Is Artificial Intelligence Gender-Free? What Does Feminist Epistemology Say About That?

Abstract

I start my contribution with some general questions: Is AI gendered? Is AI sexist or can it be? Does AI include gendered knowledge and suppositions? If so, how?

After that, I proceed to develop my theoretical (i.e., qualitatively based) starting points with the main referential authors, Donna Haraway and Alison Adam. The fact that impersonal does not mean observer-independent (as Haraway described it in a slightly different context [1997]) is a good reason to turn to feminist epistemology, especially its concept of situated knowledges. Since knowledge (or the representation of knowledge) lies at the very centre of AI research, this makes it an appropriate vehicle for a gendered critique of AI (Adam, 2000).

The concept of situated knowledges entails knowledge that reflects a perspective on a subject which is necessarily partial and limited, not universal (this is Haraway’s famous critique of the “view from nowhere”). Namely, there is no way to be simultaneously in all of the epistemologically privileged positions structured by gender, class, nation, etc.

I then proceed to AI research. The knowledge engineers build systems that contain knowledge reflecting their own interests and competencies. While this representation of knowledge is usually regarded as being universal (Adam, 2000), it is hierarchical since it does not grant epistemic authority to all. Most importantly, social exclusivism and biological essentialism are re-inscribed in the ontology of AI (Adam, 2000). I address the question of which effects social and political contaminations and prejudices can bring for the development of AI.

I suggest that unless we commit to deconstructing the harmful essentialisms that govern our human lives, we might just be perpetuating the same (i.e., our own and others’) practices of domination and unequal parts of privilege and oppression (Haraway, 1991) in developing AI.