Cognition-Inspired Interactive Frameworks for Human-AI Alignment

Ensuring human-AI alignment is crucial as AI systems increasingly influence high-stake decision-making.However, existing approaches often rely on static interpretations of user intentions, failing to account for the dynamic and evolving nature of human goals, values, and cognitive biases.In my dissertation, I draw on cognition-inspired interactive frameworks that foster bidirectional and co-adaptive alignment between humans and AI systems.By integrating cognitive theories into the design of both model training and interactive systems, we facilitate co-adaptive learning between humans and AI, empowering users to iteratively define, refine, and adapt alignment criteria.My work thus far highlights novel approaches for balancing model and human learning, enhancing collaboration, and supporting alignment in both individual and multi-stakeholder contexts. CCS Concepts Human-centered computing Systems and tools for interaction design.