In open environments in which autonomous agents can break contracts, computational models of trust have an important role to play in determining who to interact with and how interactions unfold. To this end, we develop such a trust model, based on confidence and reputation, and show how it can be concretely applied, using fuzzy sets, to guide agents in evaluating past interactions and in establishing new contracts with one another. Agents generally interact by engaging in some form of negotiation process which results in them making commitments to (contracts with) one another to carry out particular tasks (Jennings et al. 2001). However, in most realistic environments, there is no guarantee that a contracted agent will actually enact its commitments (because it may defect to gain higher utility or because there is uncertainty about whether the task can actually be achieved). In such situations, computational models of trust (here defined as the positive expectation that an interaction partner will act benignly and cooperatively in situations where defecting would prove more profitable to itself [Dasgupta 1998]) have an important role to play. First, to help determine the most reliable interaction partner (i.e., those in which the agent has the highest trust). Second, to influence the interaction process itself (e.g., an agent’s negotiation stance may vary according to the opponent’s trust level). Third, to define the

Abstract. This research aims to develop a model of trust and reputation that will ensure good interactions amongst software agents in large scale open systems in particular. The following are key drivers for our model: (1) agents may be self-interested and may provide false accounts of experiences with other agents if it is beneficial for them to do so; (2) agents will need to interact with other agents with which they have no past experience. Against this background, we have developed TRAVOS (Trust and Reputation model for Agent-based Virtual OrganisationS) which models an agent’s trust in an interaction partner. Specifically, trust is calculated using probability theory taking account of past interactions between agents. When there is a lack of personal experience between agents, the model draws upon reputation information gathered from third parties. In this latter case, we pay particular attention to handling the possibility that reputation information may be inaccurate. 1

The ability to create reliable and scalable virtual organisations (VOs) on demand in a dynamic, open and competitive environment is one of the major challenges that underlie Grid computing. In response, in the CONOISE-G project, we are developing an infrastructure to support robust and resilient virtual organisation formation and operation. Specifically, CONOISE-G provides mechanisms to assure effective operation of agent-based VOs in the face of disruptive and potentially malicious entities in dynamic, open and competitive environments. In this paper, we describe the CONOISE-G system, outline its use in the context of VO formation and perturbation, and review current efforts to progress the work to deal with unreliable information sources. 1

Cooperation among autonomous agents involves an inherent degree of uncertainty. Agents determine for themselves when to initiate cooperation or to assist others, when to rescind commitments, and how to conduct cooperative tasks. For example, an agent may delay the execution of a cooperative task, execute it to a reduced quality, or simply fail to complete it. In this paper, we describe how experience-based trust can be used to minimise the risk associated with cooperation. In particular we propose a mechanism, called multi-dimensional trust, which allows agents to model the trustworthiness of others according to various criteria. This trust information is combined with other factors to enable the selection of cooperative partners. Agents ’ preferences are represented by a set of factor weightings, which allow trust information to be tailored to the current cooperative priorities. We also describe the experimental validation of our proposed approach. 1

Decentralized peer-to-peer (P2P) applications are characterized by the absence of a central authority or infrastructure that coordinates the behavior of entities in the system. These entities, called peers, interact directly with each other and make local autonomous decisions in order to achieve their individual goals. In the absence of a single authority that maintains all the data and handles all the queries, peers themselves are responsible for seeking, storing, and sharing information efficiently. Placing a large amount of information on every peer or broadcasting a request to every peer greatly reduces the performance and efficiency of the system. Hence, it is essential that decentralized applications employ efficient storage mechanisms and reliable search mechanisms. Further, an open decentralized system that does not regulate the joining of peers can be subject to grave risks. In particular, malicious peers may be encouraged to resort to a variety of attacks, including sending spurious information, posing as other peers, etc. It is important for each peer in the system to defend against such attacks. This survey discusses these two essential issues that characterize P2P decentralized applications: storage and discovery mechanisms, and trust management. It identifies and defines key properties for each of these and also summarizes the efforts of the P2P community in addressing these properties by categorizing and discussing relevant technologies and approaches.

In this paper, we propose a Bayesian network-based trust model. Since trust is multi-faceted, even in the same context, agents still need to develop differentiated trust in different aspects of other agents ’ behaviors. The agent’s needs are different in different situations. Depending on the situation, an agent may need to consider its trust in a specific aspect of another agent’s capability or in a combination of multiple aspects. Bayesian networks provide a flexible method to present differentiated trust and combine different aspects of trust. A Bayesian network-based trust model is presented for a file sharing peer-to-peer application. 1.

Abstract. In this paper, we propose a Bayesian network-based trust model in peerto-peer networks. Since trust is multi-faceted, even in the same context, peers still need to develop differentiated trust in different aspects of other peers ’ behaviors. The peer’s needs are different in different situations. Depending on the situation, a peer may need to consider its trust in a specific aspect of another peer’s capability or in a combination of multiple aspects. Bayesian networks provide a flexible method to represent differentiated trust and combine different aspects of trust. 1.

Abstract. In this paper, we propose a reputation oriented reinforcement learning algorithm for buying and selling agents in electronic market environments. We take into account the fact that multiple selling agents may offer the same good with different qualities. In our approach, buying agents learn to avoid the risk of purchasing low quality goods and to maximize their expected value of goods by dynamically maintaining sets of reputable sellers. Selling agents learn to maximize their expected profits by adjusting product prices and by optionally altering the quality of their goods. Modelling the reputation of sellers allows buying agents to focus on those sellers with whom a certain degree of trust has been established. We also include the ability for buying agents to optionally explore the marketplace in order to discover new reputable sellers. As detailed in the paper, we believe that our proposed strategy leads to improved satisfaction for buyers and sellers, reduced communication load, and robust systems. In addition, we present preliminary experimental results that confirm some potential advantages of the proposed algorithm, and outline planned future experimentation to continue the evaluation of the model.

Abstract. Trust is a fundamental factor when people are interacting with each other, hence it is natural that trust has been researched also in relation to applications and agents. However, there is no single definition of trust that everybody would share. This, in turn, has caused a multitude of formal or computational trust models to emerge to enable trust use and dependence in applications. Since the field is so diverse, there also exists a confusion of terminology, where similar concepts have different names and, what is more disturbing, same terms are also used for different concepts. To organize the research models in a new and more structured way, this paper surveys and classifies thirteen computational trust models by the trust decision input factors. This analysis is used to create a new comprehensive ontology for trust to facilitate interaction between business systems. 1

The vision of ubiquitous computing is becoming a reality thanks to the advent of portable devices and the advances in wireless networking technologies. It aims to facilitate user tasks through seamless utilization of services available in the surrounding environments. In such distributed environments featuring openness, interactions, especially service provision and consumption, between entities that are unknown or barely known to each other, are commonplace. Trust management through reputation mechanism to facilitate such interactions is recognized as an important element of ubiquitous computing. It is, however, faced by the problems of how to stimulate reputation information sharing and honest recommendation elicitation. We present in this paper an incentive compatible reputation mechanism to facilitate the trustworthiness evaluation in ubiquitous computing environments. It is based on probability theory and supports reputation evolution and propagation. Our reputation mechanism not only shows robustness against lies, but also stimulates honest and active recommendations. The latter is realized by ensuring that active and honest recommenders, compared to inactive or dishonest ones, can elicit the most honest (helpful) recommendations and thus suffer the least number of wrong trust decisions, as validated by simulation based evaluation. 1