Specification and Validation of Normative Rules for Autonomous Agents

Abstract

A growing range of applications use autonomous agents such as AI and robotic systems to perform tasks deemed dangerous, tedious or costly for humans. To truly succeed with these tasks, the autonomous agents must perform them without violating the social, legal, ethical, empathetic, and cultural (SLEEC) norms of their users and operators. We introduce SLEECVAL, a tool for specification and validation of rules that reflect these SLEEC norms. Our tool supports the specification of SLEEC rules in a DSL [1] we co-defined with the help of ethicists, lawyers and stakeholders from health and social care, and uses the CSP refinement checker FDR4 to identify redundant and conflicting rules in a SLEEC specification. We illustrate the use of SLEECVAL for two case studies: an assistive dressing robot, and a firefighting drone.

Social, Legal, Ethical, Empathetic, and Cultural Rules: Compilation and Reasoning

“…Thus, simple formalisms, supported by a wide range of readilyavailable software support, are sufficient for developing SLEEC-sensitive AI systems. Yaman et al (2023) have presented SLEECVAL, a tool designed for the validation of SLEEC rules through the mapping of rules to tock-CSP processes. Subsequently, a CSP refinement checker can be employed to detect redundant and conflicting rules.…”