This post is part of a series titled “OEDP’s Fieldnotes on AI”, where we offer reflections about AI in environmental and participatory contexts. In our second-to-last installment, this week’s reflections center on the question:
How can prototyping sharpen inquiry—or capture it?
We consider how fast, imperfect prototypes can clarify thinking (rather than replace it) and environmental questions (without locking communities into technical pathways).
A thread that runs through my work has been open physical tools for environmental monitoring, part of open science hardware (OScH). When experimenting and prototyping with physical objects, the results can be seemingly quirky, but they are actually a perfect example of what we should be aiming for in digital spaces: things that are responsive to the environments in which they’ll be used.
I spent a decade immersed in this kind of daily prototyping while working alongside the Public Lab community. Public Lab’s aerial mapping rig is a good example. To capture images once the camera was high up in the air, you basically had two options: use a hacked SD card that allowed for continuous shots, or find something to depress the shutter button. Once, on a trip into the Gulf of Mexico, I failed to bring my usual “depressors”, which were the head of a Q-tip or a small stone. They were easy-to-find objects, but less easy to imagine as components that enable the capture of hundreds of images from thousands of feet up. It was a critical miss on my part. Looking around the boat, though, the solution presented itself: a tiny bone from a fish, caught on a previous voyage. That spirit of prototyping persisted.
In the prototyping I’ve done, ideas are bound to a specific time or place, built with the materials at hand, shared digitally, and then repurposed in other contexts with their own questions and constraints. The point is never perfection: it is about learning along the way, asking better questions, noticing what use (or a thing or an idea) reveals, and allowing the object itself to reshape how we understand the problem. In prototypes, we learn and begin to see our engagement with the world differently.
There is something important here for our approach to AI. In physical prototyping, imagination is not abstract. It is grounded and emerges from constraint, from proximity to problems and questions, and from the need to make something work with what’s available. It is iterative, revisable, and accountable to the conditions in which it operates. The prototype does not close the question; it opens it up further. In contrast, many AI systems arrive already formed by presenting outputs that feel complete, even when they are built on partial data or unexamined assumptions. They compress the space in which imagination might otherwise operate. Instead of asking, what could this be?, we are asked to evaluate what has already been generated. This is where the risk of capture begins.
When a prototype is treated as a solution, meaning a model output is taken as an answer rather than a prompt, it can narrow the range of possible interpretations before communities have had the chance to interpret, contest, or redirect what is being proposed. And yet, prototyping in AI could look very different if we looked more closely at other examples of how prototyping unfolds in the physical world. There are instructive examples from earlier “emerging technology” moments, including open science hardware, the early Internet of Things, and even personal computing kits of the 1970s and 80s, where devices were often situated, iterative, and shaped through use. And in the AI space, embodied AI work such as CurrentAI’s “Suno Sutra” points toward a vision for locally-run, shared AI devices that are less dependent on centralized control.
Rather than building toward scale or optimization, we might build toward clarity. As we discussed in the intro to our OEDP fieldnote series, we’re interested in small, local models, and collaborative workflows at a localized scale. And through experience what can help us get there are events like clinic-style engagements where the goal is not to produce a definitive output, but to surface assumptions, test interpretations, and make space for disagreement. We want our spaces to look at things that are perhaps temporary, and are designed to be revised, or even discarded, but always documented.
This is where imagination matters. Though you might be surprised to find imagination as a key thread in a book largely about bureaucracy, David Graeber’s writing on this in The Utopia of Rules is something I’ve been returning to. He points to older understandings of imagination as “a kind of middle ground, a zone of passage connecting material reality and the rational soul.” That framing resonates well with prototyping. A prototype sits in that middle ground, being not fully real, and not fully abstract. It is a way of thinking with materials, and of moving between idea and thought to the creation of what could be.
What I worry about is how little room we are leaving for that middle ground even in the techo-scientific landscape that we interact with, which wasn’t as polluted with AI debates just a few years earlier. In many institutional contexts, the pressure to move quickly from idea to implementation leaves little space for this kind of work. The messy stages, where questions are still open, and imagination has not yet been maneuvered into a solution, are treated as inefficiencies rather than as necessary preconditions for understanding what is at stake and what is possible. Without that space, we lose the ability to ask different questions and notice what doesn’t fit, or imagine alternatives to what’s already encoded in the systems we have. In the early years of OEDP, this was always a key part of our work, alongside the very pragmatic work that data can funnel us into. While we live in existing systems, we still have to question why, and imagine other ways.
If AI is to support environmental knowledge and decision-making, it needs to operate within that middle ground rather than collapse it, functioning less as an answer machine and more as a companion in environmental inquiry. It has to be something that helps us test, reflect, and revise without shutting down the interpretive work that humans need to drive. Prototyping can help to protect that space by keeping part of the work lightweight, and reminding us that the goal is not to get to the answer faster, but to understand more fully what the question actually is.
