Transparent Visual Reasoning via Object-Centric Agent Collaboration

AI systems can explain their visual decisions using objects as building blocks: OCEAN breaks down image analysis into identifiable objects rather than abstract patterns, making it easier for people to understand why an AI system reached a particular conclusion—for example, identifying specific features in a medical image rather than pointing to unclear pixel regions.

Multiple AI agents working together can produce more trustworthy explanations: The framework uses game-theoretic reasoning where different agents must agree on evidence before making a decision, similar to how a team of experts might debate findings. This collaborative process resulted in explanations that users rated as more intuitive and trustworthy compared to existing explanation tools like GradCAM and LIME.

Interpretability doesn't require sacrificing accuracy: OCEAN achieved competitive performance with standard black-box visual classifiers while maintaining a transparent reasoning process, suggesting that explainability and accuracy are not inherently at odds—a key concern for deploying AI in high-stakes domains like healthcare.

User studies validate that better explanations matter: When tested with human participants, people consistently found OCEAN's explanations more intuitive and trustworthy than those from popular post-hoc explanation methods, demonstrating that technical improvements in interpretability translate to real improvements in human understanding and trust.