Trust Alignment: A Sine Qua Non of Open Multi-agent Systems Andrew Koster 1,2 , Jordi Sabater-Mir 1 , and Marco Schorlemmer 1,2 1 IIIA - CSIC 2 Universitat Aut`onoma de Barcelona Bellaterra, Spain Abstract. In open multi-agent systems trust is necessary to improve co- operation by enabling agents to choose good partners. Most trust models work by taking, in addition to direct experiences, other agents’ communi- cated evaluations into account. However, in an open multi-agent system other agents may use different trust models and as such the evaluations they communicate are based on different principles. This article shows that trust alignment is a crucial tool in this communication. Furthermore we show that trust alignment improves significantly if the description of the evidence, upon which a trust evaluation is based, is taken into ac- count. 1 Introduction A prerequisite for cooperation is that an agent may reasonably expect this coop- eration to succeed. The cooperating agents need to know that their interaction partner will perform the action it agreed to. In many systems this can be en- forced by the architecture of the system, however in open systems in which the individual agents maintain their autonomy, such as e-Commerce or smart elec- tricity grids, this type of guarantee is not available and agents may be capable of cheating, lying or performing other unwanted behaviour. In such open multi- agent systems the agents need to choose selectively whom to cooperate with and trust is a fundamental tool for performing this selection. Unfortunately, it is more complicated than equipping an agent with one of the available computational trust models [1] and expecting it to function in a social environment. Using trust as a method for picking successful cooperation partners relies not only on having a good trust model, but also on communica- tion of trust evaluations with other agents [2]. This communication is far from straightforward, because trust is an inherently subjective concept [3]. In this pa- per we show that to communicate trust evaluations between agents some form of trust alignment is needed. The subjectivity of trust can be seen in the following example, which also demonstrates why this is problematic for communication: consider an e-Com- merce environment in which two agents buy the same bicycle via an online auction. One may evaluate the sale as very successful, because the bicycle was R. Meersman, T. Dillon, and P. Herrero (Eds.): OTM 2011, Part I, LNCS 7044, pp. 182–199, 2011. c  Springer-Verlag Berlin Heidelberg 2011