Understanding the Effect of Accuracy on Trust in Machine Learning Models

Abstract

This study investigates how stated and observed accuracy levels influence human trust in machine learning models. The findings reveal that trust is not a simple linear function of accuracy — people have complex mental models of what accuracy means and how it should affect their reliance on AI systems.

Key Findings

• Stated accuracy significantly influences initial trust in ML models
• Trust is resilient to observed inaccuracies when stated accuracy is high
• There is a threshold effect where trust drops sharply below certain accuracy levels

Source