Energizing an Alternative Ecosocial Avenue for AI Applications
Signal & Noise
Imagine a view through a telescope. Backyard, nearby hilltop, observatory, deep space: whatever fits your personal experience. You see the dome of the galaxy, a field of stars. The individual lights gleam brilliantly against the void-black backdrop. Sometimes patterns emerge: Orion, Aries, The Ursas, constellations drawn from projected lines through negative space. The finer your telescope, the greater fidelity you get in the starfield: more stars emerge from further away, changing the constellations you see, the patterns you identify, the meanings you draw. The universe is large, complex, somewhat ineffable but not so much so that we lose hope of understanding it.
Now, as we look through our telescopes, I want to imagine one more thing: that each of those stars is the solution or answer to a specific environmental problem or question. The constellations are the interwoven solution sets we need to halt biodiversity loss, reverse climate change, clean our water and air, and reset our relationship with wildland fire. The telescope, of course, is the tool we use to find those solutions and identify those constellations. For our purposes today: that telescope is the field of machine learning, data science, artificial intelligence – call it what you will.
Environmental scientists and practitioners have been applying algorithmic tools to interpret and act on environmental data for decades. Using their data scientific telescopes, they peer into complex troves of environmental observations, draw lines amongst them, and pull out critical insights that inform their actions. But across those decades they’ve largely been working within a deeply fragmented field. For a variety of structural and technical reasons, the use of machine and deep learning to accelerate environmental progress was constrained by knowledge of the capabilities and investment in their resourcing, experimentation, and development. In effect, the starfield was largely empty with handfuls of shining pinpricks where there could have been marvelous constellations of solutions coming together.
Then, in November of 2022, ChatGPT 3.5 was released to the public, and the collective understanding of - and imagination for - the capabilities of algorithmic tools blossomed. Suddenly, artificial intelligence - a term that had fallen out of favor among serious researchers for its science-fictive tone - was being explored as a solution to everything. Every sector on the planet sought applications in their domain through some combination of innovative optimism and anxiety at being left behind. Seemingly overnight the starfield’s color palette inverted. If we were looking through our proverbial telescope when the acceleration began we would have been blinded by the supernova blooming from the dark.
Three years later, and that overwhelming light continues to complicate the interpretation of signal from noise. The environmental sector - like most others - is actively seeking to sort out the snake oil and focus on efficacy. Unlike many other sectors, once the snake oil is discarded environmental organizations must cross another hurdle to ensure that their outcomes are meaningfully positive by accounting for the resource consumption, pollution, and other negative environmental and public health externalities of hyperscale computing. It’s exceedingly difficult for the environmental sector to perform those analyses with the speed, complexity, and scale of the AI market. EPIC’s work with data science and artificial intelligence is longstanding, but our program is new. Newly designed to tackle the coordination, talent, and information challenges that stand between environmental organizations and a clear, actionable view of the starlit sky
Vision & Goals
We envision a near-future where environmental organizations are delivering on their missions faster and more effectively with properly scoped, developed, and deployed technology solutions empowered by rigorously tested and efficiently run AI components. That’s…wow, I mean, quite a few words, but I like all of them so lemme just do a quick explainer:
Properly Scoped - AI tools work best when optimized for a purpose. They are not “Do Anything” machines. Define purpose through explicit, on-the-ground needs.
Properly Developed - This is infrastructure, design it with the people and build it to last. Imagine this classic ad with the slogan “You wouldn’t vibe-code a bridge.”
Properly Deployed – Equip practitioners and communities with tools they need and sustain those tools over the long term.
Rigorously Tested – Establish benchmarks and run tests to understand what the system is capable of, plan around its weaknesses, and compare it to alternatives.
Efficiently Run – Resource intensive approaches to AI are harmful, unnecessary, and dragging down environmental potential, we must innovate for better.
