Rationalize framework lets humans and AI reason together in role pairs

A new research paper accepted at the ACM CHI 2026 BiAlign Workshop introduces Rationalize, a framework for shared semantic reasoning between humans and AI models. The authors—Aritra Dasgupta, Naga Datha Saikiran Battula, Avina Nakarmi, Sohom Sen, Subhodeep Ghosh, and Xun Song—conceptualize human-AI interaction as a series of complementary role pairs: Explorer-Guide, Investigator-Informant, Teacher-Student, and Judge-Advocate. Each pair operates in a shared reasoning space where both sides make their purposes, questions, assumptions, evidence, inferences, and implications explicit.

This approach moves beyond aligning AI to humans at the output level (e.g., generating correct answers) to aligning at the rationalization level—making the reasoning process itself transparent and negotiable. The paper relates these role pairs to a bidirectional alignment framework, distinguishing between 'aligning AI to humans' and 'aligning humans to AI' depending on the role. For instance, in the Teacher-Student pair, the human teaches the AI; in the Judge-Advocate pair, the AI serves as a critic. The authors propose a collaborative research agenda for designing and assessing alignment using element-level and role-specific approaches, with applications in data-driven sensemaking tasks.