Agents Corner Opening the black box of trust: reasoning about trust models in a BDI agent ANDREW KOSTER and MARCO SCHORLEMMER, Artificial Intelligence Research Institute IIIA-CSIC, Universitat Autònoma de Barcelona, Bellaterra, Catalonia, Spain. E-mail: andrew@iiia.csic.es; marco@iiia.csic.es JORDI SABATER-MIR, Artificial Intelligence Research Institute IIIA-CSIC, Bellaterra, Catalonia, Spain. E-mail: jsabater@iiia.csic.es Abstract Trust models as thus far described in the literature can be seen as a monolithic structure: a trust model is provided with a variety of inputs and the model performs calculations, resulting in a trust evaluation as output. The agent has no direct method of adapting its trust model to its needs in a given context. In this article, we propose a first step in allowing an agent to reason about its trust model, by providing a method for incorporating a computational trust model into the cognitive architecture of the agent. By reasoning about the factors that influence the trust calculation the agent can effect changes in the computational process, thus proactively adapting its trust model. We give a declarative formalization of this system using a multi-context system and we show that three contemporary trust models, BRS, ReGReT and ForTrust can be incorporated into a BDI reasoning system using our framework. Keywords: Trust, BDI, multi-context systems. 1 Introduction Research into trust models is currently an active topic in the domain of multi-agent systems (MAS). Many computational trust models have been proposed [27], based on theoretical foundations from many different disciplines. For example, some trust models have cognitive foundations [8, 14, 30], others are based on mathematical methods, such as statistical models [36, 37] or game theory [2] and still others use a social network-oriented approach [11] or are oriented towards specific applications, such as negotiation [35] or the semantic web [34]. These trust models have in common that they are computational methods for calculating an agent’s trust in a trustee based on the agent’s own interactions with the trustee, as well as on information that is available in the environment about the trustee. Such information may be direct communications from other agents in the system, giving their own trust evaluations of the trustee; it may be reputation information; or it may be any other source of information available in the system. The trust model then aggregates this information, using the chosen mathematical method, and calculates the evaluation of the trustee. The problem with the trust models discussed so far in the literature is that an agent is unable to change its trust model if it discerns a change in the environment. If we were to view the trust model from the agent’s perspective, it would appear to be a ‘black box’ with as input the various information sources and as output an evaluation of how trustworthy the trustee is. However, as argued in [4], trust is not just an evaluation of a trustee, but an integral part of the decision making process of an agent in a social environment. For a trust evaluation to be meaningful in this process, it may be necessary to customize the evaluation process to the decision that is being made. This is especially so in an open MAS, where the environment may change. Vol. 23 No. 1, © TheAuthor, 2012. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com Published online March 16, 2012 doi:10.1093/logcom/exs003 at CSIC on February 27, 2014 http://logcom.oxfordjournals.org/ Downloaded from