Algorithmic Collective Action in Machine Learning

Institute Homepage

Institute Homepage Sign In

Back

Research Overview

Performativity in machine learning

Measuring performative power of online search

Performative Prediction: Past and Future

Applications of algorithmic collective action

Living artefact of collective action cases

Large language models and surveys

Questioning the survey responses of LLMs

Folktexts

Evaluating LLMs as risk scores

Social Foundations of Computation Members Publications

Algorithmic Collective Action in Machine Learning

Aca — The collective plants a hidden rare signal in the data and exploits the signal at test time. Success grows rapidly as a function of the fraction of workers participating. Experiments with language models on freelancer resumes confirm our theoretical prediction.

We propose a theoretical model of algorithmic collective action. Participating workers modify their data in a coordinated manner in order to steer the learning algorithm toward a desired outcome. Even small collectives can have significant influence [].

Members

Algorithms and Society

Celestine Mendler-Dünner

Hector Endowed Fellow of the ELLIS Institute

Social Foundations of Computation

Moritz Hardt

Director

Publications

Social Foundations of Computation Algorithms and Society Conference Paper Algorithmic Collective Action in Machine Learning Hardt, M., Mazumdar, E., Mendler-Dünner, C., Zrnic, T. In Proceedings of the 40th International Conference on Machine Learning (ICML 2023), PMLR, The Forty International Conference on Machine Learning (ICML), July 2023 (Published) URL BibTeX