Systems of autonomous and self-interested agents interacting to achieve individual and collective goals may exhibit undesirable or unexpected properties if left unconstrained. Using deontic concepts of obligations, permissions and prohibitions to describe, what must, may and should not be done, norms have been widely proposed as a means of defining and enforcing societal constraints. Recent efforts to provide norm-enabled agent architectures that limit plan choices suffer from interfering with an agent’s reasoning process, and thus limit autonomy more than is required by the norms alone. In response, in this paper we describe nu-BDI, an extension of the BDI architecture, which enables normative reasoning, providing agents with a means to choose and customise plans (and their constituent actions), so as to ensure compliance with norms. The paper makes three significant contributions, in providing: fine-grained tailoring of plan restrictions; a plan annotation mechanism to identify violating plans, and limit possible plan instantiations; and a technique permitting the selective and incremental violation of norms in cases where goal achievement would not otherwise be possible.
|Title of host publication||Proceedings of the 4th Annual Conference of the International Technology Alliance|
|Publication status||Published - 2010|