arXiv cs.AIThursday · May 28, 2026FREE

Reasoning and Planning with Dynamically Changing Norms

A new research paper, "Reasoning and Planning with Dynamically Changing Norms," published on arXiv cs.AI on May 28, 2026, addresses a critical challenge in AI agent development: enabling safe interaction with humans by incorporating dynamic social norms into planning. Traditional norm-guided planning for AI agents has often overlooked the evolving nature of human norms and primarily focused on interactions within communities of artificial agents. This paper shifts the focus to human-AI settings, proposing a novel framework for agents to understand and adapt to changing normative landscapes. The core contributions include a defeasible calculus designed to effectively resolve conflicts that arise when multiple norms are present or when norms change over time. Additionally, the research introduces an approach where these dynamically changing norms function as "guard rails," guiding and constraining an AI agent's planning process to ensure its actions remain socially acceptable and aligned with human expectations. The efficacy of this approach is supported by formal theoretical proofs and empirical validation. An AI agent named SocialBot was used to demonstrate the system's capabilities in a natural language dialogue task, showcasing its ability to navigate complex social interactions while adhering to evolving human norms. This work aims to foster more robust and context-aware human-AI collaboration.