IntElicit: Eliciting and Assessing Contextualized Creativity via Dialogue Policy Optimization
Researchers have developed IntElicit, a novel framework designed to assess creativity in interactive, AI-mediated learning environments. This system uses dialogue policy optimization to adaptively provide knowledge and agency scaffolds, thereby reducing confounding factors like domain expertise and willingness to engage. Through a decomposed process reward mechanism, IntElicit encourages participants to generate creative content and reasoning, rather than simply providing answers. Experiments, including a study with 64 human participants, demonstrated that IntElicit elicits better creative outcomes compared to existing methods. AI
IMPACT Introduces a new method for evaluating creativity in AI-assisted learning, potentially improving educational assessments.