A new position paper argues that explainable AI (XAI) research needs to shift focus from developing numerous ad-hoc methods to addressing fundamental challenges. The paper highlights that current XAI techniques, such as feature attributions and sparse autoencoders, rarely influence real-world workflows due to a lack of methodologies for integrating explanations into human-in-the-loop systems. The authors propose a move towards clearer problem formulations, better evaluation objectives, and pipelines for explanation-driven feedback to create more actionable AI systems. AI
IMPACT This research advocates for a shift in AI explainability, aiming to make AI systems more actionable and human-centered by focusing on foundational methodologies rather than isolated techniques.
RANK_REASON The cluster contains a research paper published on arXiv discussing foundational aspects of explainable AI. [lever_c_demoted from research: ic=1 ai=1.0]
- arXivLabs
- Conference on Neural Information Processing Systems
- explainable AI
- feature attributions
- Hugging Face
- International Conference on Learning Representations
- International Conference on Machine Learning
- Sparse Autoencoders
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
