Fostering motivation through AI techniques in educational serious games

Abstract

The term serious game refers to games that have a bigger impact than entertainment and have been growing over the last years. Serious games usually are developed for a target audience and a target teaching goal. This results in a high cost development for a small audience. One of the most relevant problems in serious games is the need to adapt and balance the game, this balance is restricted to one person and it can't be extrapolated to a bigger audience, turning the small audience to just a few peoples. In order to have an efficient control of learning, it is necessary to understand the latent relation between the cognitive capacity, motivation and performance of each person. The more personalized the material given to each student, the better their learning will be, because the balance is tailored to your needs. Machine learning, especially reinforcement learning (RL) can be used for automated NPC behaviour generation and it can be applied to an agent that controls the difficulty of a game in an unknown, unsupervised environment. For that reason, the application of algorithms like Q-learning may help on the creation of personalized learning curves in a game. The broad objective of this thesis is exploring how Artificial Intelligence can help monitor and adapt a game in real time to a player's needs and profile. Specifically, we want to study the state-of-the-art in the context of serious games as well of adaptative games. We aim to create a game with real time adaptation in the area of Mathematiques, that can create a reliable profile of any player inside our target audience and adapt to the needs of that profile. Ideally, this will be the next step on e-learning and serious games development as we can expand our audience to a bigger number with the same resources. There is a broad range of work in serious games and how they are the answer to motivate the students, however it is need to keep a flow state in the student for better results. Some papers also explore how AI can help with these questions but there is no concrete answer to this need and no definite study of how to do it. Nonetheless, the research work help us define some parameters need for the success of this dissertation work. This work starts by analysing the state of art in education and games, as well AI adapted to these games. After there will be the layout and plan of the game design that was made with the help of an expert. After that the exploration of the integration of Q-learning algorithms within the game we provide some alterations to the normal Q-learning. Finally there is the analyse of the data collected from the target audience and conclusions.