Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 31·5 min read
Not All Explanations Are Sought: Information-Seeking Psychology for Human-Centered XAI
TL;DR
Human-centered explainable AI (HCXAI) should integrate insights from information-seeking psychology, focusing on when and why users desire explanations. This approach aims to enhance user engagement with AI explanations by addressing cognitive biases that affect decision-making.
✦ Why It Matters
Design AI systems that prioritize user motivations for seeking explanations to enhance decision-making.
Key Takeaways
- Incorporate user motivations into AI explanation design.
- Address cognitive biases to improve information-seeking behavior.
- Focus on when users want explanations, not just availability.
- Enhance decision-making by anticipating user needs for understanding.
Full Summary
Human-centered explainable AI (HCXAI) can benefit from understanding how people seek information, as proposed by Sharot and Sunstein's framework. Users assess explanations based on three expected utilities: instrumental (improving actions), hedonic (enhancing feelings), and cognitive (boosting understanding).
These evaluations are influenced by cognitive biases such as overconfidence and confirmation bias, which can lead to either excessive or insufficient information-seeking. This is particularly critical for agentic AI systems, where users must anticipate consequences and assess risks.
By focusing on making explanations sought rather than merely available, HCXAI can better align with user needs and improve decision-making processes. The paper advocates for designing systems that consider the motivations behind users' desire for explanations.