Rashomon in the Streets: Explanation Ambiguity in Scene Understanding

2025-09-03

AI Summary

In safety-critical tasks such as autonomous driving motion prediction, eXplainable Artificial Intelligence (XAI) confronts the Rashomon effect—multiple high-performing models yield substantially divergent explanations for identical predictions, undermining explanation credibility and uniqueness. Method: This work presents the first empirical quantification of the Rashomon effect in real-world driving scenarios. We model scenes using Qualitative eXplainable Graphs (QXGs), integrate paired gradient-boosted trees with graph neural networks, and systematically assess cross-model and cross-architecture explanation consistency via feature attribution methods. Contribution/Results: Despite comparable predictive performance, models exhibit pervasive explanation divergence—both within and across architectures—demonstrating that explanation ambiguity is an intrinsic property of the task, not an artifact of modeling bias. Our findings expose fundamental reliability limits of XAI in autonomous driving and establish a novel benchmark for explainability evaluation and robust XAI design.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Transparent, Interpretable, Explainable ML

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Explainable AI (XAI) is essential for validating and trusting models in safety-critical applications like autonomous driving. However, the reliability of XAI is challenged by the Rashomon effect, where multiple, equally accurate models can offer divergent explanations for the same prediction. This paper provides the first empirical quantification of this effect for the task of action prediction in real-world driving scenes. Using Qualitative Explainable Graphs (QXGs) as a symbolic scene representation, we train Rashomon sets of two distinct model classes: interpretable, pair-based gradient boosting models and complex, graph-based Graph Neural Networks (GNNs). Using feature attribution methods, we measure the agreement of explanations both within and between these classes. Our results reveal significant explanation disagreement. Our findings suggest that explanation ambiguity is an inherent property of the problem, not just a modeling artifact.

Problem

• Quantifying explanation ambiguity in action prediction
• Evaluating Rashomon effect across interpretable and complex models
• Assessing explanation reliability in autonomous driving systems

Innovation

• Uses Qualitative Explainable Graphs for symbolic scene representation
• Trains Rashomon sets with gradient boosting and Graph Neural Networks
• Measures explanation agreement using feature attribution methods