Artificial Intelligence in Gaming: Creating a Living World and its NPCs

EasyChair Preprint 5812

Abstract

The paper looks at the basic history of AI in gaming and the purpose of establishing true AI for video games. The shortfalls of current AI technology and the end goal of AI are analysed to determine the ideal requirements for true gaming AI. An alternative to the Turing Test, based on the Marcus Test was proposed to test how well an AI meets these requirements. The paper then reveals the findings from its own experimentation into two AI approaches, bottom-up, and top-down. From this experimentation, it was concluded that bottom-up AI is the most effective means of simulating natural behaviour in NPCs. However, players also preferred interacting with the top-down NPC despite generally finding it to be less realistic. It is likely that the simple style of the sprites used had an affect on what behaviour players perceived as realistic. The paper titles this effect the "uncanny divergence." Future papers should look further into the uncanny divergence and a combined top-down/bottom-up approach.

Keyphrases: Artificial, Artificial Intelligence, Computational Intelligence, Living world, NPCs, Turing Test, gaming, intelligence, machine learning, top-down, video game

@booklet{EasyChair:5812,
  author    = {Laura Savaglia},
  title     = {Artificial Intelligence in Gaming: Creating a Living World and its NPCs},
  howpublished = {EasyChair Preprint 5812},
  year      = {EasyChair, 2021}}