C. Stephanidis and M. Antona (Eds.): UAHCI/HCII 2014, Part II, LNCS 8514, pp. 475–486, 2014. © Springer International Publishing Switzerland 2014 Learning from Each Other: An Agent Based Approach Goran Zaharija, Saša Mladenović, and Andrina Granić Faculty of Science, University of Split, Nikole Tesle 12, 21000 Split, Croatia {goran.zaharija,sasa.mladenovic,andrina.granic}@pmfst.hr Abstract. This paper presents an agent based approach to knowledge represen- tation and learning methods. Agent architecture is described and discussed, together with its advantages and limitations. Main purpose of the proposed approach is to gain further insight in current teaching methods with a foremost aspiration for their improvement. Two different experimental studies were con- ducted; the first one addressing knowledge representation and the second one regarding knowledge transfer between agents. Obtained results are presented and analysed. Keywords: learning, artificial intelligence, machine learning, agent based systems. 1 Introduction There are many different approaches in agent based learning like distributive [1], cooperative [2], [3], reinforced [4] and collaborative [5] learning, but most of these approaches make strict differentiation between teacher and learner agents. We intend to present an agent based approach in which, depending on different circumstances, agents possess the ability to act both as a teacher and as a learner. Although agents will not be differentiated by their role, each of them could possess individual charac- teristics (dimensions, mobility, number and type of sensors) making them unique or at least different from each other. As a result a system that is more flexible than those aforementioned should be designed. It should also enable much simpler and efficient transfer of knowledge among all agents acting within the system. This paper aims to present a type of agent that can act both as a teacher and a learner, while using robots as physical representation of those agents. Primary reason for developing such kind of agents is to discover new or improve existing teaching methods. Accordingly, we are proposing a framework that could be used for those purposes. To successfully act as a teacher, it is desirable that agents are able to switch their role from the teacher to the student. Desired effect of such change of roles is an embracement of a same student mental model, thus allowing successful knowledge transfer between subjects and avoiding traps in form of potential misconceptions. Every single individual has its own perspective of the surrounding world (egocen- tric view) that differs from the collective or global representation of that same world (allocentric view) [6], [7]. This should be taken into consideration when talking about teaching; anyone taking the role of the teacher should be aware that not everybody shares his/hers view of the world.