Aprendizado ativo em raciocínio baseado em casos para o emprego do engano em jogos de cartas

Abstract

Deception is omnipresent in everyday social interactions. Deception and automatic deception detection and exploration are research subjects in many different fields, such as cyber security, computer games, military operations and fake news. Despite these works, the implementation of intelligent agents capable of deceiving in computer systems remains a challenging task. In particular, it is not trivial to capture and label the intention of a human strategist when making a certain deceptive decision, especially if we consider passive learning techniques in Artificial Intelligence. In this sense, this work proposes a new approach that combines active learning and case-based reasoning - CBR, in which an agent when faced with situations that require deceptive decision-making asks a human expert to make a review of the solution suggested by the CBR algorithms. In this active learning process, if necessary, the expert presents a more appropriate solution to the current problem. Thereby, this work shows how to systematically capture experiences of problem solving that involve deception and later use the acquired knowledge in order to make better decisions when faced with opportune situations for the use of deception. Experimental results in the domain of a card game called Truco has demonstrated that the use of active training techniques, compared to imitation learning techniques, enables a Truco player agent, even using case bases with a reduced number of cases, to play at higher levels than agents who use much larger case bases