AI for High-Stakes Decision Making with Hima Lakkaraju - #387
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
00:00
Exploring Weaknesses in AI Explanation Methods
This chapter focuses on the shortcomings of local and global explanation techniques in AI, particularly perturbation-based methods like LIME. It delves into how adversarial classifiers can mislead interpretability tools, making it difficult to identify biases in AI models, especially when race influences predictions. The discussion emphasizes ongoing research aimed at developing more robust explanation methods to improve fairness and transparency in machine learning.
Transcript
Play full episode
