A Logical Understanding of Legal Interpretation

If compliance with a norm does not achieve its purpose, then its applicability must dynamically be restricted or expanded. Legal interpretation is a mechanism from law allowing norms to be adapted to unforeseen situations. We model this mechanism for norms regulating computer systems by representing the purpose of norms by social goals and by revising the constitutive rules defining the applicability of norms. We illustrate the interpretation mechanism by examples. Introduction and Motivation Norms regulating computer systems can be modelled in different ways, see, for example, (Boella, van der Torre, and Verhagen 2008). If norms are represented by hard constraints, then computer systems are designed to avoid violations. Alternatively, if norms are represented by soft constraints, then they provide standards that can be violated, and violations should result in sanctions. Soft constraints allow agents to optimize their performance by reasoning about the trade off between respecting the norm and the risk of being sanctioned. However, an agent can respect a norm even when the compliance of the norm does not benefit the system, thus wasting his resources while the system achieves only a suboptimal state. If norm compliance does not achieve the purpose of a norm, then its applicability must dynamically be restricted or expanded. Legal interpretation is a mechanism from law allowing norms to be adapted to unforeseen situations. The research question of this paper is: • How to model the interpretation mechanism of law, in order to design more flexible computer systems regulated by norms? We model norms as plans aiming to achieve the social goals the members of a society decided to share (Boella and van der Torre 2007). The legislator could try to specify all the circumstances to which a norm applies and all the exceptional contexts where it does not apply, but, as well known in the planning community of AI, universal plans rarely are a practicable strategy. An agent should rather produce a partial plan and revise it when part of it becomes unfeasible. Copyright c © 2010, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. In the same way as replanning allows an agent to revise its plans while keeping fixed its original goals, legal interpretation allows norms to be adapted after their creation to the unforeseen situations in order to achieve the social goal they have been planned for. To model legal interpretation we use the distinction between regulative norms (obligations, prohibitions and permissions) from constitutive rules. While the former, which are changed only by the legislative system, specify the ideal behaviour, the latter ones provide, by means of counts-as definitions, an ontology of institutional concepts (Boella and van der Torre 2004). The applicability conditions of legal rules very often refer to these institutional concepts, rather than to so called brute facts. To simplify the notation we refer to the former as constitutive rules, and the latter simply as norms. To model the revision of norms, we use an extension of Defeasible Logic (DL) (Governatori and Rotolo 2008). Legal Interpretation Norms have a conditional structure such as b1, . . . ,bn ⇒O l (if b1, . . . ,bn hold, then l is obligatory); an agent is compliant with respect to this norm if l is obtained whenever b1, . . . ,bn is derived. Most logical models of legal reasoning assume that conditions of norms give a complete description of their applicability (Sartor 2005). However, this assumption is too strong, due to the complexity and dynamics of the world. Norms cannot take into account all the possible conditions where they should or should not be applied, giving rise to the so called “penumbra”: a core of cases which can clearly be classified as belonging to the concept. By a penumbra of hard cases, membership of the concept can be disputed. Moreover, not only the world changes, giving rise to circumstances unexpected to the legislator who introduced the norm, but even the ontology of reality can change with respect to the one constructed by the law to describe the applicability conditions of norms. See, e.g., the problems concerning the application of existing laws to privacy, intellectual property or technological innovations in healthcare. To cope with unforeseen circumstances, the judicial system, at the moment in which a case concerning a violation is discussed in court, is empowered to interpret, i.e., to change norms, under some restrictions not to go beyond the purpose from which the norms stem. The distinction between norms and constitutive rules suggests that legal interpretation does not amount to revising norms, but to interpreting legal concepts, i.e., to revising constitutive rules. In this paper we adopt the view that the ontology of legal concepts is built via constitutive rules having the so-called counts-as form (Searle 1995): r : a1, . . . ,an ⇒c b. For example, a bicycle is considered as a vehicle by the following constitutive rule: r0 : Bike(x)⇒c Vehicle(x) This counts-as rule, if instantiated by any bicycle a, says that a counts as a vehicle. Constitutive rules have a defeasible character, for example, a bicycle for children cannot be considered as a vehicle: r1 : Bike(x),ForChildren(x);c ¬Vehicle(x)