Intention recognition, commitment and their roles in the evolution of cooperation

Dissertação para obtenção do Grau de Doutor em Informática

The goal of this thesis is twofold. First, intention recognition is studied from an Arti cial Intelligence (AI) modeling perspective. We present a novel and e cient intention recognition method that possesses several important features: (i) The method is context-dependent and incremental, enabled by incrementally constructing a three-layer Bayesian network model as more actions are observed, and in a context-dependent manner, relying on a logic programming knowledge base concerning the context; (ii) The Bayesian network is composed from a knowledge base of readily speci ed and readily maintained Bayesian network fragments with simple structures, enabling an e cient acquisition of the corresponding knowledge base (either from domain experts or else automatically from a plan corpus); and, (iii) The method addresses the issue of intention change and abandonment, and can appropriately resolve the issue of multiple intentions recognition. Several aspects of the method are evaluated experimentally, achieving some de nite success. Furthermore, on top of the intention recognition method, a novel framework for intention-based decision making is provided, illustrating several ways in which an ability to recognize intentions of others can enhance a decision making process. A second subgoal of the thesis concerns that, whereas intention recognition has been extensively studied in small scale interactive settings, there is a major shortage of modeling research with respect to large scale social contexts, namely evolutionary roles and aspects of intention recognition. Employing our intention recognition method and the tools of evolutionary game theory, this thesis explicitly addresses the roles played by intention recognition in the nal outcome of cooperation in large populations of self-regarding individuals. By equipping individuals with the capacity for assessing intentions of others in the course of social dilemmas, we show how intention recognition is selected by natural selection, opening a window of opportunity for cooperation to thrive, even in hard cooperation prone games like the Prisoner's Dilemma. In addition, there are cases where it is di cult, if not impossible, to recognize the intentions of another agent. In such cases, the strategy of proposing commitment, or of intention manifestation, can help to impose or clarify the intentions of others. Again using the tools of evolutionary game theory, we show that a simple form of commitment can lead to the emergence of cooperation; furthermore, the combination of commitment with intention recognition leads to a strategy better than either one by itself. How the thesis should be read? We recommend that the thesis be read sequentially, chapter by chapter [1-2-3-4-5-6-7-8]. However, for those more interested in intention recognition from the AI modeling perspective, i.e. the rst subgoal of the thesis, Chapters 6 and 7 can be omitted and Chapters 4 and 5 are optional [1-2-3-(4)-(5)-8]. In addition, for those more keen on the problem of the evolution of cooperation, i.e. the second subgoal of thesis, Chapter 3 and even Chapter 2, can be omitted [1-(2)-4-5-6-7-8].

Fundação para a Ciência e Tecnologia - PhD grant (ref. SFRH/BD/62373/2009)