures sophisticated and believable gameplay experiences, which are crucial for immersive digital environments. In UE5, different maps and layouts are handled as separate levels, a feature that allows developers to manage various parts of a game world more efficiently. Levels in UE5 can range from small, contained environments to expansive open worlds, each serving a unique purpose within the game. This modular approach facilitates better organization and enables teams to work on different sections of a game simultaneously without causing conflicts. It also allows for more efficient loading and streaming of game content, as only the necessary levels are loaded into memory at any given time, improving performance and reducing load times (Epic Games Level Streaming, 2024). Figure 1. An example snippet of a blueprint from Trash Pandamonium 7 Unreal Engine 5 offers two main methods for game programming: traditional C++ coding and the custom Blueprints system. C++ offers a powerful and robust system, granting granular control over most parts of the engine for maximum performance. However, it is also quite complex even for experienced programmers and is not very approachable for artists and other non-programming developers. Addressing this accessibility issue is where the Blueprints system excels (Epic Games Blueprints, 2024). The Blueprints system utilizes a visual scripting language, in which lines of code are replaced with nodes and connections between nodes (see Figure 1). This visual approach makes Blueprints much easier to work with for developers who have little to no experience with programming, allowing artists and designers to work within the engine itself to implement game content and features without requiring the intervention of a dedicated programmer. Figure 2. A Simple Decision Tree showcasing the sequence, selector, and task nodes. 8 The main way that any form of AI is handled in UE5 is with decision trees and the Blackboard system built into the engine. Decision trees provide a hierarchical structure for AI decision-making, enabling developers to define behaviors based on various conditions or inputs (see Figure 2). They work by evaluating conditions at each node to determine the appropriate branch to follow, ultimately guiding AI behavior through a series of logical choices (Champandard, 2019). Blackboard serves as a shared memory space where AI agents can store and access relevant information during runtime, facilitating contextually aware behaviors. When combined with decision trees, these two systems allow for complex AI systems within a game using simplistic logic when compared to the Neural Network Engine that Unreal Engine offers. The next step in improving the performance of AI players is using the Environmental Query System (EQS). The EQS allows AI to make informed decisions based on various environmental factors, enhancing its ability to interact with the game world in a more realistic and nuanced manner (Epic Games Environment, 2023). The EQS operates by allowing developers to set up queries within a decision tree that can include a range of environmental checks, such as finding the current closest player. By processing these queries in real-time, AI characters can react to changes in the environment more fluidly, improving their overall performance (Epic Games Environment, 2023). Incorporating EQS into AI development helps bridge the gap between static, pre-scripted behavior and dynamic, responsive gameplay. It enhances the realism of AI interactions, making the AI feel more responsive to player actions. In UE5, controllers are fundamental classes that manage character actions within the game (Epic Games Controllers, 2024). The player controller serves as an 9 intermediary between the player’s input device (such as a game controller or keyboard) and the character they are controlling. It processes the inputs from the player and translates them into movement, attacks, and interactions with objects. The AI controller is quite similar, though it does not handle inputs from any physical controllers. Instead, it takes inputs from the decision tree and translates it into movements and interactions with other players (Epic Games AI, 2024). This enables AI players to navigate the world and play as though they were human players without the need for a physical controller. The fact that both classes are children of the overarching controller class also means that they are interchangeable, allowing code for handling human players to also handle AI players without extensive refactoring. With this solid groundwork in place, the following section will detail the specific development process of Trash Pandamonium. We will look at how the use of the Blueprints scripting language allowed for greater iteration during the development process, enabling rapid prototyping and collaboration between programmers and nonprogrammers within the game engine. We will also explore how decision trees and Blackboards were used in creating intelligent and context-aware AI players for Trash Pandamonium. By examining how these systems were leveraged to shape Trash Pandamonium, it will become apparent how these tools were integral in developing a captivating and enjoyable experience for players. 