On Human Predictions with Explanations and Predictions of Machine Learning Models

Abstract

Examines how different types of AI explanations affect human prediction accuracy. The study finds that not all explanations are created equal — some improve human performance while others introduce new biases.

Key Findings

• Example-based explanations outperform feature-based explanations for human decision-making
• AI explanations can introduce anchoring bias in human predictions
• The effectiveness of explanations depends on the user's domain expertise

Source