Who Should Join the Decision-Making Table? Targeted Expert Selection for Enhanced Human-AI Collaboration
Keywords: human-ai complementarity, calibration, rule learning
Abstract: Integrating AI and human expertise can significantly enhance decision-making across various scenarios. This paper introduces a novel approach that leverages the Product of Experts (PoE) model to optimize decision-making by strategically combining AI with human inputs. While human experts bring diverse perspectives, their decisions may be constrained by biases or knowledge gaps. To address these limitations, we propose an AI agent that provides probabilistic, rule-based insights, complementing and filling human experts' knowledge gaps. A key feature of our approach is the strategic selection of human experts based on how well their knowledge complements or enhances the AI’s recommendations. By dynamically adapting the expert selection process, we ensure that decisions benefit from the most impactful and complementary inputs. Our PoE model calibrates inputs from both AI and human experts, leveraging their combined strengths to improve decision outcomes. Furthermore, operating in an online setting, our framework can also continuously update the AI’s knowledge and refine expert selection criteria, ensuring adaptability to evolving environments. Experiments in simulation environments demonstrate that our model effectively integrates logic rule-informed AI with human expertise, enhancing collaborative decision-making.
Primary Area: interpretability and explainable AI
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 12984
Loading