10 TRASH PANDAMONIUM Figure 3. The logo for Trash Pandamonium, drawn by William Freeman. As the brainchild of game director William Freeman, Trash Pandamonium (see Figure 3 for game logo) draws from two main sources for inspiration: Kirby Air Ride (Kirby Air Ride, 2003) and the Smash Run minigame from Super Smash Bros. for Nintendo 3DS (Super Smash Bros. for Nintendo 3DS: Smash Run, 2014). In Trash Pandamonium, up to four players navigate a 3D level (see Figure 4), competing to collect trash and earn powerups while fighting each other, before facing off in a final showdown. This fast-paced action mirrors Smash Run’s focus on collecting items to increase your stats (Super Smash Bros. for Nintendo 3DS: Smash Run, 2014) and Kirby Air Ride’s frantic scramble to collect powerups and fight the other racers (Kirby Air Ride, 2003). 11 Figure 4. A screenshot from the City level in Trash Pandamonium. In Trash Pandamonium, four players battle as raccoons trying to collect the most trash at the end of the main round. Just like collecting stat items in Smash Run, the collected trash boosts each player’s stats1 a certain amount, meaning that the player who has collected the most trash will have the highest stats at the end of the round. This then leads to the “Final Showdown,” which can take several forms: x King of the Hill – Players must knock each other off of a platform, and the last racoon remaining on the platform wins. x Dodge Bomb – Similar to “King of the Hill,” but players are also given bombs to throw at each other. x Keep the Crown (see Figure 5) – Players must maintain posession of a crown for 30 cumulative seconds to win, but the crown can be stolen by attacking the player in control of the crown. In this context, “stats” refer to changeable player attributes, such as speed, bag size, and attack power. 1 12 x Hot Potato – Players pass around a giant bomb until it explodes and eliminates the player holding it, with the last player standing winning the game. Figure 5. A screenshot showcasing one of the Final Showdown maps for the city map, with bounce pads on each truck. The development of Trash Pandamonium exemplifies the iterative design process that is commonplace in commercial game development. Development occurred during the 2023-2024 academic school year. The initial game concept was originally pitched as a smaller scale prototype, which was meant to be completed in two weeks. This prototype focused on the core mechanics, such as attacking other players, picking up trash, and having the trash modify the player's stats, and it only featured one desert-themed level. While Kirby Air Ride and Smash Run served as starting points for the ideas developed in Trash Pandamonium, new and updated ideas were explored as development continued after the prototype phase. An attack combo system replaced the basic single attacks of the prototype, while VFX were added to the attacks as well, giving the game more visual flare (see Figure 6). New levels were added, increasing the number of levels 13 to three main levels: Mesa (previously the desert level), Beach, and City. Environmental hazards were added to each level, such as fire tornados, moving vehicles, and a rising tide. Advanced movement options were added as well, with bounce pads being placed in every level, and an improved dash mechanic allowed for more advanced movement combos. Figure 6. A player attacking another player during a Final Showdown, showcasing the attack VFX. AI IN TRASH PANDAMONIUM I. IMPLEMENTATION Throughout the development process, a common issue was not being able to test the game with a full set of four players present, clearly highlighting the need to have some form of AI player. Additionally, the implementation of AI players also greatly broadened the game’s accessibility, allowing solo players and groups of two or three players to experience the true frenetic competition of Trash Pandamonium. While the AI players are not able to fully emulate the experience of playing with a co-located