Shifting the power dynamics of data in the social sector

In the weeks following the ICT4D Conference in Nairobi this May 2026, I have been reflecting on how my relationship with Technology and Data has evolved throughout the years, particularly at this turning point marked by the rising power of AI.

My name is Pallas, and I am a Domain Researcher and Scientific Writer at Avandar Labs. Since this is the launch of my blog posting activity with Avandar (the first of many to come), I want to introduce myself with some background context. I am a scientific researcher by training, with a PhD in Biological Sciences and Public Health from Harvard University. My first experience of the power of technology was through the light microscope. As an undergraduate, I discovered a fascination with how microscopes enabled us to see what the bare human eye could not: the intricate wonders of the cell, the basic building block of life. Each cell contained its own world of complex, dynamic, interdependent machinery, one only visible to us because of technology. Advances in biological imaging have continually transformed our knowledge of subcellular networks, each generation of technology revealing a deeper understanding of what we are made of. It was during my PhD that I first encountered the transformative potential of machine and deep learning as a force actively reshaping how science was done. In microscopy, the shift in how data was handled and analyzed was seismic. We moved from the constraints of human interpretation alone– a researcher assigning qualitative values to images– to algorithms that could be trained to classify and quantify data at a scale no individual eye could match. The gain was profound: greater analytical depth, expanded throughput, and partial liberation from human bias.

With that power, however, came a new set of questions. New constraints, new guardrails, new safeguards, new and often unanticipated forms of bias to reckon with. The tools demanded as much scrutiny as they enabled discovery.

Now, four months into my work at Avandar, I find myself returning to parallel questions, only this time, the scale has shifted. What I once witnessed in the microcosm of a research lab is now legible in the macrocosm of society. The tensions that defined how I approached AI in biological imaging — who controls the model, what assumptions are encoded within it, what safeguards are necessary — are the same tensions shaping how AI is transforming the social sector.

The power dynamics of data

At ICT4D, this tension was unmistakable. The exponential pace of AI development was not merely a backdrop to the conference, it was the animating question running through it. I spent the week immersed in conversations with people who have devoted years to building digital tools for communities, all of them now working to understand what the rapidly shifting landscape of data and AI means for their work. Practitioners examined the opportunities and the pitfalls of AI across sustainable development, humanitarian response, climate change, global health, education, and financial inclusion. My days were filled with talks from thought leaders, panels convening experts across sectors, demonstrations of digital tools in action, and workshops with practitioners testing, iterating, and refining solutions in the field. As a researcher, I luxuriated in the opportunity to observe, listen, and learn from the community around me. Across multiple sessions on responsible AI, data privacy, data use, interoperability, monitoring & evaluation, I noticed that conversations often eluded to the underlying power dynamics of our data ecosystems. Who controls data; who is it designed to serve; who does the data fail to represent; who bears the cost when data is mishandled? It began to dawn on me how deeply these power structures weave into the entire lifecycle of data, and ultimately, the decisions we make, the tools we build.

At Avandar, our founding mission is to build the best data tools for the social sector from the bottom up. This means we aim to provide software that is affordable and open-source for the smallest, most resource-constrained teams without ever sacrificing quality. These teams, often the local actors at the frontline doing the most life-changing work for their communities, are dispossessed of power by a data ecosystem never designed to serve them. In one of the keynote talks at ICT4D, I was introduced to the term “data colonialism” by Jim Fruchterman of Tech Matters. Data colonialism is a concept that connects historical colonial resource extraction to what is happening now with data.¹ He described the times he saw foreign organizations come, collect data with local help, and then leave when they’re done or when the funding runs out. The data goes with them. Local actors have no control over the data they helped collect, no input in how their data is used, no access to the publications that inform decisions about their own communities. We see this colonialist model of data collection systematically extracting value from people in already marginalized communities, rendering those most at risk when data is misused with the least power to do anything about it.

It doesn’t help when tech solutions for data management also perpetuate this power dynamic, orienting itself toward those who already hold the most capital. The result is a digital divide that compounds existing inequities, where the organizations closest to the work are handed the weakest tools.

