This paper presents a new approach for leveraging ChatGPT in fighting game commentary generation task. Commentary generation often relies on deep learning techniques, which typically demand extensive data to achieve effectiveness. Large language models (LLMs) have become essential due to their remarkable ability to process data efficiently, thanks to their extensive training on vast datasets. Our proposed approach integrates the use of LLMs, specifically the GPT-3.5 model, for generating commentaries through the utilization of various prompts with data from the open-source fighting game, DareFightingICE. Four prompt variants are employed to assess the effectiveness of each prompt components. Objective evaluation using natural language metrics reveals that different prompt components significantly affect the generated commentaries. Additionally, subjective evaluation through a questionnaire reveals that prompts without parameter definitions received the highest preference from human evaluators. These results suggest that LLMs exhibit versatility in generating fighting game commentaries and hold promise for broader applications.
This paper presents Enhanced Sonic Doom, a substantial extension of prior conference study, which delivers a comprehensive accessibility framework for first-person shooter (FPS) games. The novel contributions of this research are fourfold: (1) a large-scale human validation with 56 sighted participants (47 male, 8 female, 1 non-binary, aged 18-44), substantially expanding upon initial findings; (2) a dedicated investigation demonstrating the generalization of our auditory design principles to spatial navigation; (3) the introduction of an AI-driven ablation framework to deconstruct the impact of isolated sound cues; and (4) the observation of performance trade-offs suggestive of cognitive load thresholds for effective sonification, where simpler auditory schemes consistently outperformed complex ones. Results demonstrate that our enhanced sound design significantly improves navigation and combat performance. Furthermore, while an automated aim-assist achieved superior objective performance, users strongly preferred a less intrusive, audio-based system, underscoring the critical balance between assistance and user autonomy. This work provides empirically validated design principles advocating strategic minimalism in accessible sound design and introduces a scalable, fine-grained methodology for its evaluation.
This paper presents our work to enhance the background music (BGM) in DareFightingICE by adding an adaptive BGM. The adaptive BGM consists of five different instruments playing a classical music piece called "Air on G-String." The BGM adapts by changing the volume of the instruments. Each instrument is connected to a different element of the game. We then run experiments to evaluate the adaptive BGM by using a deep reinforcement learning AI that only uses audio as input (Blind DL AI). The results show that the performance of the Blind DL AI improves while playing with the adaptive BGM as compared to playing without the adaptive BGM.
This study provides a symmetrical, cross-genre comparison of human expertise transfer and “blind” artificial intelligence (AI) performance in audio-only gaming environments. Although previous research has focused on human performance in audio games and the feasibility of blind agents trained on auditory inputs separately, a direct comparison of these two forms of expertise is missing. We fill this gap with a robust experimental design, involving 37 human players (aged 18–44), grouped by gaming experience and specialized blind AI agents. We measured key performance variables, including win ratios, health differences, and task completion times across two genres: a fighting game (DareFightingICE) and a first-person shooter (SonicDoom). Our findings show a complex, task-dependent relationship. In DareFightingICE, expert humans (73.0% win ratio) significantly outperformed the AI (54.0% win ratio), demonstrating effective cognitive transfer. Meanwhile, the AI’s performance matched the overall human average (54.0% vs. 53.0%). Conversely, in SonicDoom, AI achieved superhuman speed in simple tasks (1.55 s vs. 5.35 s) but underperformed compared to expert humans in complex scenarios, highlighting that the AI’s proficiency is specialized but fragile, whereas human expertise is more robust and adaptable. The results provide practical insights for audio-rich game design and highlight the crucial need for AI models beyond reactive policies.
This study provides a symmetrical, cross-genre comparison of human expertise transfer and "blind" artificial intelligence performance in audio-only gaming environments. Although previous research has focused on human performance in audio games and the feasibility of reinforcement learning agents trained on auditory inputs separately, a direct comparison of these two forms of expertise is missing from the literature. We fill this gap with a robust experimental design involving 37 human players (5 females, 31 males, 1 non-binary individual, aged 18-44), grouped by gaming experience, and specialized blind Artificial Intelligence (AI) agents across two different genres: a fighting game (DareFightingICE) and a first-person shooter (SonicDoom). Our findings show a complex, task-dependent relationship between performance. In DareFightingICE, expert humans significantly outperformed the AI, demonstrating effective cognitive transfer. Meanwhile, AI’s performance matched the overall human average, establishing a key benchmark. Conversely, in SonicDoom, AI achieved superhuman speed in simple tasks but underperformed compared to expert humans in complex scenarios. This difference highlights that the AI's proficiency is specialized but fragile, whereas human expertise is more robust and adaptable. The results provide practical insights for designing audio-rich games and highlight the crucial need for AI models that extend beyond reactive policies.