ExplainabilityTransparencyTrustworthinesscs.LG
Towards Analogy-Based Explanations in Machine Learning
Eyke Hüllermeier
May 23, 2020
Modeling Decisions for Artificial Intelligence
23 citations
2 influential
Key Takeaways
-
Analogies offer a human-friendly explanation method: The paper argues that analogy-based explanations—where a model's decision is explained by comparing it to similar past cases—can make AI decisions more understandable to non-experts, complementing existing explanation approaches in explainable AI.
-
Analogical reasoning goes beyond just improving accuracy: While analogy-based methods have been used to build more accurate machine learning models, the paper emphasizes their overlooked value for interpretability, suggesting that explainability should be considered equally important when designing AI systems.
-
Analogies can work alongside other explanation techniques: Rather than replacing current explainability methods, analogy-based explanations are positioned as a complementary tool that can work together with similarity-based explanations to provide more complete insights into how AI systems make decisions.
-
This approach bridges AI and human reasoning: By grounding explanations in analogical reasoning—a natural way humans justify their own decisions—this method could help make AI systems more transparent and trustworthy to the people who use and are affected by them.
Abstract
Principles of analogical reasoning have recently been applied in the context of machine learning, for example to develop new methods for classification and preference learning. In this paper, we argue that, while analogical reasoning is certainly useful for constructing new learning algorithms with high predictive accuracy, is is arguably not less interesting from an interpretability and explainability point of view. More specifically, we take the view that an analogy-based approach is a viable alternative to existing approaches in the realm of explainable AI and interpretable machine learning, and that analogy-based explanations of the predictions produced by a machine learning algorithm can complement similarity-based explanations in a meaningful way. To corroborate these claims, we outline the basic idea of an analogy-based explanation and illustrate its potential usefulness by means of some examples.