Shifting power back to the people

To serve these communities is, by necessity, to take a position on where power ought to sit. It is not enough to build the best tools and consider the work done. If data power dynamics determine who benefits from data and who bears its risks, then building responsibly means continually interrogating those structures, and asking with every design decision, whether we are reinforcing the concentration of data power or working to redistribute it back to the people at the center of the work.

At the conference, it was exciting to see how ethical critiques of power were beginning to reshape the social sector's Responsible Data and AI standards.

Decolonizing data

Jim Fruchterman pushed for sector-wide reform to “decolonize data." He shared Tech Matters’ ongoing initiative to develop a Better Deal for Data (BD4D), a practical governance framework for the ownership, control, privacy, sharing, consent, and monetization of data. It is a set of commitments designed to shift power back to the communities where data is collected.

Listening to our community

Power can also be given back to communities through practices as simple as listening to them, and making changes based on what we learn. I was particularly moved by one of the panel discussions on “Tech-enabled community listening.” Some of the questions I had been pondering in abstraction were echoed in the discussions in the room. The speakers presented tools their organizations were implementing to reach marginalized voices and how AI was used to analyze this data at scale. But where the tools succeeded, were also followed by grounded critiques on where the tools fail. Lauren Ropp, from the Institute for Development Impact (I4DI), highlighted that tech struggles with capturing human context. Chatbots will not detect when a woman lowers her voice when her husband enters the room. Voice surveys miss when groups laugh before answering, or when silence is the answer. AI misses intonation, hesitation, nuance in human understanding. It under performs on low-resource languages, local dialects, and women’s voices. AI models lack representation of humanity’s diversity, and are biased towards the loudest voices who already have the most power.

Linda Raftree of MERL-Tech Initiative also pointed out that much of the Responsible Data work we now hear on the main stage at ICT4D is not new. It’s built on years of thinking by feminist organizations and African women who have been asking these questions long before they became conference themes. Most of us just weren’t listening to what they had to say.

Lessons, questions, and what comes next

Note

To recap the themes that have come into focus:

1. Power structures weave into the entire lifecycle of data. It informs the decisions we make, and the tools we build. The people doing life-changing work for their communities, often have the least power. By continually interrogating the existing structures in which we operate, asking with every design decision who benefits from data and who bears its risks, only then can we truly build the best tools to serve them.

2. Power must be redistributed back to the people at the center of the work. The social sector's Responsible Data and AI standards are evolving. Emerging data governance frameworks, like BD4D, are some of the first operational standards we can collectively adopt to shift the power dynamics of the social sector.

3. AI dominates social sector tech developments across the board, but comes with a new set of questions to reckon with. AI is powerful for scale, but fails in human understanding. It misses tone, hesitation, group dynamics, and silence. AI models lack representation of humanity’s diversity, and carry their own intrinsic biases.

Questions that remain

Every lesson about power, technology, and AI has, in turn, revealed a longer list of questions. How do power dynamics shape bias in the social sector, and how do we identify and mitigate it? If data behind most AI models lacks diversity, what assumptions and biases are encoded in the models themselves? And if each new layer of technology inherits and compounds biases of the last, how do we address them at the root? What are the responsible practices for using AI on data of vulnerable populations? Can we build new AI models for the specific needs of the social sector, with a stronger commitment to privacy? These questions guide my ongoing work at Avandar.

What this means for my work at Avandar

I am not afraid to ask the uncomfortable questions, as it is in the places we have not dared to go that our work truly begins.

As a researcher and a writer, I am devoted to continuous questioning of what we think we know. I am not afraid to ask the uncomfortable questions, as it is in the places we have not dared to go that our work truly begins. My work at Avandar sits between two practices I care deeply about: continuous learning through rigorous research, and helping knowledge reach the people who need it. I am excited to integrate these ongoing questions in the work I carry forward, in hopes of helping the social sector feel more resourced.

Footnotes

1. Ramanathan N, Fruchterman J, Fowler A, Carotti-Sha G. Decolonize data. Stanford Social Innovation Review. 2022;20(2):59–60. https://doi.org/10.48558/FSYC-2N32 ↩