Human-AI Decision Making

Core Contribution

Empirical research on how humans make decisions with AI assistance. Examines conditions for appropriate reliance and complementary performance.

Key Findings

Complementarity Challenge: Human-AI teams often underperform best individual performer Reliance Calibration: Users struggle to know when to trust AI Explanation Effects: Explanations can increase trust without increasing accuracy

Design Implications

Uncertainty communication matters more than explanation detail
Training on AI limitations improves calibration
Interface design shapes reliance patterns

Appropriate Reliance

The goal: users trust AI when it’s right, override when it’s wrong. Achieving this requires:

Accurate confidence communication
User understanding of AI limitations
Low-cost override mechanisms

>heyMHK

Human-AI Decision Making

Human-AI Decision Making

Core Contribution

Key Findings

Design Implications

Appropriate Reliance

Properties

Graph view

Table of Contents