Don’t Be Afraid – an AI Controlled Affective Horror Game

For my bachelor thesis, my group researched whether the use of an affective Game AI would increase the horror experiences in video games.
The project was completed in the summer of 2016, and was called ‘Do not Be Afraid – an AI Controlled Affective Horror Game’ and it was implemented as a videogame in the game engine Unity3D. In the project group, I was responsible for the implementation of an AI that measured biofeedback (perspiration and heart rate) from the users, thereby actively altering the game.
The project research and evaluation gave me deep insight in user engagement in Human-computer Interaction, as well as the theoretical and practical framework of using bio-feedback in videogames.

Abstract:

In traditional horror games, players are often limited to predetermined choices. The experiences are very much alike, no matter who plays the game or wwhen they play it. However, there are several instruments to measure an user’s emotional state(biofeedback), which could send data to the game, and thereby changing the game through e.g. an artificial Intelligence agent. Games, that take advantage of this, are called “Affective Games” as they change and respond to the players’ emotional state. To create a horror game with this type of AI, would individualise the experience for every player. Through research in AI and horror media, a problem statement was formed:

How can we measure the effect of a game AI in an affective horror game, which utilize players physiological signals as input for alterations within the game world?”

Through a total of three iterations, including design, implementation and evaluation, an AI controlled horror game was created. The game took the users’ Galvanic Skin Response, Heart Beats per minute and Heart-rate Variability as input to determine which one of three horror theme packages each particpant found most frigtening.
The third iteration was tested on 37 participants at Aalborg Univeristy Copenhagen. The data proved, that there was no significant difference between the affective game and the non-affective version of the game in terms of how scared the participant felt. AFter discussing and analysing numerous bias factors, an assumption was made that the procedure of testing was appropriate. Given more time and a larger sample size for smaller effect size, the project could have provided different results.

Link to report