Chapter 2 Related Work
2.2 Multi-agent Systems Characteristics
Multi-agent systems exist in a broad range of domains and can be applied to many different applications, from simple agents used in information retrieval and information filtering to more complex agents used in air-traffic control. In this section, we define an agent and its behaviour by describing its key characteristics. We also define the domain characteristics, which help to shape our understanding of agents, the type of environment they exist in and the nature of their interactions.
2.2.1 Agent Characteristics
The three key agent characteristics are autonomy, heterogeneity and communication. Forautonomy, we adopt the view of Luck and d’Inverno [76] that autonomous agents derive their autonomy from motivations. In comparison, an agent is defined as an in- stantiation of an object together with an associated goal or set of goals. Autonomous agents pursue their own agendas for reasoning and behaviour in accordance with their internal motivations. Based on Kunda’s work in the field of psychology [60], a motiva- tion is defined as any desire or preference that can lead to the generation and adoption of goals and which affects the outcome of the reasoning or behavioural task intended to satisfy those goals. Thus, an autonomous agent is differentiated from an agent by the goals it possesses and which are generated from its motivations, rather than adopted from other agents.
Four types of agents can be identified [122], categorised according to the agent characteristics of heterogeneity and communication: homogeneous non-communicating, homogeneous communicating, heterogeneous non-communicating, and heterogeneous communicating agents. The level of heterogeneity refers to how similar or different agents are with respect to their internal structure, goals, domain knowledge and actions. The other agent aspect ofcommunicationdefines the degree to which the agents com- municate with one another. We focus on heterogeneous communicating agent systems, which can be complex and powerful, and consequently have a number of domain-related issues that we are considering next. Based on the characteristics identified by Stone and Veloso [122], we outline those we believe are most representative of the e-commerce domain. Besides the autonomy, heterogeneity and communication characteristics, other relevant ones are as follows.
Deliberative Agents as compared to reactive agents are capable of adapting their behaviour according to their internal state, past history and decision making.
Local Perspective in decentralised systems involves not having a global view of the environment. From their local views, agents have a partial picture of the agent system.
Modelling Other Agents’ State as even though agents are able to communicate with one another, due to reasons such as privacy, agents need to model the state, actions and knowledge of other agents. Modelling involves observation of agent behaviour and interactions and predicting future moves.
Benevolence versus Competitiveness We consider agents to be primarily selfish, as they look after their own interests and aim to achieve their individual goals. In some situations, agents may choose to be altruistic and give recommendations to others, in exchange for reciprocal behaviour.
Commitment/Decommitment Agents make commitments to one another when they communicate and decide on how they are to cooperate on a particular task. The commitments provide means for agents to trust that the committing agent will do what it initially agreed to do.
Other characteristics may be expressed by agents, however in this work, we are not focussing on them and we assume that if present, there are supporting mechanisms in place. Examples include negotiation, resource management and communication method.
Negotiation is a process by which a group of agents communicate with one another to try and come to a mutually acceptable agreement on some matter [75]. The role of negotiation is to ensure that an agreement is reached for the tasks involved to be
performed, ideally in such a way that all the parties involved benefit from the negotiation outcome. Another characteristic isresource management, where agents may have some interdependent actions due to limited common resources. Communication method is also an important characteristic of the agent domain. Since heterogeneous agents are built by different designers, there needs to be a common language and protocol for agents to interact with. We assume that the method and format of communication has been established in the later sections of this work.
2.2.2 Domain Characteristics
Population Size is the number of agents in the system, which varies according to the domain and can range from a few, to several dozens and several hundreds in electronic commerce, and electronic supply chains.
Time Dependency of Actions relates to whether the generation of actions is subjec- tive to time pressures. The type of domains we are considering are real-time and agents’ behaviours and actions are influenced by the behaviour of others in the system, as well as environmental factors. For instance, customer agents in an e-commerce system will stop buying from a supplier as it becomes increasingly unreliable.
Dynamism of Agents involves agents entering or leaving the system at will, depending on their goals at various points in the transaction period. Additionally, agents can adapt their behaviour accordingly.
Communication Cost in the domains we are considering is assumed to be almost free [127], as reciprocal behaviour benefits agents when they share information.
e-supply chains are medium. In contrast, air traffic control is a domain with high cost of failure.
User Involvement pertains to the degree of human involvement in the multi-agent systems. We assume that humans are involved, for instance, to give user feedback on the performance of its representative agents, whose behaviour can consequently be updated.
Environmental Uncertainty can result from the domain itself, from agents not know- ing the actions of other agents, and from the agents not knowing the outcome of their own actions [17].