Synthergy
Date of Completion
5-5-2024
Degree Type
Honors Thesis - Campus Access
Discipline
Computer Science (CMSI)
First Advisor
Andrew Forney
Second Advisor
Robert Johnson
Abstract
Synthergy is an artificial intelligence (AI) powered social deduction game. Synthergy consists of an interactive website, game server, database, and AI server. Like other popular games in the same genre such as Werewolf, Town of Salem, and Among Us, Synthergy is played in a sequence of phases. During these phases, players use special actions, converse with each other, and vote to eliminate players they believe belong to the opposing team. Unlike these other games, Synthergy allows players to compete without a full roster of human players by utilizing non-player character (NPC) agents powered by the large language model chat-GPT by OpenAI. These agents are able to mimic all human player actions including writing chat messages, accusing other players, utilizing special actions and abilities, and answering questions. Each of these actions is formatted as a custom prompt, prepended with important game and character information. These prompts are then used to query the Chat-GPT 3.5 application program interface (API). Through the use of the LangChain library, the format of the plain-text responses is made consistent and predictable. In preliminary tests, the AI NPC agents have proven initially promising at mimicking human like play.
Recommended Citation
Moore Odell, Abraham; Forney, Andrew Ph.D.; Srouji, Aidan; Armknecht, Alex; and Ramanathan, Arusha, "Synthergy" (2024). Honors Thesis. 526.
https://digitalcommons.lmu.edu/honors-thesis/526
