Explainable artificial intelligence (XAI) is a rapidly growing research field that has received a lot of attention during the last few years. An important goal of the field is to use its methods to detect (social) bias and discrimination. Despite these positive intentions, aspects of XAI can be in conflict with feminist approaches and values. Therefore, our conceptual contribution brings forward both a careful assessment of current XAI methods, as well as visions for carefully doing XAI from a feminist perspective. We conclude with a discussion on the possibilities for caring XAI, and the challenges that might lie along the way.
relations and how (structural) discrimination emerges through them; 2) a special focus on the relationship between gender identities, and the necessity to think beyond the gender binary; 3) acknowledging care and care work in our society, its (often intentional) invisibility, and implications for the redestribution of this work.
Following these key aspects of feminism, and central work that is already existing at the
intersection between feminism, feminist epistemology, care work, data and AI [
To do this, our work is asking the following two research questions (RQ): ) How can we assess current methods of XAI from a feminist viewpoint? and ) How do we imagine future versions of XAI, and which changes to the current methodologies does this entail?
We can dissect RQ 1) using a couple of follow-up questions: Which aspects of existing XAI methods interfere with our working definition of feminism? How does existing XAI interact with power relations: do they perpetuate, or foster to challenge them? Which aspects of existing XAI methods can be considered feminist? Why? Regarding the RQ 2), we will ask questions such as: how could a feminist explanation look like? Which technical and social conditions are needed along the way? How do we imagine a feminist future towards caring XAI?
Answering the posed questions is an experimental and explorative endeavor, set out to forward a meaningful and critical contribution to the XAI community over techno-determinist solutionism [17]. As we will give answers to these questions, the issues that are raised might not be solved. Instead, they allow for further questioning and research.
This work in progress is of interdisciplinary character, bringing together perspectives from the social sciences, the computer sciences, and our shared interest in feminist theories, concepts and methodologies. While it is important to understand the global dimension of XAI and AI, we acknowledge our positionality as researchers that are based in Europe, and with European institutions.
Outlook A feminist perspective is an important and necessary addition to XAI and AI - it brings a view to the table that is historically neglected or marginalized and ofers possibilities of ”studying up” instead of ”studying down” [18, 19]. This is especially important if we reason about (social and historical) discrimination, and about how this discrimination can be counter-acted.
If we do not integrate a feminist perspective, we will build a world of (X)AI that leaves some people outside of it.
Acknowledgments This work has received funding from the European Union’s Horizon 2020 research and innovation programme under Marie Sklodowska-Curie Actions (grant agreement number 860630) for the project “NoBIAS - Artificial Intelligence without Bias” ( nobias-project.eu). This work reflects only the authors’ views and the European Research Executive Agency (REA) is not responsible for any use that may be made of the information it contains. IEEE Transactions on Cognitive and Developmental Systems (2020) 1–1. doi:10.1109/TCDS. 2020.3044366, conference Name: IEEE Transactions on Cognitive and Developmental Systems. [16] A. Kasirzadeh, Reasons, Values, Stakeholders: A Philosophical Framework for Explainable Artificial Intelligence, in: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’21, Association for Computing Machinery, New York, NY, USA, 2021, p. 14. URL: http://doi.org/10.1145/3442188.3445866. doi:10.1145/3442188.3445866. [17] E. Morozov, To Save Everything, Click Here: The Folly of Technological Solutionism,
PublicAfairs, New York, 2013. [18] L. Nader, Up the Anthropologist: Perspectives Gained From Studying Up, Technical Report, 1972. URL: https://eric.ed.gov/?id=ED065375, eRIC Number: ED065375. [19] C. Barabas, C. Doyle, J. Rubinovitz, K. Dinakar, Studying up: reorienting the study of algorithmic fairness around issues of power, in: Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, FAT* ’20, Association for Computing Machinery, New York, NY, USA, 2020, pp. 167–176. URL: https://dl.acm.org/doi/10.1145/ 3351095.3372859. doi:10.1145/3351095.3372859.