Damore you know: Expertise in AI

Catherine Stinson and Sofie Vlaad

Artificial intelligence (AI) has been widely described as being in an ethical crisis. Some examples include involvement in targeted harassment of immigrants at the US border, increasing use of facial recognition for surveillance and policing despite evidence that the tools are racially biased, and evidence that recommendation algorithms drive radicalization and political division. Alongside these reports are calls to action. Both critics and tech CEOs call for more regulation. Codes of ethics and ethics boards are springing up. Many people are calling for more diversity in the field.

This paper addresses diversity as a means to addressing AI’s ethical crisis. That AI needs more diversity in its workforce is frequently announced in headlines, but the discussion under these headlines often fails to specify what is meant by diversity, or to connect the dots between diversity and AI’s ethical problems. While diversity initiatives based on identity politics show small hope of bettering ethical outcomes in AI, feminist epistemology suggests approaches with considerably more promise.

There remain the practical problems of finding effective means of increasing diversity, as well as making the case for diversity palatable to decision-makers in AI. To a lot of people in the field, talk of diversity in AI comes off as a PR move, and diversity initiatives look like pandering to special interest groups. This point of view can be seen in a leaked internal discussion by Microsoft employees that questions the importance of diversity, claims that diversity initiatives are discriminatory against white men, and worries that diversity hires are bad for the bottom line.

Some skepticism about measures claiming to address AI ethics is understandable given that high profile AI ethics groups lack the transparency, oversight powers, and independence needed to be effective, making them look very much like empty PR exercises. Furthermore, despite decades of investment in women’s participation in computing, the gender gap has grown, not shrunk. The culture of academic computer science and the tech industry, which lionizes genius and devalues non-technical expertise, poses a barrier to change driven from outside criticism. To address these practical hurdle to increasing diversity in AI, the most promising avenue may be to frame the argument for diversity in mathematical form, following the example of work in social epistemology.