Design of Generative AI Game NPCs Based on Human-Computer Interaction

Abstract

The generative artificial intelligence (Generative AI) and the large language models (LLMs) are developing at a very fast pace, which has led to the fundamental change in the way of interaction with the non-player characters (NPCs) in the games. The conventional NPCs have no dynamic reaction mechanisms or long-term memory, and therefore they cannot be used to satisfy the requirements of the authenticity and immersion in the modern open-world games as it is not possible to use them effectively. In the past few years, the topic of designing an NPC based on LLMs has been discussed in detail by many researchers within the academic world including topics like personality consistency, dialogue creation, contextual adjustment, etc. Most current works however deal with the technical aspects and do not provide systematic description of how to evaluate and design using human-computer interaction (HCI). The paper presents and applies the generative AI NPC design paradigm that uses the LLMs through the lens of HCI. This can be done through the introduction of such methods as contextual design, low-fidelity. The design-evaluation-redesign process, which is based on the prototyping, think-aloud and heuristic evaluation of the possibilities of AI NPCs in the dynamic dialogue, contextual adaptation and behavior prediction, will be developed as a series of iterative steps. The system has been prototyped with an open-domain dialogue, with a React front-end and an LLM API that supports contextual memory. The mixed-methods study approach to the experimental research includes the quantitative analysis of the immersion of the players and the efficiency of their task performance and qualitative measurement of the user interaction. There are expected results of having a prototype system of an interactive AI NPC, a collection of guidelines to interact with generative AI NPCs, and a set of experimental and assessment toolkits to conduct research at the intersection of HCI-AI. This paper gives theoretical basis and practical directions to the application of the concept of generative AI to game interaction.