All around us, the pace of technology has accelerated, and the rapid development of AI is no exception. In just a few weeks, we’ve moved from AI conversations centered around how realistic images and generation model images appear to ones compounded by users experimenting with OpenClaw. In policy spaces, seemingly “responsible” AI companies are now face-to-face with federal government demands. And in environmental circles, the AI conversation has largely become flattened to questions of energy consumption and the physical footprint of data centers (though important, it doesn’t tell the full story).
At the same time that these larger, highly visible debates unfold, AI is creeping into the everyday spaces where OEDP works. We don’t subscribe to the inevitability narrative, but we do recognize that AI is already, and will continue to be, a disruptive force in open data, participatory science, and environmental decision-making. Regardless of how one feels about it, AI is changing the way environmental knowledge is produced, shared, and acted upon.
What seems absent from many of these debates are the places where OEDP has been working for years, such as data stewardship grounded in participatory design to shape the means and methods by which communities understand their environments and themselves.
The loudest conversations praise speed or warn about scale: faster modeling, bigger datasets, more time saved, larger infrastructure. Meanwhile, we’re pausing to look closely and saying, “Hm…”
From this pause, we’re beginning to develop a blueprint that aligns with OEDP’s values that orients how we approach AI:
Participation by design. Participatory science has shown that good science is social: consent and context matter, and the right to question is essential. To avoid participation-washing, AI tools for environmental research must treat these as requirements—not afterthoughts. Communities should be able to see what a model was trained on, contest its behavior, and reshape it for local reality: we should be treated as something more than support agents in the deployment and use of AI. When people contribute data, they should help determine how models learn from it.
Open stewardship for training data. Open data principles must evolve: access alone is no longer enough. In the era of models, “open” without stewardship risks extraction. Licensing, provenance, and governance can enable shared benefit while protecting local knowledge, language, and cultural context. Environmental data should be stewarded, rather than owned, so it doesn’t become the basis for systems that communities can’t examine or govern.
Small + local AI is a public option. Not all value requires scale, and scale can often erase the local context that gives knowledge its meaning. Much of what communities need can run on a laptop or low-cost device: detecting turbidity in a local stream, transcribing oral histories in a minoritized language, or retrieving documents from a civic archive offline. In these instances, AI can actually be helpful for scientific discovery or for preserving and accessing local knowledge. Smaller, local models are easier to audit, cheaper to maintain, easier to contextualize, and more accountable to the places they serve. They shift AI from a rented service (think shoes at a bowling alley) to a stewarded asset (such as your grandmother’s cast-iron pan).
OEDP’s work lives in the space between community monitoring and decision-making, where data becomes meaning, and meaning becomes action. In that space, AI is not just a tool—it is a governance choice. It can recenter power or redistribute it. It can multiply knowledge or overwrite it.
Emerging technologies are part of the systems in which we operate. So we’re beginning to ask practical questions that help clarify our collective responsibilities, and the kinds of resources we may need to build. Questions like:
What does AI miss when it summarizes lived experience?
How does refusal function as a protective practice in participatory science?
What kinds of patterns can humans perceive or create that machines cannot; and why does that matter for environmental knowledge?
Over the next few months, we—Shannon Dosemagen, Emelia Williams, and Cathy Richards—will share a series of posts offering reflections on AI in environmental and participatory contexts. They won’t be polished or perfected. But they will be our fieldnotes from attentively wandering through the AI landscape. Sometimes the process has meant staring at a [physical] cloud to understand perspective; sometimes it’s meant digging into the mechanics of Claude. Each week represents an engagement with the ideas of other people thinking, working, and living with these questions, too. You’ll see them referenced in each of our posts.
This isn’t a formal research program (though we hope others take up these questions in deeper ways). Instead, our team is using these questions as prompts to think about our own work: ways to surface where AI and its owners will exert influence, where communities may need preparation, and where design choices shape power. You’ll hear ongoing work of ours, such as the Digital Toolkit for Collaborative Environmental Research (or DIGITCORE) and data stewardship, referenced frequently.
The goal isn’t perfection, or even answers. It’s to begin poking at the questions that are being skimmed over in the rush, and to slow down long enough to see what remains when the fog clears.
