A Very Short History of Statistics, Machine Learning, and Artificial Intelligence from a Social Sciences Perspective

Abstract

This article argues that modern artificial intelligence and classical statistical methods share common methodological foundations. Starting with Gauss’s least-squares method (1801), the paper shows that ordinary least-squares regression is mathematically equivalent to the simplest neural network. The narrative traces the parallel evolution of statistical tools in social sciences from Galton and Pearson through structural equation modeling, item response theory, and causal inference and machine learning, from early neural networks to transformers and large language models. Recognizing these shared roots can facilitate productive exchange: AI methods extend the regression toolkit to new data types, while social science methodology can bring rigor to AI research on human behavior.