The Impact of Power and Norms on Agent Behaviour

We understand society as a set of autonomous agents sharing a common environment and making interactions in order to overcome their own limitations by using other agent capabilities, and where all the members are controlled by the same social regulations. Representing agents that live there requires, first, the specification of an agent architecture, which includes elements concerning the social behaviour of agents, and second, to provide different strategies that agents may use in order to take advantages of the opportunities that society gives them.

Previous analysis to existing agent architectures, enabled us to identify some of the key points to be taken into account for such purpose. Firstly, agents not only must work to achieve goals, they must be able to generate them. Secondly, an agent has to model other coexisting agents to exploit their capabilities. Thirdly, agent limitations create dependence relationships which, in turn, define power relationships that are neither static nor unrelated to the chosen agent architecture, these relationships always change according to the current goals, plans, capabilities, intentions and norms of the involved agents. Fourthly, social agents must internally represent the norms of the society where they live and must be willing to adopt and fulfil them according to their own motivations and preferences. Finally, reasoning about dependence, normative and power relationships, gives agents different motives to delegate some of their goals or adopt the goals of others and therefore the needed abilities to make agreements, form coalitions and even create complex organisations.

As a result, the main aims of this research project are first, to analyse how dependence, normative and power relationships may affect the behaviour of agents, and second, to create a framework, which allow their formal representation. Although previous work has been focused on these social concerns, it has been done in a discrete and somewhat arbitrary fashion. By contrast, we have preferred a basic architecture where new elements and operations become easily integrated by following the methodology provided by the SMART agent framework described in [“Understanding Agent Systems”. d'Inverno and Luck. Springer. 2001]. So far, we have specified an architecture for autonomous social and normative agents, which besides having traditional characteristics of deliberative agents, allow them to represent and reason about the norms that control the society where they lives. Current research is being done in four main issues: norm reasoning, power relationships, goal adoption and finally goal delegation and plan selection. In the following paragraphs a summary of each one is presented.

a). Norm reasoning

Norm adoption and norm compliance processes. These processes must decide when a new norm can be recognised as a duty and when a norm already adopted must be fulfilled respectively. Adopting a norm does not necessarily mean that the agent has to do what a norm says. This latter must be decided until a norm becomes applicable, and considerations about the current agent’s goals and motivations become included. In order to do that, first an analysis of the reasons why an agent adopts/abandons a norm must be provided. In fact, we consider that an agent has to be motivated in order to adopt a norm. Mostly, agents adopt norms as a means to achieve other important goals, and therefore motivations associated with norms could be inherited from these goals. In this way, once the goals become fulfilled, agents may have no reasons for complying with norms and they could be abandoned. Now, regarding norm compliance, the fulfilment of a norm is generally linked to deliberations about punishments (fear to be fined), rewards (desires to get something) and social motivations (honesty, social responsibility, etc). By contrast, its violation is associated with considerations about the cost of compliance, possible conflicts with important current intentions and goals, or with the impossibility of being punished.

b). The Power Model

Social Power Theory [Castelfranchi 90] sets up the basis to explain why many forms of social interactions occur. However, although some of the main principles have been already formalised, the dynamics of dependence and powers have not been expressed yet, mainly because its formalisation has been done independently of the agent architecture and, consequently, its effects on different processes of deliberation have not been studied yet. In particular, we believe that these relationships can be used in all the process of reasoning where other agents have to be involved.

c). Plan Selection and Goal Delegation

A rational agent must select a plan that guarantees the satisfaction of the selected goal. It is sensible to think that the best plan may be one in which all actions and sub-goals can be satisfied by the agent itself. However, this rarely can be done and some external help may be needed. This is precisely the moment when agents may require to delegate one of their goals and consequently, to initiate interactions. By selecting a plan, an agent is thinking about its future activities and therefore, criteria such as the possibility of success in the delegation of a sub-goal seem to be important. In this way, an agent may select a plan including subgoals that although cannot be satisfied by itself, can be easily delegated to other agents. Four important issues are related here: the mechanisms for influencing other agents, the selection of a plan, the selection of a cooperation partner and the delegations of goals.

d). Goal Adoption

An autonomous agent has to have good reasons to work for a goal that it does not generate. Sometimes, for example, due to its reliance on others, an agent is prone to be influenced, and in order to achieve one of its goals, the agent is willing to adopt an external goal. This happens, for example, when an agent agrees to work for another agent in exchange for money to satisfy its other goals, or when an agent agrees to do something for someone else expecting a future reciprocation. Other times, goals are adopted because a norm exists and obliges an agent to accept the other’s goals. In any case, once an external goal becomes adopted, that is, the agent is convinced that it has to do it, the goal must be treated as any other of its goals and consequently, external goals must also be motivated. Representing this characteristic of agents requires the following activities. An analysis of the different motives to adopt external goals must be provided and then formalised. This part must be the counterpart of goal delegation, we have to represent here how an agent is influenced. Implications of goal adoption must also be considered. In particular how external goals are related to internal motivations must be explained and then formalised. Social commitments are possible involved in this phase because, sometimes, although the main motivations to adopt an external goals disappear, the adopted goal must remain until it become satisfied as part of the social responsibility of agents.

Publications

· F. Lopez y Lopez, M. Luck and M. d'Inverno, A Framework for Norm-based Inter-Agent Dependence, in Proceedings of the Third Mexican International Conference on Computer Science, 2001.