Here’s an overview of the questions we’re asking and responding to (we’ll update this each week with links to the blogs and updates on what each person is writing about).
#1. What problem is AI being offered as a solution to, and for whom?
This set of blogs was released April 1-3.
The first week, we reflect on what problems AI is being offered as a solution to, and for whom, in the context of participatory science and environmental research. Where do AI narratives misalign with the actual needs and constraints of EJ and participatory science?
In “Who gets to define participation in the age of AI?”, Shannon writes about on protecting human judgement through participation rather than automation.
In “A sparkle, not a silver bullet”, Emelia sits with the ubiquity of ✨ and the dominant narratives we can push back on.
In “Data is not enough”, Cathy focuses on the opportunity cost of the “lack of time to process information” refrain.
#2. When does AI help, and when does it increase risk?
This set of blogs was released April 8-10.
In the second installment of OEDP’s Fieldnotes on AI, we look at the specific situations where AI could reduce burden, increase access, or improve decision-making without increasing risk or dependency? And also where AI actively degrades judgment, care, and accountability.
In “Are your tools ultralight?”, Emelia considers what it looks like to carefully inventory AI tools as an act of stewardship.
In “The Things We Lose Along the Way”, Cathy wonders about the thin line between reducing real burden and overly simplifying the world we live in.
In “Prediction Without Political Will”, Shannon looks at how AI can help surface environmental risks/patterns, but can risk deepening harm without corresponding political will and resources.
#3. What is a point of view—and why can’t AI have one?
This set of blogs was released April 15-17.
During the third week, we spend time grounding ourselves in perspective. What is a “point of view” in environmental research and EJ work? How does a point of view emerge from place, exposure, and lived environmental experience?
In “AI is… basic”, Cathy wonders how we maintain our uniqueness when using a technology meant to find the average.
In “POV is a position, not a preference”, Shannon writes about how environmental justice shows that point of view is not a preference but a position shaped by lived exposure to harm and power in ways AI systems cannot replicate.
In “Weird sight against epistemic dominance”, Emelia ponders about how human embodiment, interaction, and knowledge-creation cannot be replicated by tools that do not have a place, stake in the game, or a way of being that experiences the natural world.
#4. What kinds of patterns can humans create that machines cannot?
This set of blogs was released April 29-May 1.
We build on our previous reflections about POV to look at patterns that humans can create that machines can’t, and why this matters for environmental knowledge. What habits help people respond to environmental uncertainty without outsourcing judgment to AI?
In “The work of noticing patterns”, Shannon writes about how human pattern-making is rooted in memory and place, allowing us to both notice and act.
In “Humans are the loop”, Emelia considers how humans make patterns in precarity, with survival in mind.
In “Mimicry is the greatest form of flattery”, Cathy writes on humor and grief, two instances where not following a pattern is fundamental to understanding it.
#5. Prototyping without capture: how can we use AI to clarify questions?
This set of blogs was released May 6-8.
In our second to last installment, we consider how prototyping can sharpen inquiry—or capture it. How can fast, imperfect prototypes clarify thinking rather than replace it? How can prototyping clarify environmental questions without locking communities into technical pathways?
In “Prototypes, purpose, and not driving off the road”, Emelia elevates the idea of putting up guardrails while you experiment (quickly! carefully! purposefully!).
In “From failure you learn—from success, not so much”, Cathy writes about leveraging AI early on in the process in order to learn from failure as quickly as possible.
In “Against premature answers: what prototyping teaches us”, Shannon thinks about her work with open science hardware in relation to AI, and how hands-on and place-based prototyping shows how meaningful inquiry emerges through imaginative engagement.
#6. Refusal, stewardship, and obligation in the model era
In “Who has the keys?”, Cathy writes about the power researchers hold and the many ways we can responsibly steward valuable things.
In “When participation becomes raw material”, Shannon writes about how stewardship can no longer end at collection or sharing as data becomes training material for AI.
In “To what end?”, Emelia writes about the knowns of data stewardship and how it plays out or expands when AI enters the equation.
It's nice to see others are writing about this. I got one magazine article in the works about how to make data centers eco-friendly...and a newsletter about the difference between streaming services data centers vs AI data centers.......I look forward to reading your newsletter.