human player, the 14 core gameplay loop and gaming experience remain intact and as the developers intended it to be experienced. To implement AI players into Trash Pandamonium within the given time constraint for development, much of the built-in AI functionality in Unreal Engine 5 was used, such as the previously detailed Blackboards and decision trees. However, due to time constraints, it was not possible to implement greater AI functionality through the Environment Query System. The decision to not utilize the EQS helped reduce the complexity of the AI player development, allowing more time for design. In the end, human players are still able to engage with the complete gameplay experience, even without the EQS. One of the main difficulties during the implementation of the AI players was working with the current systems that were created for Trash Pandamonium. Although the AI systems in UE5 are intended to be robust, there are certain tools that need to be used with the Character Blueprint class to allow for the AI controllers to work. As Trash Pandamonium was developed using the Blueprints system rather than C++, there were further restrictions on how the AI systems could be utilized. This issue primarily arose from the methods used to spawn the player in at the beginning of the game, which did not work with the AI controller being used in place of the player controller. The problem was eventually resolved by modifying many different references throughout the game code to use a generalized controller rather than specifically a player or AI controller. Several other small changes had to be made to the game’s code to have AI players spawning in and running correctly. While each fix was simple, identifying every instance of something throughout the code and updating it was very time consuming. 15 The functionality of the AI players was designed using the conceptual design framework introduced by Firas Safadi, et al. in their paper “Artificial Intelligence in Video Games: Towards a Unified Framework” (Safadi, 2015). This framework was used to ignore the specifics of how UE5 handles AI and instead focus on how the AI should work abstractly. Some code for the abstraction of a basic enemy AI is presented in Figure 8. Figure 8. A short code snippet showing an abstract form of an AI player. Looking at Figure 8, it starts with defining an enemy and a trash variable, before entering an endless loop of the core logic. Here, the AI finds the nearest player and nearest piece of trash. This is then used to check the nearest player to see if they are in range and have more than 10 pieces of trash. This check is done using the nearby_enemy function, which returns 1 if the conditions are met. If the conditions are met, the player moves to the nearest player and attacks them. If either condition is not met, then the player moves to the nearest piece of trash and picks it up. One major flaw here is that the AI is unable to deposit trash, as it never explicitly moves to 16 the deposit location. While it is a simple model, it is possible to expand on it to create a significantly more advanced AI player. Figure 9. The final decision tree in the game utilizing three distinct states. By the end of development, the version of the AI player that was implemented in the game, while not as simplistic as the model shown above, was still limited in its capabilities. The AI has three states that it can be in: Scrap Collection, Attack, and Deposit (see Figure 9). The default state, Scrap Collection, only focuses on collecting scrap as this is the focus of the game. Depending on how close the nearest opponent is, the AI can then swap to the Attack state, where it chases down the nearest player and attacks them. The Deposit state is only reached when the AI has more than a certain threshold of scrap. Due to the EQS not being used with this model, the main issue the AI player runs into is that it is unable to navigate the launchpads on the levels. II. MAIN TAKEAWAYS 17 The development of AI players in Trash Pandamonium truly underscored the advanced capabilities of engines such as UE5. With just the integrated tools available in UE5, it is possible to create complex AI behavior. Despite the relatively recent emergence of truly complex AI behavior in games, it is clear that the games industry plans to continue to march full steam into more and more experimental territory. This can be seen with tools such as the EQS, which is still experimental in UE5 at this time, offering even more options for increasing AI complexity. The usefulness of the conceptual design framework introduced by Firas Safadi, et