The Stochastic Parrot in the Coal Mine: Model Collapse is a Threat to Low-Resource Communities
Abstract
Model collapse, the degradation in performance that arises when generative models are trained on the outputs of prior models, has largely been studied in the abstract. This work argues that model collapse is not just a theoretical curiosity but a threat to low-resource and marginalised communities. By reducing training efficiency and skewing data distributions away from the tails of their support, model collapse disproportionately impacts these communities, both environmentally and culturally. We examine these implications and propose strategies to mitigate the disparate impact of model collapse on low-resource communities.
Devon Jarvis
Lecturer
I am a lecturer at Wits interested in studying systematic generalization and the emergence of modularity in the brain and machines.
Richard Klein
PRIME Lab Director
I am an Associate Professor in the School of Computer Science and Applied Mathematics at the University of the Witwatersrand in Johannesburg, and a co-PI of the PRIME lab.
Benjamin Rosman
Lab Director
I am a Professor in the School of Computer Science and Applied Mathematics at the University of the Witwatersrand in Johannesburg. I work in robotics, artificial intelligence, decision theory and machine learning.
