ExplainabilityTransparencyTrustworthinesscs.AI
Bridging the Gap Between Explainable AI and Uncertainty Quantification to Enhance Trustability
Dominik Seuß
May 25, 2021
arXiv.org
21 citations
Key Takeaways
-
Explainability and uncertainty quantification are complementary tools that have been developed separately but should work together: The paper argues these two research areas have evolved independently despite their potential to reinforce each other in building trustworthy AI systems, suggesting a significant gap in current responsible AI frameworks.
-
Combining explanations with uncertainty measures creates more honest AI systems: By pairing explanations of why an AI made a decision with quantified measures of how confident the system is in that decision, users get a more complete and realistic picture of the AI's reliability rather than false certainty.
-
Current responsible AI frameworks are incomplete without addressing both interpretability and confidence: The paper positions explainability and uncertainty quantification as essential but underexplored components of responsible AI, alongside fairness and other properties, indicating that trustability requires a more integrated approach.
-
This integration would directly enhance user trust through transparency: When AI systems can both explain their reasoning and acknowledge the limits of their knowledge, users can make better-informed decisions about whether and how to rely on those systems.
Abstract
After the tremendous advances of deep learning and other AI methods, more attention is flowing into other properties of modern approaches, such as interpretability, fairness, etc. combined in frameworks like Responsible AI. Two research directions, namely Explainable AI and Uncertainty Quantification are becoming more and more important, but have been so far never combined and jointly explored. In this paper, I show how both research areas provide potential for combination, why more research should be done in this direction and how this would lead to an increase in trustability in AI systems.