DESIGNING NORMATIVE BEHAVIOUR VIA LANDMARKS Huib Aldewereld, Davide Grossi, Javier V´ azquez-Salceda, and Frank Dignum Institute of Information and Computing Sciences Utrecht University, The Netherlands {huib, davide, javier, dignum}@cs.uu.nl Abstract. In highly regulated environments, where a set of norms de- fines accepted behaviour, protocols provide a way to reduce complex- ity by giving direct, step by step guidelines for behaviour, as long as the protocols comply with the norms. In this work we propose a formal framework to design a protocol from a normative specification. In order to be able to connect (descriptive) norms with (operational) protocols, an intermediate level is created by the use of landmarks. 1 Introduction In the last years there has been an explosion of new approaches, both theoretical and practical, focusing on normative specifications as a flexible way to structure, restrict and/or impose behaviour in multiagent systems (MASs).In particular, recent developments focus on norm languages, agent-mediated electronic insti- tutions, contracts, protocols and policies. Our work focuses on a normative ap- proach based on the use of norms in electronic institutions (eInstitutions). Norms are high-level specifications of acceptable behaviour within a given context. De- finitions of norms range from very philosophical, in deontic logic, to precise specifications of protocols in agent-mediated eInstitutions. One of the questions that arises is how to properly connect the norm spec- ification with the behaviour of the agents. Norms are usually defined in some form of deontic logic [19], in order to express accepted (legal) behaviour through obligations, permissions and prohibitions. However, it is hard to directly connect this kind of norms with the practice as: 1. Norms in law are formulated in a very abstract way, i.e., the norms are ex- pressed in terms of concepts that are kept vague and ambiguous on purpose. 2. Norms expressed in deontic logic are declarative, i.e., they have no operational semantics (they express what is acceptable, but not how to achieve it). 3. As Wooldridge and Ciancarini explain in [24], in those formalisms and agent theories based in possible worlds, there is usually no precise connection be- tween the abstract accessibility relations used to characterise an agent’s state and any computational model. This makes it difficult to go directly from a formal specification to an implementation in a computational system.