al. in their paper “Artificial Intelligence in Video Games: Towards a Unified Framework” (Safadi, 2015) is also quite apparent. By focusing on a conceptual approach, this framework facilitated a flexible approach to AI design. It provided a structured foundation that abstracted away the complexities of the UE5 AI tools, allowing for the greatest amount of freedom during the design process without having to consider implementation details. The framework also encourages an iterative approach, ensuring that better and better AI players can be created. Artificial intelligence has and will continue to be a major part of the conversation around technological advances, including the modern gaming space. As technology continues to advance, AI in games is poised to play a pivotal role in enhancing player experiences. The integration of sophisticated AI tools in one of the most popular public game engines demonstrates a commitment to pushing the boundaries of what is possible in a video game. These tools not only streamline the development process, but they also empower developers to create more immersive and intelligent worlds without having to become masters of AI. 18 CONCLUSION AND FUTURE WORK AI in video games has been around since some of the very first video games in the 1970’s. With time, the power and capabilities of AI in video games have increased exponentially. What started as a paddle moving up and down a screen in Pong (Pong, 1972) has turned into AI that can dynamically adapt to the playstyle of the player (Echo, 2017) or control entire armies in large scale battles (The Forever Winter, Unreleased). In the case of Trash Pandamonium, while the AI players did not end up being quite that advanced, these technological advancements facilitated the ease of creation and integration into the game. Using Safadi et al.'s conceptual framework also allowed for flexibility with how the AI players were designed, highlighting the importance of taking an iterative approach to game development. The next steps in development are to continue working on improving the ability of the AI players in Trash Pandamonium. Though development of the game has formally concluded, a few members of the team have continued development. The easiest way to see a marked improvement in the AI player ability would be to include the EQS in the decision tree. This would allow for greater control of the AI player, specifically regarding creating more complex decision trees with more branches. The implementation itself would be straightforward, as the EQS is designed specifically to work with the decision trees. Including the queries at branch points will allow for more specific state change requirements, as well as allowing more information about the state of the game around the player to be gathered. With this change, actions such as using the jump pads and effectively playing every final showdown would be possible. 19 Another possible path forward is with the Neural Network Engine (NNE) that was recently introduced in UE5. While this thesis has only focused on algorithmic AI implementations, neural networks and machine learning are a promising area of future research. Using the NNE would allow for perhaps the most complex AI players. However, it would be difficult to do any sort of planning for the abilities of such an AI player given how machine learning works. It would be possible to nudge the AI in a certain direction by using specifics metrics to gauge how well it is performing, but following a specific order of actions would be impossible. This results in perhaps a more lifelike AI player but takes away any control from the developer as to how the AI acts during the game. AI in games will continue to advance, with new and better tools being developed. However, with the conceptual design approach, AI modeled today could be used to power a game 10 years from now. By not getting bogged down in the specific tools available at a certain time, the conceptual approach allows AI designs to remain relevant far beyond a specific implementation in a game. This reusability is great for the video game industry, as it can save hours of work, allowing focus to be shifted to other parts of the game. In the end, having a completed and published game is the most important factor, so having an approach to AI that allows for time saving while still being complex and interesting is essential. 20 REFERENCES Adams, Ernest. Fundamentals of Game Design. New Riders, 2014. Assaf, Muhammad. "From Pong to Narrative: The Evolution of AI in Gaming" ART 108: Introduction to Games Studies (2023). Champandard, Alex J., and Dunstan, Philip. "The behavior tree starter kit." Game AI Pro 360: Guide to Architecture. CRC Press, 2019. 