This post is part of a series titled “OEDP’s Fieldnotes on AI”, where we offer reflections about AI in environmental and participatory contexts. This week’s reflections center on the question:
What problem is AI being offered as a solution to, and for whom?
We identify the pressures communities feel right now that AI will supposedly solve and delve into where dominant AI narratives misalign with the actual needs and constraints of EJ and participatory science (and how they narrow the questions EJ groups are “allowed” to ask).
I came of professional age during the 2010s boom of interest in civic science, media, and technology. It was a moment that leaned heavily toward participation, broader connectivity, and the idea that communities—no longer just local or proximate—could operate seamlessly across the globe, dancing in concert as they solved the world’s problems. All these years on, the problem, I’ve come to believe, is not connectivity itself, but the assumption that scale and seamlessness are prerequisites for meaningful participation.
Participation is not just about scale. It is about responsibility. It’s about engagement and understanding the problem. It’s about being better allies, listeners, and collaborators.
Science and open technologies in the environmental space, at their best, are tools for helping us understand complex issues. They are not (or should not be) simply data collection devices. Yet participatory science has often asked us to foreground the research first: the tools as primary, and the datasets as the source of impact. In doing so, technology can slip in front of the actual issue.
I see a familiar pattern emerging with AI. It’s increasingly suggested as useful for setting agendas, anticipating questions, and even designing research methodologies. As Dawn Nafus argues in SparkleTech, we do have the opportunity to shape where and how AI might be useful, but only if we also put clear “boxes” around where it should not touch. There are places that must remain with human judgment where point of view matters, local knowledge is essential, and where decisions require finesse rather than mechanized compilation and rigidity. Scholars and advocates have similarly critiqued dominant AI narratives that prioritize technical objectivity over human rights and justice, calling on us to “move beyond AI safety narratives” that too often center on efficiency, inevitability, or techno-solutionism, rather than on lived realities.
Years ago, I wrote an article for Science for the People that tried to make sense of the jumbled terrain of citizen, community, and civic science by wrapping it in a neat, bow-tied framing called “participatory science.” It was meant to be lightweight, a way to draw attention to shared tensions and aspirations. In hindsight, that framing risks making the field sound like an elective expansion of science, rather than what it often is: an unfortunate requirement for communities forced to generate evidence just to have their harms acknowledged at all. It subtly reinforces the idea that communities are being given the opportunity to participate.
That history matters when we think about AI entering participatory science. We are already starting from a misaligned place. As with earlier, top-down models of “citizen science,” the danger is not simply that AI might be misused, but that AI begins to define what questions can and should even be asked.
What happens when AI overlooks or flattens local knowledge and lived experience? What happens when AI shapes which questions are deemed legible or worth pursuing? When AI enters the picture, nuance can disappear as problems become universalized and framed as inevitable—stripped of history, and power. Many of the injustices EJ work grapples with are structural and ongoing; they cannot be automated away without obscuring their root causes. Complicating matters further, AI is already embedded in the infrastructures we rely on. Even if EJ practitioners are cautious, opting out is rarely a viable option; avoiding it is difficult and often requires more time, energy, and resources than participation itself.
AI is often offered as a solution to capacity gaps, urgency, and governance complexity. But environmental justice and participatory science are not primarily problems of efficiency. They are problems of power, accountability, and trust.
I live in New Orleans. Earlier this winter, during Mardi Gras season, I was reminded how seductive sparkle can be. I love glitter as much as the next person. I also recognize how easily it lets the eye linger on the surface, drawing attention away from what’s underneath. I’ve felt the pull myself: how easy it would be to synthesize long-form notes into a tidy summary, or to outsource the sense-making after a long day of facilitation. But what do I lose when I do that? I lose contextualization, my ability to insert history, to connect the dots, to feel where something in the room shifted.
Likewise, what does justice become when the person reviewing public comments uses mechanized interpretation? What happens when “good” technical writing leads AI systems to prioritize one comment over another, amplifying some voices while muting others (or when AI simply sways a decision through an overwhelming volume of submissions)? When AI enters participatory work, what does participation get redefined as?
In environmental justice and participatory science, the problem AI is often offered as a solution to is not injustice itself, but the pressure to move faster, to handle more data, to manage complexity, to compensate for shrinking public infrastructure and limited capacity. These pressures are real. But speed is not accountability, and scale is not solidarity. In many cases, friction is not a flaw in the system, but a feature of democratic process.
This does not mean AI is irrelevant to our work. But at OEDP, our role is not to deploy AI tools into participatory processes. It is to shape the infrastructures and governance models that determine how such tools enter those processes in the first place. Through work like DIGITCORE and our focus on data stewardship, we aim to make visible the differences between human deliberation and AI-mediated synthesis: which concerns emerge through lived experience, where trust shifts the direction of inquiry, and how shared values create clarity that no dataset alone can produce.
By articulating those differences, we can define where AI might assist us, and where it should not operate at all. We can translate participatory science and environmental justice principles into design requirements and governance rules for AI systems themselves. I’m referring to this as Participation-First AI. It’s not an endorsement of use or a new field label. It is a boundary condition ensuring that consent, framing power, interpretive authority, and accountability remain anchored in communities. For AI to have a role, it must enter as infrastructure that is stewarded, bounded, and accountable, not as a substitute for the relational work justice demands. If participation becomes data input, we will have lost the point.
