A Computational Framework for Norm-Aware Reasoning in Autonomous Systems

Abstract

Autonomous agents are increasingly deployed in complex social environments where they not only have to reason about their domain goals but also about the norms that can impose constraints on task performance. To accomplish this, robots must be able to reason not only on how to perform their tasks, but also incorporate societal values, social norms and legal rules so they can gain human acceptability and trust. The intelligent trade-offs that these systems must make between domain and normative constraints is the key to any system being socially responsible and acceptable. Integrating task planning with norm aware reasoning is a challenging problem due to the curse of dimensionality associated with product spaces of the domain state variables and norm-related variables. In this work, we propose a Modular Normative Markov Decision Process (MNMDP) framework that is shown to have orders of magnitude increase in performance compared to previous approaches. The MNMDP framework applies normative reasoning considering only the norms that are activated in appropriate contexts, rather than considering the full set of norms, thus significantly reducing computational complexity.

Since norms are both context-dependent as well as context-sensitive, we must model context in an expressive, scalable and compact manner in order to find the activation and deactivation conditions for norms. To this end, we propose a generalizable context modeling approach to understand norm activations in social environments combining the expressiveness of propositional logic with the compactness of decision trees. We show how we can combine our context model with our MNMDP framework to support norm understanding as well as norm enforcement for real systems. We discuss the inferences obtained from the human experiments data that we conducted confirming the complexity of the relationship between contexts and norms. We demonstrate the effectiveness of our approach through scenarios in simulated social environments in which agents using the framework display norm-aware behavior. We also show the significant computational improvements that we obtain when using our proposed approach for computationally modeling social interactions.