Generative AI in Mafia-like Game Simulation

Munyeong Kim,Sungsu Kim

2023-09-21

Abstract:In this research, we explore the efficacy and potential of Generative AI models, specifically focusing on their application in role-playing simulations exemplified through Spyfall, a renowned mafia-style game. By leveraging GPT-4's advanced capabilities, the study aimed to showcase the model's potential in understanding, decision-making, and interaction during game scenarios. Comparative analyses between GPT-4 and its predecessor, GPT-3.5-turbo, demonstrated GPT-4's enhanced adaptability to the game environment, with significant improvements in posing relevant questions and forming human-like responses. However, challenges such as the model;s limitations in bluffing and predicting opponent moves emerged. Reflections on game development, financial constraints, and non-verbal limitations of the study were also discussed. The findings suggest that while GPT-4 exhibits promising advancements over earlier models, there remains potential for further development, especially in instilling more human-like attributes in AI.

Artificial Intelligence,Human-Computer Interaction

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to explore the efficacy and potential of generative AI models (especially GPT - 4) in role - playing simulations, specifically through a gang - style game named Spyfall. The main objectives are as follows:

1. **Evaluate the performance of generative AI in complex social - reasoning tasks**:
- Research the ability of generative AI models (such as GPT - 4) to understand and participate in social - reasoning games (such as Spyfall). These games require players to reason, make decisions, and interact based on natural - language communication.

2. **Compare different versions of generative AI models**:
- Compare GPT - 4 with its predecessor GPT - 3.5 - turbo in terms of adaptability in the game context, relevance of questions, and ability to form human - like responses. Research shows that GPT - 4 has significant improvements in understanding the game environment, asking relevant questions, and generating human - like responses.

3. **Explore the limitations of generative AI in games**:
- Some challenges were found in the research, such as the limitations of GPT - 4 in bluffing and predicting opponents' actions. In addition, the financial constraints in game development and the impact of the lack of non - verbal cues on the research were also discussed.

4. **Look forward to the future development of generative AI**:
- Explore how to further develop generative AI to make it have more "human - like" attributes, so that it can be more flexible and widely applied. The research also emphasizes the importance of interdisciplinary cooperation to promote the development and wider application of generative AI.

Through these objectives, the paper hopes to reveal the potential of generative AI in simulating human behavior and complex decision - making tasks and provide directions for future research and development.

### Key Formulas

This article does not involve specific mathematical, physical, or chemical formulas, but it does involve some technical indicators and performance comparisons, such as:

- **Correct - format - response rate**: Used to measure the proportion of questions and answers generated by the model that meet the requirements.

\[
\text{Correct - format - response rate}=\frac{\text{Number of correct - format responses}}{\text{Total number of responses}}
\]

- **Human - like - response quality**: Used to evaluate whether the responses generated by the model have human - like characteristics.

\[
\text{Human - like - response quality}=\frac{\text{Number of responses considered human - like}}{\text{Total number of responses}}
\]

These indicators help researchers quantify and compare the performance of different generative AI models.