Twin Systems for DeepCBR: A Menagerie of Deep Learning and Case-Based Reasoning Pairings for Explanation and Data Augmentation

Deep learning models can be made more explainable by pairing them with case-based reasoning systems: The research demonstrates that opaque deep learning models can be paired with transparent case-based reasoning (CBR) models to provide explanations for AI decisions, addressing a key challenge in responsible AI implementation.

Multiple types of explanations are possible through this approach: The twin-system framework enables three distinct explanation strategies—factual (what actually happened), counterfactual (what would need to change for a different outcome), and semi-factual (alternative scenarios)—giving stakeholders different ways to understand AI decisions.

Deep learning's internal features can be leveraged for transparency: The research shows that functional abstractions from deep learning models, such as feature weights, feature importance, and decision boundaries, can be extracted and used to power explainability solutions, suggesting that transparency doesn't require completely replacing existing AI systems.

This approach may also improve AI training data quality: Beyond explanation, the researchers suggest that combining deep learning with case-based reasoning could help with data augmentation—the process of expanding training datasets—which could lead to more robust and fair AI systems.

Responsible AI doesn't require choosing between performance and interpretability: By creating "twin systems" that keep the original deep learning model functional while adding a transparent reasoning layer, this approach suggests organizations can maintain AI performance while gaining the explainability needed for responsible deployment.