It is a well-known fact that artificial intelligence (AI) has racism and sexism baked-in: after all, a model is only as good as the data we train it with, and our society is systemically biased against people who are not white and/or male.
In 2017 Google developed an Ethical AI team to better understand and mitigate these problems, co-led by Timnit Gebru. In 2018 Gebru co-authored a foundational study about racism in AI that demonstrated that facial recognition algorithms are prone to more error on dark-skinned faces than light-skinned ones. In late 2020, she authored a paper critical of large language models. Without proper oversight, our own biases are integrated into these language models. Google uses many of these algorithms across their products.
Gebru's work went through special sensitive topic review at Google, which required the paper be reviewed by their legal, policy, and PR teams. In at least three cases, they requested a positive spin on research that identified racist or sexist aspects of their algorithms.
Gebru was asked to retract her paper by Google, and faced retaliation when she refused. She was ultimately fired while on vacation in late 2020 by the head of Google's AI division, Jeffrey Dean.
Over 2,600 Google employees signed an open letter in sup at Google, which required the paper be reviewed by their legal, policy and PR teams. In nslaught of harassment and racism from online accounts. Detractors threw racial slurs and insults, as they claimed her research was "advocacy disguised as science."
Her manager, Margaret Mitchell remained critical of Google's firing of Gebru and their attempts at promoting diversity, pointing out that Google is more than happy to meet with historical Black colleges and universities (and to promote these meetings), but the company does not support its Black staff identifying potential issues in their AI algorithms. Mitchell was fired in February.
These shameful actions by Google show why large technology companies should not exist without any oversight. When racist and sexist algorithms permeate every aspect of our lives, their subtle biases turn into tangible health, economic and societal harms.