To interact safely with humans, AI agents must not only know social norms but also consider them during planning. A new study introduces an approach that helps AI plan with dynamically changing norms.
Developments
The research team contributed a defeasible calculus framework to resolve norm conflicts and use them as "guardrails" for action plans. This approach was tested through SocialBot, an agent handling natural language conversational tasks.
Unlike previous systems that often treat norms as static, this model allows agents to adapt to the dynamic nature of social regulations in real-world environments.
Why it matters
Integrating social norms into the core of the planning process, rather than just as a post-processing layer, helps AI behave in a more human-like and safer manner. In Vietnam, as chatbots and service robots begin to weave into daily life, the ability to understand and adhere to flexible rules of conduct will be key to the widespread social acceptance of AI.