Agents That Work in Harmony by Knowing and Fulfilling Their Obligations

Societies constrain the behavior of agents by imposing multiple, often contradictory, obligations and interdictions amongst them. To work in harmony, agents must find ways to satisfy these constraints, or to break less important ones when necessary. In this paper I , we present a solution to this problem based on a representation of obligations and interdictions in an organizational framework, together with an inference method that also decides which obligations to break in contradictory situations. These are integrated in an operational, practically useful agent development language that covers the spectrum from defining organizations, roles, agents, obligations, goals, conversations to inferring and executing coordinated agent behaviors in multi-agent applications. One strength of the approach is the way it supports negotiation by exchanging deontic constraints amongst agents. We illustrate this and the entire system with a negotiated solution to the feature interaction problem in the telecommunications industry. Introduction and Motivation Working together in harmony requires that everybody fulfils their obligations and respects everybody else’s rights. In other words, it requires that everybody respects the social laws of their community. To build agents that can be trusted to work with and on behalf of humans in organizations requires the same thing, that agents know and fulfil their obligations while respecting the rights and authority of humans and of other agents in the organzation. Multiple symultaneous obligations and interdictions require agents to find the right behavior that achieves the goals induced by obligations without violating the interdictions. Often, there is no way to find the right behavior without violating less important obligations or interdictions in order to ensure the more important ones are fulfilled. Current models of collective behavior often oversimplify this situation. The Cohen-Levesque account of teamwork (Levesque, Cohen &: Nunes 90) for example, and the implemented systems based on it (Jennings 95; 1Copyright (~)1998, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. Tambe 97) assume that all members of a team have essentially a single, implicit, obligation towards a common mutual goal. In this paper we build on a different model of social interaction, one that explicitely represents and integrates multiple obligations. 1. At the social level, our model assumes that societies and organizations constrain the social behavior of agents by imposing social laws, representable as networks of mutual obligations and interdictions amongst agents. Not fulfiling an obligation or interdiction is sanctioned by paying a cost or by a loss of utility, which allows an agent to apply rational decision making when choosing what to do. Social laws are objective forces motivating social behavior and to a large extent determine the ’attitudes’ at the individual agent level. Agents ’desire’ and ’intend’ the things that are requested by their current obligations, knowing that otherwise there will be a cost to pay. At the individual agent decision level, each agent decides what behavior to adopt to satisfy the applicable social laws as well as its own goals and priorities. In particular, at this level agents determine how to solve conflicting obligations and interdictions. Having decided on the general behavior in terms of what to do or not, agents need to plan/schedule the activities that compose the selected behavior. This determines the precise sequencing of actions to be executed, consistent with time, resource and possibly other constraints on action execution. Finally, actions have to be executed as planned, with provisions for handling exceptions and violations. These may be dealt with at any of the above levels, for example by retrying, replanning, deciding on different actions or even (in an extreme case that we do not deal with) trying to modify the social laws. To integrate obligations in this framework we rely on (1) a representation semantically founded on dynamic deontic logic of social laws as obligations, permissions and interdictions among the roles that agents play in an organization and (2) a constraint propagation reasoning method allowing agents to infer the applicable obliga,2.