Learning capabilities of agents in social systems

In a social computational system, there exist not only social interactions between software agents but also between humans and agents. Through interactions with humans, agents can acquire more knowledge, e.g., in problem solving. Usually, agents are hard-coded with anticipated abilities and their knowledge cannot evolve dynamically. In this paper, we propose a strategy-based approach to enable agents learning from humans in conflict situations. The learning process consists of four phases: 1) the conflict between a human and an agent is detected, 2) the human initiates a communication with the agent and proposes a strategy to solve the conflict, 3) the human´s strategy is evaluated, and 4) the agent applies the most effective strategy in a new similar situation. The contribution of the paper is two-fold: it presents a new agent learning approach in the area of multi-agent learning and proposes a way of cooperation between humans and agents in a social computational system to evolve agents´ abilities.