When Good Intentions Meet Reality
I made similar mistakes in India. I believed expanding internet access and social media would strengthen democratic participation and civil society. More connectivity meant more information flow, which would help communities organize and make better decisions.
Instead, I watched the same platforms enable sophisticated radicalization networks—coordinated campaigns weaponizing connectivity to mobilize violence and accelerate polarization. The infrastructure I'd championed as democratizing was simultaneously being used to tear communities apart.
The pattern was clear: technologies aren't neutral tools that work the same way everywhere. They enter existing power structures and get shaped by them. The question isn't whether something works in theory, but whose interests it serves in practice and what happens beyond what you intended.
The Real Problem Wasn't Implementation
These weren't implementation failures that better project management could fix. The problem was deeper: arriving with predefined solutions to problems defined through external frameworks, then measuring success through metrics that reflected external priorities.
This is the default mode for most international development, diplomatic engagement, and tech-for-good work. The people closest to problems rarely get to define what those problems are, let alone design the solutions.
My current work in AI governance grew from wrestling with this dynamic. When powerful actors deploy algorithmic systems in vulnerable communities—for predictive policing, resource allocation, content moderation—they're not just introducing new tools. They're making decisions about how society should be organized, whose knowledge matters, and which outcomes deserve optimization. And those decisions get encoded into systems that are hard to challenge or change.
Read on to also find out what Marine calls the "Jobs that don't have names yet" where she highlights roles that would be relevant for the future of AI governance and policy in this sector.
So What Does Working Differently Actually Look Like?
Peace technology isn't about applying tech solutions to conflict. It's about recognizing that technology always reconfigures power relationships, and designing accordingly.
For example, the voice-based crisis reporting system we're building in Malawi. The hard part isn't speech recognition in Chichewa—it's figuring out: Who owns the data communities generate about emergencies? Who can access it? Who profits from it? Who has the power to shut it down or repurpose it?
These aren't secondary questions to address after the technology works. They determine whether the system serves communities or becomes another way to extract information from them.
Models like data commons approaches flip the default model. Instead of data about communities becoming corporate or government property, communities retain collective ownership and control. The concept is straightforward—the challenge is building governance structures that make it work when you're dealing with populations that existing systems systematically exclude.
Participatory AI can get reduced to consultation—asking communities what they think about systems someone else designed. Real participation means communities have actual power: the authority to say no to deployment, demand transparency about how systems work, and modify or terminate them based on what they observe.
These aren't just nicer ways to work. They're responses to recognizing that when external actors control technological decisions—however well-intentioned—we reproduce patterns of extraction and control.
That Feeling of Not Quite Belonging Anywhere
If you're reading this feeling professionally illegible—too technical for policy work, too political for technical roles, too critical for implementation, too practical for research—you might be positioned to do something valuable.
The sector's real challenges sit in the spaces between professional specializations. We need people who understand why technical solutions fail when they ignore power. Who can translate between community knowledge and institutional frameworks. Who've worked inside enough different systems to see their blind spots.
This isn't about being a generalist. It's about learning to see how different types of expertise create blind spots, and working in the uncomfortable spaces where the most important dynamics actually happen.
Jobs That Don't Have Names Yet
These don't have clear job descriptions yet:
Conflict-sensitive technology governance: Understanding how digital systems reshape power in fragile places, and building accountability before deployment rather than after harm.
Community data sovereignty: Designing frameworks that enable collective ownership and control over information, particularly for populations excluded from data governance decisions.
Participatory accountability: Creating mechanisms where affected communities have real power over algorithmic systems—not token input, but authority to investigate, contest, and override automated decisions.
Cross-sector translation: Explaining to technologists why community organizing matters, to policymakers why technical architecture is political, to funders why participatory processes produce better outcomes than rapid deployment.
These roles need people who've learned to question their own frameworks—ideally through the uncomfortable experience of having them challenged by people facing consequences of your work.
What I Wish I'd Developed Earlier
Learn to recognize your blind spots: The journalist in Bangui understood information politics my frameworks missed. That's not about self-deprecation—it's recognizing that different positions reveal different aspects of problems. Your expertise is valuable and limited.
Develop literacy across domains: You don't need expertise in everything, but understand enough about technology development, policy, institutional power, and community organizing to see where they intersect and where your knowledge becomes a limitation.
Get comfortable being uncomfortable: The ability to hold contradictions—the imperative to act, the recognition that action causes harm, the necessity of working within flawed systems while pushing to change them.
Make power analysis routine: Always ask: Who benefits? Who bears risks? Who decides? Whose knowledge counts? These aren't radical questions—they're basic due diligence most work skips.
Document what didn't work: Your most valuable asset is often understanding why something failed. The CAR and India experiences shaped my research more than successes. Learning to discuss failure honestly makes you credible to communities who've experienced harmful interventions.
Where This All Leads
AI systems are being deployed in vulnerable communities by actors with minimal accountability. Digital infrastructure is reshaping power in ways governance frameworks can't capture. The gap between where technology gets developed and where its consequences land keeps widening.
We don't need better experts operating comfortably in existing categories. We need people who can work across the boundaries of expertise, who've learned through experience why external solutions often fail, and who can build alternatives that take power seriously.
If your career feels hard to explain, if you keep hitting resistance because you don't fit hiring categories cleanly, you might be developing exactly what we need. The roles that matter are the ones we're still figuring out how to name—and they require people uncomfortable enough to question the categories themselves.
About the writer
Marine Ragnet is AI Lead at NYU's Peace Research and Education Program and Senior Fellow at the Portulans Institute, focusing on AI governance and community-centred technology development in fragile contexts.
This is a MyImpactStory, a series by Impactpool where your Impact Story is what you do to make the world better. Share yours to gain invaluable exposure across the sector and inspire others committed to social impact. The most relevant stories will be shared on Impactpool and in our social media channels.