WHAT PLACE DOES FEMINISM HAVE IN IMPROVING MACHINE LEARNING? (DOES IT EVEN HAVE A PLACE?)

Hildah Nyakwaka
Nov 6 · 4 min read

Have you ever wondered why all the chatbots have female personas? Ever wanted your personal assistant to have an accent like yours? To understand your local language? Me too! This past weekend at the annual Mozilla Festival hosted at Ravensbourne University in London, I was privileged to sit on a panel discussing the importance of feminist principles in changing who and how machines read.

Historically and even currently, a lot of positive technology is geared towards serving and improving the lives of a certain demographic; white and mostly male, while the negative effects are felt across the board. This is made deliberate through two main ways; the makers of the technology and the datasets collected.

Four women sitting at a panel discussion on feminism and AI at MozFest 2019
Four women sitting at a panel discussion on feminism and AI at MozFest 2019
Feminism and AI panel at MozFest 2019 (Photo Credits to Philo van Kemande @phivk)

Does feminism have the capability to reshape these already existing notions and inventions? Short answer, yes. This however requires delving into more than just general feminist politics. Pushing aside everyone’s individual definition of feminism; we can reach a consensus that feminism not only advocates for the empowerment of women, but also other marginalized groups and individuals, including but not limited to people living with disabilities and the LGBTQI community. This essentially involves erasing the patriarchal limits that have traditionally bounded out and disempowered all these individuals from equal access and usage. The disempowerment is strongly manifested in their socio-economic and political lives and has extended to their digital lives.

This brings us to fundamentally important questions; Who do machines recognize as human beings? Who are left out? Why? And how can we change this narrative? Question one can be answered this way; data is to AI what food is to humans, i.e. what we feed our machines matters. Machines work in binaries; 1s and 0s. In this case, what data are we feeding our machines? Datasets could possibly look like {white, male, cis, straight hair, fair skin} which means any deviation away from these descriptions would be read by the machines as entities that are dangerous or that simply do not exist. Black women experience this regularly when they get profiled by airport body scanners for sporting their natural hair in all forms. Yet, this is only on the surface level and for something that could be considered trivial. This then means that everyone who looks different in many other different and deeper ways, is not only left out of any progressive and developing technology, but is also discriminated against by the existing technology. We must not think this is not accidental, it is by design.

The existing technology that seems to “represent” women only embody sexist stereotypes. This can be seen in the invention of female chatbots who all have the persona of a white woman as pointed out by Gretchen Andrew when she spoke of the servitude stereotype behind having female chatbots such as Alexa, Cortana, Siri and (my personal addition) Zuri. They are designed primarily by male developers and engineers and this is why they embody the same misogynistic patterns that can be observed in real life (you know, only there to take your orders and not have an opinion of anything outside of what she’s required to say?) This narrative is being challenged daily by women like Josie Young who are redefining what the design process of building feminist chatbots looks like and how to build ethical AI using feminist principles of the internet.

If you know me, you know this is my one constant mantra; a top to bottom approach to problem solving, an inward going outward cleanse of any system. We cannot pretend to want to change a system by tweaking a few features to appear “woke” or progressive. Instead of developing technology that replicates the old retrogressive narratives, let’s hire female engineers, female data scientists, female developers, female sociologists to assess the impact breaking down these systems would have on the technology we develop and the products we put out for consumers.

But beyond this black and white representation, there needs to be representation of other individuals who have been left out for too long and for whom, negative technology impacts flow even deeper. If we as an example teach a machine that a woman looks a particular type of way {straight haired, amongst other physical features}, we are teaching these machines to again ignore nuances that exist among human beings. We are locking out trans and intersex individuals. We can break such barriers by having increased representation. We have developers, engineers and data scientists who are either differently abled or that identify as trans or are intersex. This is what intentional diversity and inclusion looks like; including people who were deliberately excluded by the existing patriarchal and capitalist systems.

Aside from just having representation in machine learning and the building of technology, it is important to have diversity in the data sets. As an example, at MozFest we had less representation from African women both living in their home countries and in the diaspora. We had even less representation of people living with disabilities. Majority of the attendees spoke English. How then, can our datasets look different? How can our conversation sound different if everyone in the room looks and behaves exactly like us? I know funding everyone to come to such an event is logistically impossible, but we can redirect resources (funds, skills, networks) to those less privileged than us so that they can by themselves learn and build products and experiences that fulfill them better.

However, even as we make positive strides in the diversity, equity and inclusion movement, we must be very aware of what we are including people in. While AI has its positives, it also has its negatives. All these developments have a long way to go but we can start from here. If you missed MozFest 2019, check out their highlights.

My entire existence is intertwined in all I’m trying to do and creating a similar safe space for all. Consciously learning and unlearning. @nyakwakah on Twitter

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade