Interpretable Directed Diversity: Leveraging Model Explanations for Iterative Crowd Ideation

AI explanations can scale human feedback: The paper demonstrates that automated AI explanations (Attribution, Contrastive Attribution, and Counterfactual Suggestions) can replace human facilitators in providing feedback to large crowds, addressing a key scalability challenge in creativity support tools.

Explanations improve outcomes beyond scores alone: User studies showed that providing explanations alongside quality and diversity scores resulted in measurably better ideation diversity compared to either no feedback or scores-only feedback, suggesting that understanding why matters for improvement.

Users value actionable, directed guidance: Participants appreciated that AI explanations "helped focus their efforts and provided directions for improvement," indicating that explainable AI can make automated feedback feel more useful and trustworthy than opaque algorithmic judgments.

Multiple explanation types serve different purposes: The paper employs three complementary explanation approaches—Attribution (what factors mattered), Contrastive Attribution (what would change the score), and Counterfactual Suggestions (how to improve)—suggesting that responsible AI feedback benefits from multi-faceted transparency rather than single explanation methods.

Iterative improvement requires interpretability: The research shows that when people can understand AI reasoning through explanations, they can use that feedback to iteratively refine their work, demonstrating that interpretability is not just an ethical requirement but a practical enabler of human-AI collaboration.