ExplainabilityFairnessPrivacycs.AI
Beyond XAI:Obstacles Towards Responsible AI
Yulu Pi
September 7, 2023
arXiv.org
3 citations
Key Takeaways
-
Explainability alone is insufficient for responsible AI: While techniques to make AI systems more transparent have received significant attention, the paper emphasizes that responsible AI requires addressing multiple dimensions beyond just explainability, including privacy, fairness, and contestability.
-
Current explainability methods have real-world limitations: The paper identifies that existing explainability techniques and their evaluation strategies present numerous limitations when applied in actual practice, suggesting that transparency tools may not work as effectively outside controlled settings as developers expect.
-
A holistic approach is necessary: Responsible AI cannot be achieved by focusing narrowly on making AI decisions understandable; organizations must simultaneously consider how to protect user privacy, ensure fair treatment across different groups, and allow people to challenge AI decisions.
-
Evaluation of explainability is problematic: The paper highlights that not only do explainability methods have limitations, but the strategies used to evaluate whether these methods actually work are also flawed, raising questions about whether we can reliably measure transparency improvements.
Abstract
The rapidly advancing domain of Explainable Artificial Intelligence (XAI) has sparked significant interests in developing techniques to make AI systems more transparent and understandable. Nevertheless, in real-world contexts, the methods of explainability and their evaluation strategies present numerous limitations.Moreover, the scope of responsible AI extends beyond just explainability. In this paper, we explore these limitations and discuss their implications in a boarder context of responsible AI when considering other important aspects, including privacy, fairness and contestability.