Generación procedimental de comportamiento para videojuegos narrativos basada en grandes modelos del lenguaje
Loading...
Official URL
Full text at PDC
Publication date
2024
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
En los últimos años, la evolución de la Inteligencia Artificial Generativa está provocando cambios en una sociedad que cada vez tiene más presente esta tecnología en su día a día. La popularidad de los Grandes Modelos del Lenguaje y de los modelos de generación de imágenes ha despertado un fenómeno, en el campo de la Informática, que parece que no ha hecho más que empezar.
Este trabajo aborda esta potente tecnología desde el punto de vista del desarrollo de videojuegos, explorando la posibilidad de que estos modelos se apliquen en el ámbito de los videojuegos narrativos de manera satisfactoria.
El objetivo del trabajo es poner a prueba las capacidades de los nuevos chatbots y modelos de lenguaje para generar automáticamente comportamientos para un videojuego narrativo, y estudiar los puntos fuertes y débiles de esta tecnología al usarla sobre este juego, a modo de sistema que interpreta la interacción del jugador para tratar de avanzar en la trama. Se busca también calibrar el nivel de autonomía que se le puede dar al modelo para que sea factible implementarlo, pero sin condicionar la variabilidad que puede aportar.
Se ha desarrollado este sistema en Unity que se comunica, mediante la API de ChatGPT, con el modelo GPT-3.5 para la generación de comportamientos a lo largo de una aventura gráfica en la que el jugador debe guiar a una joven protagonista a través de un viaje onírico para superar el dolor de una gran pérdida que ha sufrido.
Como queda demostrado a lo largo de las muchas pruebas realizadas, gracias a las intervenciones de ChatGPT, en la mayoría de los casos cada parte de la historia se resuelve de una manera razonable, teniendo en cuenta las intenciones del jugador y el modelo de juego provisto inicialmente.
In recent years, the evolution of Generative Artificial Intelligence is bringing about changes in a society that is increasingly aware of this technology in its daily life. The popularity of Large Language Models and image generation models has awakened a phenomenon, in the field of Computer Science, that seems to have only just begun. This paper approaches this powerful technology from the point of view of video game development, exploring the possibility that these models can be applied in the field of narrative video games in a satisfactory way. The aim of the work is to test the capabilities of the new chatbots and language models to automatically generate behaviors for a narrative video game, and to study the strengths and weaknesses of this technology when used on this game, as a system that interprets the player’s interaction to try to advance the plot. We also seek to calibrate the level of autonomy that can be given to the model so that it is feasible to implement it, but without conditioning the variability that it can provide. This system has been developed in Unity that communicates, through the ChatGPT API, with the GPT-3.5 model for the generation of behaviors throughout a graphic adventure in which the player must guide a young protagonist through a dreamlike journey to overcome the pain of a great loss she has suffered. As demonstrated throughout the many tests performed, thanks to ChatGPT interventions, in most cases each part of the story is resolved in a reasonable way, taking into account the player’s intentions and the game model initially provided.
In recent years, the evolution of Generative Artificial Intelligence is bringing about changes in a society that is increasingly aware of this technology in its daily life. The popularity of Large Language Models and image generation models has awakened a phenomenon, in the field of Computer Science, that seems to have only just begun. This paper approaches this powerful technology from the point of view of video game development, exploring the possibility that these models can be applied in the field of narrative video games in a satisfactory way. The aim of the work is to test the capabilities of the new chatbots and language models to automatically generate behaviors for a narrative video game, and to study the strengths and weaknesses of this technology when used on this game, as a system that interprets the player’s interaction to try to advance the plot. We also seek to calibrate the level of autonomy that can be given to the model so that it is feasible to implement it, but without conditioning the variability that it can provide. This system has been developed in Unity that communicates, through the ChatGPT API, with the GPT-3.5 model for the generation of behaviors throughout a graphic adventure in which the player must guide a young protagonist through a dreamlike journey to overcome the pain of a great loss she has suffered. As demonstrated throughout the many tests performed, thanks to ChatGPT interventions, in most cases each part of the story is resolved in a reasonable way, taking into account the player’s intentions and the game model initially provided.
Description
Trabajo de Fin de Grado en Ingeniería Informática, Facultad de Informática UCM, Departamento de Ingeniería del Software e Inteligencia Artificial, Curso 2023/2024