Abstract
This chapter presents organizations as a macro-micro notion and device; they presuppose autonomous proactive entities (agents) playing the organizational roles. Agents may have their own powers, goals, relationships (of dependence, trust, etc.). This opens important issues to be discussed: Does cooperation require mentally shared plans? Which is the relationship between individual powers and role powers; personal dependencies and role dependencies; personal goals and assigned goals; personal beliefs and what we have to assume when playing our role; individual actions and organizational actions? What about possible conflicts, deviations, power abuse, given the agents’ autonomy? MultiAgentSystems discipline should both aim at scientifically modeling human organizations, and at designing effective artificial organizations. Our claim is that for both those aims, one should model a high (risky) degree of flexibility, exploiting autonomy and pro-activity, intelligence and decentralized knowledge of roleplayers, allowing for functional violations of requests and even of rules.
TopIntroduction
The main thesis of this chapter is that organization is a notion, a model, and a technical device that presupposes individual agents1; it is a way of exploiting and improving the co-powers2 of agents, producing efficient cooperative results. Organizations have to be funded and grounded in the powers of the agents and in particular on and into their minds (knowledge, goals, reasoning, choice, etc.).
In fact, “Organization” is organization of something: precisely of the activities of sub-units, able to get and process information, to act in a coordinated manner, to co-operate, and to do this with some local – decentralized – autonomy, and adaptation to the context. In particular, ‘organizations’ are coordination structures for cooperation among several agents, in view of iterated and typical problems or problem-solving activities. Any ‘organization’ entails some Multi-Agent (MA) plan, and the agents have to work (case by case) within some common plan.
Some level of autonomy and of purposiveness of the component units is presupposed: the active entities able to play a role within the organization and its plans, must be able to perform actions in view of specific results, that is, to pursue and locally monitor goals (their ‘tasks’), and must have some degree of ‘autonomy’ for adapting the action to its context and for solving problems.
‘Autonomy’ is a relational notion: X is autonomous from somebody else (Y) and as for something (O) (Castelfranchi, 1995a). Autonomous means that X doesn’t depends on Y as for (having/realizing) O; does not need Y for this. X can have or realize O by itself. An active entity can be autonomous on various dimensions and for various “resources” or “activities”: for example, for accessing information and data, for learning, for reasoning, for decision making, for abilities and skills, or for external material resources (money, a car, etc.). One might be autonomous as for a given O but not for another one. For example, X may be autonomous as for deciding but not as for the needed data, because she is blind; or, X may have the practical resources for doing action A, but not being able to do it, and she needs the help of Y for this.
“Agents” - as defined in MAS - are autonomous entities (“Autonomous Agents & MAS”): there is not a fully centralized processing; they have local independent information and results, and, sometimes, independent learning or reasoning; they have some pro-activity (Wooldridge and Jennings, 1995); they are regulated by internal representations (sometimes strictly ‘purposive’ ones, like in BDI Agents). The most qualifying and advanced form of autonomy obviously is autonomy in goals: when an agent has its own independent motives, and will pursue our objectives only if there is some reason and advantage for doing so.
In sum, for us, the “sociological” and “institutional” level of processing and activity must be funded on the basic notion of agents activity: action, decision, belief, plan, etc. and in particular on the individual social action. It is a typical mistake to try to found collective levels per se’ or directly on individual agent theory ignoring sociality at the interpersonal level as the very basis of sociality at the collective level (Tuomela, 1995; Castelfranchi, 2003c). So, for example, there are typical and fundamental aspects to be modeled in organization, like ‘dependence’, or ‘coordination’, or ‘delegation’ and ‘reliance’, like ‘goal-adoption’ or help, that are already present at the interpersonal level and must be defined first of all at that level (see in this chapter Section on ‘Socio-Cognitive Organization Pillars’).
Given this premise and perspective several important issues should be taken into account in organization theory and design.
