Homogenizing Effects of AI in Research

This lecture, given by Will Fleisher, PhD, aims to explain less-frequently discussed harms of generative AI systems: algorithmic monoculture & homogenization. Algorithmic monoculture occurs when the same, or very similar, AI systems are widely deployed for a particular purpose. For instance, if every human resources department in an industry uses the same application evaluation algorithm, a job candidate dispreferred by the algorithm, will be unable to find work in that industry. 

Monoculture turns noise — random errors — into bias because it makes the errors systematic. A related harm, algorithmic homogenization, occurs because large language models (LLMs) are all built using the same underlying technology, methods and data. When a large percentage of a research field is using AI, subtle influences by the LLMs can shift the way that researchers ask questions, evaluate literature, and solve problems; over time, this can add up to large distortions.

Will Fleisher, PhD, is an Assistant Professor of Philosophy and a Research Assistant Professor in the Center for Digital Ethics at Georgetown University.