Extracting PAC Decision Trees from Black Box Binary Classifiers: The Gender Bias Case Study on BERT-based Language Models
This work addresses the challenge of trust in explainable AI for binary classification, specifically in detecting bias in language models, though it is incremental as it adapts existing methods.
The authors tackled the problem of ensuring fidelity in decision trees extracted from black box AI models by applying the Probably Approximately Correct (PAC) framework to provide theoretical guarantees, and they demonstrated this on BERT-based language models, revealing occupational gender bias.
Decision trees are a popular machine learning method, known for their inherent explainability. In Explainable AI, decision trees can be used as surrogate models for complex black box AI models or as approximations of parts of such models. A key challenge of this approach is determining how accurately the extracted decision tree represents the original model and to what extent it can be trusted as an approximation of their behavior. In this work, we investigate the use of the Probably Approximately Correct (PAC) framework to provide a theoretical guarantee of fidelity for decision trees extracted from AI models. Based on theoretical results from the PAC framework, we adapt a decision tree algorithm to ensure a PAC guarantee under certain conditions. We focus on binary classification and conduct experiments where we extract decision trees from BERT-based language models with PAC guarantees. Our results indicate occupational gender bias in these models.