XAI: Techniques and Challenges
Keywords: Explainability, Artificial Intelligence, Feature Representation
Abstract: XAI is an important field as AI technologies are becoming more are more powerful and influential in our daily life. In this paper, We introduce the concept of explainability from perspectives of psychology, philosophy, and AI, and discuss its significance in bridging the gap between model interpretability and human understandability. We outline key goals for XAI and highlight the challenges in achieving these objectives, demonstrated with an empirical experiment with GPT-4. We review solutions from the literature under categories of transparent models and post-hoc explainability techniques. As a summary, we emphasize the importance of XAI in amid the growing complexity of AI models and their profound impact on society.
Submission Number: 222
Loading