ExplainabilityTransparencyTrustworthinesscs.AI
The Twin-System Approach as One Generic Solution for XAI: An Overview of ANN-CBR Twins for Explaining Deep Learning
Mark T. Keane, Eoin M. Kenny
May 20, 2019
International Joint Conference on Artificial Intelligence
14 citations
Key Takeaways
-
Twin systems create interpretable explanations by pairing black-box AI models with more transparent "white-box" alternatives: Rather than trying to make complex deep learning systems directly interpretable, this approach maps them to simpler, more understandable systems that can explain the original model's decisions.
-
Case-Based Reasoning (CBR) systems serve as practical explainers for neural networks: By pairing artificial neural networks with CBR systems, researchers can provide explanations in the form of similar past cases—a method that mirrors how humans often explain decisions by referencing comparable examples.
-
Key challenges remain in making twin systems truly useful: The paper identifies three critical open questions that need resolution: determining the best methods for weighing which features matter most, deciding how to present explanatory cases to users effectively, and identifying which types of cases actually provide meaningful explanations to people.
-
This approach offers a generic, scalable solution to the XAI problem: Rather than developing explanation methods specific to each AI application, the twin-system framework provides a reusable strategy that can be applied across different deep learning contexts.
Abstract
The notion of twin systems is proposed to address the eXplainable AI (XAI) problem, where an uninterpretable black-box system is mapped to a white-box 'twin' that is more interpretable. In this short paper, we overview very recent work that advances a generic solution to the XAI problem, the so called twin system approach. The most popular twinning in the literature is that between an Artificial Neural Networks (ANN ) as a black box and Case Based Reasoning (CBR) system as a white-box, where the latter acts as an interpretable proxy for the former. We outline how recent work reviving this idea has applied it to deep learning methods. Furthermore, we detail the many fruitful directions in which this work may be taken; such as, determining the most (i) accurate feature-weighting methods to be used, (ii) appropriate deployments for explanatory cases, (iii) useful cases of explanatory value to users.