Timnit Gebru is the founder and executive director of the Distributed Artificial Intelligence Research Institute (DAIR). Prior to that she was fired by Google in December 2020 for raising issues of discrimination in the workplace, where she was serving as co-lead of the Ethical AI research team. She received her PhD from Stanford University, and did a postdoc at Microsoft Research, New York City in the FATE (Fairness Accountability Transparency and Ethics in AI) group, where she studied algorithmic bias and the ethical implications underlying projects aiming to gain insights from data. Timnit also co-founded Black in AI, a nonprofit that works to increase the presence, inclusion, visibility and health of Black people in the field of AI, and is on the board of AddisCoder, a nonprofit dedicated to teaching algorithms and computer programming to Ethiopian highschool students, free of charge.
The past 3 years of work in NLP have been characterized by the development and deployment of ever larger language models, especially for English. BERT, its variants, GPT-2⁄3, and others, most recently Switch-C, have pushed the boundaries of the possible both through architectural innovations and through sheer size. Using these pretrained models and the methodology of fine-tuning them for specific tasks, researchers have extended the state of the art on a wide array of tasks as measured by leaderboards on specific benchmarks for English. In this paper, we take a step back and ask: How big is too big? What are the possible risks associated with this technology and what paths are available for mitigating those risks? We provide recommendations including weighing the environmental and financial costs first, investing resources into curating and carefully documenting datasets rather than ingesting everything on the web, carrying out pre-development exercises evaluating how the planned approach fits into research and development goals and supports stakeholder values, and encouraging research directions beyond ever larger language models.