Science gets stuck in local optima, just like an AI

This paper argues that scientific knowledge, at any given time, is not the best possible understanding of nature. Instead, it is shaped by historical accidents, ingrained ways of thinking, and institutional pressures. This means science often gets trapped in less-than-ideal explanations, missing better ones.

For centuries, people assumed science was always moving towards the absolute truth. This paper suggests that science, like a machine learning algorithm, can get stuck in a 'local minimum' — a good enough answer that prevents it from finding a truly optimal one. This means that entire fields of study might be built on foundations that are merely convenient, not fundamentally correct. It challenges the idea that scientific progress is a straight line to ultimate truth.