Case Study: Google DeepMind reduces gender bias in online group leadership with Prolific

Google DeepMind finds pseudonymity cuts peer exclusion by 56% with Prolific

Google DeepMind, working with researchers from the Paris School of Economics, sought to understand and mitigate gender bias in online group dynamics to inform the design of fairer AI systems. To conduct this research at scale, they partnered with Prolific to recruit nearly 1,000 verified participants from the US and UK for a complex, synchronous group study.

Using Prolific's platform to recruit and manage participants, the study measured leadership selection bias in an online task. The results revealed that while hiding demographic cues (pseudonymity) reduced bias in peer selection, a significant bias in self-nomination persisted. This research provided Google DeepMind with crucial, high-quality human data, demonstrating that Prolific is capable of supporting complex, multi-stage behavioral experiments essential for frontier AI evaluation.

Google DeepMind