Key Terms in this Chapter
Power: Internal and external resources, skills, capabilities, conditions, that make the agent ‘able’ and ‘in condition’ to do a given action and to realize a given goal. This is the “power of” doing an action, realizing a goal. There are various forms of social power; the most basic is the “Power over” the other: Y is dependent on X as for a given goal, thus X has Power over Y as for that goal.
Goal-Adoption: Goal-Adoption is when an autonomous agent X with her own goal comes to have a new goal (and possibly pursue it), since and until she believe that it is the goal of another agent Y, in order Y realizes his goal. Goal-Adoption can be there both for altruistic or for selfish reasons. A special kind of Goal-Adoption is Goal-Adhesion: where X ‘accepts’ to help Y, since there is an (implicit or explicit) request (expectation, order, will, …) that she does something for Y.
Goal-Delegation: Agent X has certain expectations (predictions + wishes) about Y’s behavior, and relies, counts upon that behavior in order to perform her action and achieve the goal; she is allocating a part of her plan to Y. X can just count on Y’s action by exploiting Y’s autonomous action; Y can be unaware. In stronger forms, X actively assigns that part of the plan to Y (by authority or exchange or cooperative agreement or request for help, etc.); Y accepts this. This notion of Delegation is more basic and broader than the ‘institutional’ notion (usually used in organizations), where X not only allocate a part of her plan to Y, but she also ‘empowers’ Y for doing this, and passes to him some resource, and (more important) some responsibility or duty which was of X or of the group or organization.
Agent: Autonomous proactive (purposive) entities. “Agents” - as defined in MAS - are autonomous entities: there is not a fully centralized processing; they have local independent information and results, and, sometimes, independent learning or reasoning; they have some pro-activity (Wooldridge and Jennings, 1995); they are regulated by internal representations (sometimes strictly ‘purposive’ ones, like in BDI Agents); they have their own powers (skills, but possibly also resources like data). The most qualifying and advanced form of autonomy and of agency obviously is autonomy in goals: when an agent has its own independent motives, and will pursue our objectives only if there is some reason and advantage for doing so (self-motivated).
Role: The role is what a member, an agent, is supposed (or better prescribed) to do within a multi-agent plan, and within an organization; the set of assigned goals, duties, powers, assumptions. His ‘part’, or ‘share’; his ‘mansion’ and contribution to the global outcome. A role is something to be dressed, to be ‘interpreted’; a mask, a character: the agent becomes a “role player”.
Autonomy: ‘Autonomy’ is a relational notion: X is autonomous from somebody else (Y) and as for something (O) (Castelfranchi, 1995a). Autonomous means that X doesn’t depends on Y as for (having/realizing) O; does not need Y for this. X can have or realize O by itself. An active entity can be autonomous on various dimensions and for various “resources” or “activities”: for example, for accessing information and data, for learning, for reasoning, for decision making, for abilities and skills, or for external material resources (money, a car, etc.). One might be autonomous as for a given O but not for another one. For example, X may be autonomous as for deciding but not as for the needed data, because she is blind; or, X may have the practical resources for doing action A, but not being able to do it, and she needs the help of Y for this.
Organization: “Organization” is organization of the activities of sub-units, able to get and process information, to act in a coordinated manner, to co-operate, and to do this with some local – decentralized – autonomy, and adaptation to the context. In particular, ‘organizations’ are coordination structures for cooperation among several agents, in view of iterated and typical problems or problem-solving activities. Any ‘organization’ entails some Multi-Agent (MA) plan, and the agents have to work (case by case) within some common plan. One of the advantages of organizations is precisely that a lot of roles in habitual plans or in possible generative plans are pre-defined and consolidated, and the performer of that part/role has not to know the whole plan or the specific higher goal, has not to negotiate about reciprocal roles, etc. This is one of the reasons organizations are efficient and convenient (reduced cognitive costs, reduced negotiation costs, pre-established general agreement, etc.). In a sense an organization is a meta-plan: both, a plan for making specific and instantiated plans, and a generic, non-instantiated plan to be applied. So the roles in the organization are generalized roles to be instantiated case by case in specific plans.