Thesis

Creativity is known as an ability that can be developed and improved. Since creative abilities are desired in most modern societies, it becomes important to develop activities that stimulate creativity at a very young age. It seems, however, there is a lack of tools to support creative activities for children. We present Cubus, a tool that uses autonomous synthetic characters to stimulate idea generation in groups of children during a storytelling activity. With Cubus, children can invent a story and use the stop-motion technique to create a movie depicting it. This work yielded a useful methodology that we consider can aid the design of tools which assist users in their task. This methodology consists in an iterative development where several user studies are carried out to inform and validate design choices during a tool's different development stages. Additionally, a methodology to evaluate the different aspects of creativity is also presented and implemented during our creativity evaluation with Cubus. To evaluate how Cubus supports creativity, we investigated the number of ideas generated by groups of children during their creative process of creating and recording a story and the creativity of the product this process originated, a stop-motion movie. Results showed that the embodied synthetic characters with autonomous behavior of Cubus contributed to the generation of more ideas in children, a key aspect of creativity. Regarding the creative product, results suggest that Cubus agents' autonomous behaviors were unable to influence children's creative products, the stop-motion movies.

Natural language processing is a complex area of computer science. During recent years natural language started to have more interference in the Games area and has been established in conjunction with speech recognition. This Master Thesis project covers the theory behind an artificial intelligent agent capable of playing the Werewolf or Mafia Game, with a more natural interaction towards human users, using natural language processing techniques and probability theory. We present our solution to create such an agent, where we take advantage of Bayesian networks, variable elimination algorithm, an embodied conversational agent and similarity measures.

Personality has been a key feature in the creation of companion characters for digital games. These characters cooperate with the player to overcome obstacles and progress through the game. In this work, we present a generic framework and a personality model inspired by Cloninger's psychobiological model of Temperament and Character to convey a companion character’s personality in the context of the cooperation between the companion character and the player. Our framework is comprised of the character's personality, a decision system based on the character's personality, and a tag system to keep track of the character's experience, knowledge, objectives, etc. We conducted a two-stage experiment to better understand (1) if the character's graphical design has an influence on personality reporting and (2) if the companion's personality conveyed by our model is adequately perceived by the player through interaction. Our results suggest that (1) in-game behaviour is more important than first impressions induced by the character's design and that (2) two of our traits (Harm Avoidance and Cooperativeness) were easily understood by the participants.

In this work, we introduce a Monte-Carlo tree search (MCTS) approach for the game “Hearthstone: Heroes of Warcraft”, the most popular online Collectible Card Game, with 50 million players as of April 2016. In Hearthstone, players must deal with hidden information regarding the cards of the opponent, chance, and a complex game-play, which often requires sophisticated strategy. We argue that, in light of the challenges posed by the game (such as uncertainty and hidden information), Monte Carlo tree search offers an appealing alternative to existing AI players. Additionally, by enriching Monte Carlo tree search with a properly constructed heuristic, it is possible to introduce significant gains in performance. We illustrate through extensive validation the superior performance of our approach against ”vanilla” Monte Carlo tree search and the current state-of-the art AI for Hearthstone.