IntElicit: Eliciting and Assessing Contextualized Creativity via Dialogue Policy Optimization

The paper introduces IntElicit, a framework designed for eliciting and assessing contextualized creativity through dialogue policy optimization. It employs a decomposed process reward mechanism to encourage participant reasoning in multi-turn interactions, addressing challenges like sparse rewards and reward hacking. The findings, based on simulations and a study with 64 participants, indicate that IntElicit enhances creative outcomes compared to traditional assessment methods, highlighting its potential for improving creativity evaluation in AI-mediated environments.