Fundamental Limitation in Explaining AI
A recent paper discusses the inherent limitations in explaining AI systems, particularly large-scale models. The authors present a quadrilemma that highlights the challenges of achieving complexity, performance, interpretability, and faithfulness in AI explanations simultaneously. This suggests that AI governance should acknowledge the incomplete nature of explanations provided by AI systems.
- ▪The paper mathematically proves a fundamental quadrilemma in explaining AI.
- ▪The quadrilemma states that AI explanations cannot satisfy complexity, performance, interpretability, and faithfulness at the same time.
- ▪The authors recommend focusing on explaining only the important parts of AI systems for practical applications.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.24727 (cs) [Submitted on 23 May 2026] Title:Fundamental Limitation in Explaining AI Authors:Atsushi Suzuki, Jing Wang View a PDF of the paper titled Fundamental Limitation in Explaining AI, by Atsushi Suzuki and 1 other authors View PDF HTML (experimental) Abstract:While large-scale models such as LLMs and diffusion models have achieved practical success, public institutions have emphasized the importance of explainability in AI. Existing methods for explaining AI, however, are not designed to provide completely faithful explanations of the behavior of large-scale AI systems.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
