Whistleblowing Stochastic Parrots: A Sense of Artificial Intelligence Ethics and Advocacy

Abstract

On December 2^nd, 2020, former Google Artificial Intelligence Researcher Timnit Gebru announced that she was fired from Google after she refused to retract a paper that called out the social, environmental, and bias risks of large language models. She raised awareness about biases in artificial intelligence (AI), which sparked conversations about the use of AI within policing systems, given the possibility for disproportionate harm marginalized communities could face from these systems. Her paper’s key contention was that AI centered on large language models, systems—like OpenAI’s ChatGPT and Google’s newly created PaLM2—that extracted harmful amounts of data, endangering the environment and pulling from biased sources. Armed with the knowledge of whistleblowing risks, Timnit was faced with an important decision: should she go public about the circumstances of her firing? What if she pushed back and exposed Google’s research culture and their mistreatment? What risks would she face in doing so? How would the public react?