Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Aug 7·5 min read
Epistemic Trustworthiness in Generative AI: A Normative Framework for Warranted Reliance in High-Stakes Workflows
TL;DR
A normative framework for assessing epistemic trustworthiness in generative AI is proposed, focusing on conditions for justified reliance in high-stakes contexts. It emphasizes epistemic humility, access, and resistance to injustice as essential for responsible AI use.
✦ Why It Matters
Evaluate your generative AI systems against the framework to ensure responsible use in high-stakes environments.
Key Takeaways
- Assess AI systems for epistemic humility to acknowledge their limitations.
- Ensure users can question and contest AI outputs for better epistemic access.
- Recognize users as legitimate knowledge holders to resist epistemic injustice.
- Focus on warranted reliance rather than just output accuracy in AI design.
How It Works
The framework defines epistemic trustworthiness through three interrelated conditions: systems must communicate their limitations (epistemic humility), allow user engagement with outputs (epistemic access), and validate user knowledge (resistance to epistemic injustice).