HAAM: Human-AI Accuracy Model

Python Version License

Implementation of the Double Machine Learning Lens Model Equation (DML-LME) for analyzing perceptual accuracy in high-dimensional settings. HAAM quantifies how humans and AI systems achieve accuracy when making judgments, decomposing their decision-making processes into interpretable components.

What is HAAM?

The Human-AI Accuracy Model addresses a fundamental question: When humans and AI achieve similar accuracy levels, are they using the same perceptual cues and cognitive strategies?

HAAM provides a rigorous statistical framework to:

  • Decompose judgment accuracy into direct and mediated pathways

  • Quantify the Percentage of Mediated Accuracy (PoMA) for any perceiver

  • Compare how humans vs AI utilize high-dimensional perceptual features

  • Handle thousands of features using debiased machine learning

Key Features

  • 🎯 DML-LME Implementation: Double Machine Learning Lens Model Equation for high-dimensional perception

  • 📊 PoMA Calculation: Quantify what percentage of accuracy flows through measured perceptual cues

  • 🧠 Human-AI Comparison: Statistical framework for comparing perceptual strategies

  • 📈 Rich Visualizations: 3D UMAP projections, PCA analysis, word clouds

  • 🔍 Topic Modeling: Automatic discovery and labeling of content themes via BERTopic

  • 📉 Comprehensive Metrics: Correlations, regression coefficients, cross-validated R², PoMA decomposition

Contents

API Reference

Indices and tables