A research paper proposes a framework for evaluating human-AI collaboration by quality and cost, showing that interaction cost can vary greatly even for sessions with identical quality ratings. This highlights the importance of considering dialogue analysis in AI system design and evaluation.