Making Do and Making Change: Why AI Governance Needs Feminist Bricolage
In 2016, a small group of residents in Pittsburgh, USA, suffered from poor air quality. They didn’t have the tools they needed to identify the sources of air pollution. In partnership with Carnegie Mellon University’s CREATE Lab, they built “Smell Pittsburgh,” a smartphone app that helped the community report foul odours and track patterns of pollution. Using what they had at their disposal, namely their own sense of smell, smartphones, a team of academics, existing air quality monitoring data from government stations, and civil society support, they built out a powerful and creative form of governance. They created a system where smell reports were documented and made accessible online, and sent it out to the local health department. This paved the way for the creation of an AI model to predict upcoming smell events and send push notifications, and for a map that visualizes crowdsourced smell reports with air quality data from monitoring stations. In the 10 years since, this approach has helped the community advocate for better air quality, and informed a nationwide mechanism called Smell MyCity.
In northern Aotearoa (New Zealand), Te Hiku Media, a broadcaster representing five Māori iwi faced a key challenge. They had to teach computers to understand te reo Māori without surrendering control of their data to big tech companies. They had 30 years of archived audio from their iwi radio station, of which 1000 hours were from native speakers. With time, a small team built on community trust and kaumātua (elder) guidance that included one engineer in 2015 built a partnership with a small data science firm called Dragonfly Data Science. Together, they built their own content platform called "Whare Kōrero" (house of speech), instead of opting for a global platform to upload their data after signing over content rights in their terms of service. They ran a crowdsourcing campaign called Kōrero Māori, where over 2,500 people read out over 200,000 phrases in a matter of 10 days, providing 300+ hours of labelled speech data under the Kaitiakitanga license. The team then developed automatic speech recognition models for te reo Māori with 92% accuracy, using NVIDIA's open-source NeMo toolkit. In the process, they outperformed global tech giants and created a whole transcription service called “Kaituhi.” Their models are being used by over 20 Māori radio stations, and inspired a whole range of similar projects by Native Hawaiians and Mohawk peoples. Their entire initiative also created powerful pathways for Māori data scientists and highly-skilled tech jobs in their communities.
Both groups started from where they were and with what they had, to solve problems specific to them. They combined lived experience with expertise, centred community knowledge and leaned on technical knowledge for support, and set up governance capacity with the tools they had at their disposal instead of waiting for a top-down regulatory framework. Both groups showcased collective power that prioritised data transparency, community rights, and prevented big tech from extracting their own data and selling them a “solution” they did not participate in creating. They questioned systemic and structural violence, subverted the trope that the master’s tools offer the sole path for engagement, and defined their own paths while protecting their interests. Both groups practiced feminist bricolage.
Feminist bricolage: Collecting, building, weaving, creating
Feminist bricolage is the convergence of bricolage as a research methodology and feminist thinking. Bricolage describes a form of work that assembles knowledge and solutions from materials and resources that are readily available, rejecting and subverting the myth of the perfectly designed, universal framework.
The term “bricolage” came from the work of French anthropologist Claude Lévi-Strauss, who distinguished between the bricoleur (the tinkerer who works creatively with available materials) and the engineer (who designs systems from scratch with purpose-built components). Through the work of Joe Kincheloe (2005), bricolage became a critical research methodology that relies on multiple data-gathering strategies and uses diverse theoretical positions to situate research purposes and meanings. This form of bricolage focuses on interconnected relationships rather than on individual units functioning in silos.
Handforth and Taylor (2016) considered bricolage an experimental feminist practice of collaborative writing that both, weaves together research narratives and challenges normalized academic practices. It rejects the god trick of seeing everything from nowhere, as Donna Haraway (1988) put it, meaning that it subverts universalism and disembodied objectivity. Instead, it makes the case for pluriversal ideas, built on a commitment to actioning intersectionality and situating multiple perspectives. However, it is not just the mere art of “making do” with available resources. Feminist bricolage recognizes that in working with all the tools available and at hand, a community is presenting a knowledge or approach or framework that is meaningfully situated and assembled from available materials. It recognizes that these outcomes are just as rigorous and valuable as outcomes produced through well-resourced and linear designs.
Governing AI through Feminist Bricolage
The emergence of Artificial Intelligence has posed more policy, legal, and regulatory challenges than have been successfully met. From outright harm that is clearly in need of being checked, to ethical dilemmas, there are many questions that a limited, universal legal or policy framework can meaningfully address. The dual use nature of AI as a tool, as well as the burden it imposes on marginalised people both in its production and use, make a compelling case for any attempt at AI governance to prioritize intersectionality, decoloniality, and multidimensionality.
Rigid, universal frameworks ignore contextual realities and peddle colonial approaches that homogenize and atomize human beings into unidimensional datapoints. Add to this is regulatory arbitrage (the practice of avoiding heavy legal consequences in one region by shifting to a region where there is little to no regulation), which continues to enable colonial digital infrastructure to keep its extractive approaches going.
In practice, feminist bricolage can look like a community combining algorithmic auditing tools, traditional cultural protocols, and lived experience to evaluate a facial recognition system. By using diverse and seemingly incompatible knowledge systems together, they prioritize a feminist approach of centring marginalised knowledge while rejecting technocratic monopoly. Institutionally, feminist bricolage could look like practitioners adapting laws in collaboration with civil society pursuing several levels of organising and cultural boycotts to challenge exploitative AI training data practices. By assembling governance from disparate existing resources, these efforts rely on what is available instead of waiting for the perfect new top-down regulation to make a change. In doing so, the endeavour recognizes power inequities and works strategically within a state of constraint.
Looking to the future
On the one hand, there is a race to build AI and its related hardware. On the other hand, there is a race to regulate AI. Between both these races, a large number of communities that lack formal regulatory power are confronted by the pressing need to govern AI that affects their lives and livelihoods. Mainstream approaches to regulation follow traditional approaches of law and policymaking, and produce mechanisms that operate in a top-down, engineered fashion. It is meant to apply to an individual regardless of their context. Unlike this, a feminist bricolage approach is bottom up. Rather than serving as perfect universal frameworks, they offer situated practices that are adaptive. It holds plenty of promise for the future of AI.
However, it is important not to reduce feminist bricolage to “making do,” and romanticizing that approach when structural change is foundational to meaning change. Adaptation can easily pave the way for accommodation and co-optation by systemic and structural factors, meaning that those in power can avoid accountability and continue leaving communities with frugal resources. Acknowledging these tensions inherent in operationalizing feminist bricolage is productive, as it keeps us open, honest, and committed to pursuing a feminist vision for governance without avoiding pragmatism. The goal shouldn’t be governing with the perfect law, but governing with community.
References:
Haraway, D. (1988). Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective. Feminist Studies, 14(3), 575-599.
Handforth, R., & Taylor, C.A. (2016). Doing academic writing differently: a feminist bricolage. Gender and Education, 28(5), 627-643.
Kincheloe, J.L. (2005). On to the Next Level: Continuing the Conceptualization of the Bricolage. Qualitative Inquiry, 11(3), 323-350.
Hsu, Y. C., Cross, J., Dille, P., Tasota, M., Dias, B., Sargent, R., Huang, T. H., & Nourbakhsh, I. (2020). Smell Pittsburgh: Engaging community citizen science for air quality. *ACM Transactions on Interactive Intelligent Systems*, 10(4).
Mahelona, K. (featured in NVIDIA Blog, 2024). "Māori Speech AI Model Helps Preserve and Promote New Zealand Indigenous Language."