27-46. Counter Strike 2. PC, Valve Corporation, 2023. https://www.counter-strike.net/cs2 Csepregi, Lajos Matyas. The Effect of Context-Aware LLM-Based NPC Conversations on Player Engagement in Role-Playing Video Games, Unpublished Manuscript, 2023, https://www.projekter.aau.dk/projekter/files/536738243/The_Effect_of_Context_aware_ LLM based NPC Dialogues on Player Engagement in Role playing Video Games. pdf. Accessed 10 July 2024. Drachen, Anders, Canossa, Alessandro, and Yannakakis, Georgios N. "Player Modeling Using Self-Organization in Tomb Raider: Underworld." Proceedings of the 2009 IEEE Symposium on Computational Intelligence and Games, 2009, pp. 1-8. ECHO. PC, Martin Emborg, 2017. Epic Games. (2024) AI Controllers. Unreal Engine 5.4 Documentation, https://dev.epicgames.com/documentation/en-us/unreal-engine/ai-controllers-in-unrealengine. Accessed 11 July 2024. 21 Epic Games. (2024) Blueprints Visual Scripting in Unreal Engine | Unreal Engine 5.4 Documentation | Epic Developer Community. Unreal Engine 5.4 Documentation, https://dev.epicgames.com/documentation/en-us/unreal-engine/blueprints-visualscripting-in-unreal-engine. Accessed 25 May 2024. Epic Games. (2024) Controllers. Unreal Engine 5.4 Documentation, https://dev.epicgames.com/documentation/en-us/unreal-engine/controllers-in-unrealengine. Accessed 10 July 2024. Epic Games. (2023) Environment Query System. Unreal Engine 5.3 Documentation, https://docs.unrealengine.com/5.0/en-US/environment-query-system-in-unreal-engine/. Accessed 19 May 2024. Epic Games. (2024) Level Streaming. Unreal Engine 5.4 Documentation, https://dev.epicgames.com/documentation/en-us/unreal-engine/level-streaming-in-unrealengine. Accessed 10 July 2024. Epic Games. (2024) Unreal Engine 5.4 Is Now Available, www.unrealengine.com/enUS/blog/unreal-engine-5-4-is-now-available. Accessed 18 May 2024. Goh, Edward, Al-Tabbaa, Omar, and Khan, Zaheer. ‘Unravelling the Complexity of the Video Game Industry: An Integrative Framework and Future Research Directions’. Telematics and Informatics Reports, vol. 12, 2023, p. 100, https://doi.org/10.1016/j.teler.2023.100100. Half-Life. PC, Valve Corporation, 1998. 22 Kirby Air Ride. GameCube, HAL Laboratory, 2003. https://en.wikipedia.org/wiki/Kirby_Air_Ride. Kristiadi, D.P., Udjaja, Yogi, Supangat, Budiman, Prameswara, Randy Yoga, Warnars, Harco L.H.S., Heryadi, Yaya, and Kusakunniran, Worapan, "The effect of UI, UX and GX on video games," 2017 IEEE International Conference on Cybernetics and Computational Intelligence (CyberneticsCom), 2017, pp. 158-163, https://doi.org/10.1109/CYBERNETICSCOM.2017.8311702. Larche, Chanel J, and Dixon, Mike J. “The relationship between the skill-challenge balance, game expertise, flow and the urge to keep playing complex mobile games.” Journal of behavioral addictions vol. 9(3), pp. 606-616. 6 Oct. 2020, https://doi.org/10.1556/2006.2020.00070. McGee, Kevin, and Abraham, Aswin Thomas. ‘Real-Time Team-Mate AI in Games: A Definition, Survey, & Critique’. Proceedings of the Fifth International Conference on the Foundations of Digital Games, Association for Computing Machinery, 2010, pp. 124– 131, https://doi.org/10.1145/1822348.1822365. Merrill, Kelly, Kim, Jihyun, and Collins, Chad. "AI companions for lonely individuals and the role of social presence." Communication Research Reports 39.2 (2022): 93-103. https://doi.org/10.1080/08824096.2022.2045929. Oxner, Bill. "The Chessmaster 2000”. Computer Gaming World. No. 32, 1986, p. 44. Pac-Man. Arcade cabinet, Namco, 1980. 23 PLAYERUNKNOWN’S BATTLEGROUNDS (PUBG). PC, KRAFTON, 2017. https://pubg.com Pong. Arcade cabinet, Atari, 1972. Safadi, Firas, Fonteneau, Raphael, and Ernst, Damien. ‘Artificial Intelligence in Video Games: Towards a Unified Framework’. International Journal of Computer Games Technology, 2015, p. 1-30, https://doi.org/10.1155/2015/271296. Super Smash Bros. for Nintendo 3DS. Nintendo 3DS, Masahiro Sakurai, 2014. https://www.smashbros.com/wiiu-3ds/us/3ds/. The Chessmaster 2000. PC, The Software Toolworks, 1986. The Forever Winter. PC, Fun Dog Studios. Unreleased. https://www.fundogstudios.com. Tom Clancy’s The Division. PC/PS4/Xbox, Magnus Jansén and Julian Gerighty, 2016. https://www.ubisoft.com/en-us/game/the-division Trash Pandamonium. PC, William Freeman, 2024. https://www.trashpandamonium.com. Yannakakis, Georgios N., and Julian Hallam. "Artificial intelligence and games." IEEE Transactions on Games, vol. 9, no. 4, 2017, pp. 308-311. Name of Candidate: Jonathan Jenkens Date of Submission: August 5, 2024