AI technologies are tools on the toolbelt. Designed and selected for their proper purpose they empower people to repeatedly succeed and improve at the appropriate task. Actual hammer, meet actual nail. Unfortunately, “pragmatic and purposeful” are not the first two words that come to mind when describing the hype-driven AI adoption market we’re all currently operating in. Right now it’s more of a bazooka-meet-moving-target kind of world. To get this in line - at least for the environmental sector - we have a few short and medium term goals:
Pilot and evaluate tools to test out the best applications for these technologies within the environmental sector. As we identify them, we will engage with leaders in those environmental domains to connect them with the knowledge, resources, and principles to develop and deploy.
Drive meaningful innovation for AI and its enabling infrastructure that aligns with an ecosocial ethic, optimizes for resource efficiency, and distributes technology’s benefits to people and our planet.
Apply our ongoing research at the intersection of AI and the environment to develop and propagate policies for environmental agencies and lawmakers to harness the best of these technologies and mitigate their harms.
Principles & Axioms
We are guided by our dedication to accelerating sustainable solutions for ecosocial flourishing. This means finding faster pathways to provide ongoing benefits to people and the planet. AI technologies have the potential to contribute sustained speed to the delivery of positive impacts in forestry, drinking water, environmental permitting, and restoration initiatives to name a few of our domains of interest. Due to the combination of a weak regulatory environment and consolidating market incentives: those broad-based benefits are not guaranteed and many of the harms are already glaringly obvious. We adopted the following principles and axioms to guide our own work, as well as our engagements with the rest of the environmental sector:
Align with environmental stewardship and restoration – We will promote the tools and tactics that measurably advance our mission to accelerate the pace of environmental progress. Where those tools and tactics are unproven, we will do the hard work of proving their positive impact. We will increase that pace and widen our net positive impacts by measuring, communicating, and mitigating AI’s inflicted harms that hold us back.
Cultivate a culture of objective exploration – EPIC is a science and progress first environmental policy do-tank. We believe in evaluating and experimenting with opportunities as they arise. We accept that shortfalls, false starts, and stumbles are critical ingredients to the discovery and implementation of policies, technologies, and projects that accelerate the speed of environmental progress and ecosocial flourishing. We follow the evidence, sustaining our effort where it makes a meaningful impact and allowing less effective work to fall by the wayside.
Learn continuously – The only promised output of an experiment is learning. We are committed to revisiting what we learn as change occurs. We will learn from a variety of information streams, remaining open to unconventional, unexpected, and unsolicited opportunities to widen and deepen our understanding. We will apply these learnings to our own work and disseminate them to our partners, audiences, and the public.
Embrace collaboration – We don’t go it alone. We have our greatest impact in collaboration with one another and our external partners. We will seek out, and be open to, partnerships that expand the breadth and depth of our impact, knowledge, and resources. We will lead whenever necessary and support wherever possible to ensure the adoption of tools and tactics that have the greatest benefit for the environment and the most people.
The future of these technologies and our relationship with them is being written today. We have the agency and responsibility to direct the path towards ecosocial flourishing.
“AI” is a marketing term encompassing a wide variety of technologies, domains, use cases, and implementations. The differences between them are meaningful. It is valid to support some applications/pursuits within the umbrella and not others.
Technological innovation must prioritize safety, sustainability, equity, and measurable impact.
AI regulation is a nuanced and mutable topic. Policy experimentation is critical to developing a regulatory environment that promotes ecosocial flourishing.
Empowering communities, researchers, and civil servants will lead to a more effective innovation ecosystem and better outcomes from AI technologies than current monopolistic approaches.
Secrecy surrounding the environmental impacts of hyperscale data centers prevents us from understanding the environmental and social harms of AI technologies and undercuts our capacity to innovate better solutions.
All technologies, including those under the AI umbrella, have their greatest possible positive impacts when they are tightly scoped, rigorously reviewed, and continuously maintained.
Projects & Partnership
We are rolling out this program to coordinate our pre-existing projects, and to expand our work in conjunction with evolving needs. Our ongoing work at the intersection of AI technologies and the environment fall into three broad buckets: Landscape Environmental Strategy, Talent Development and Support, and Tool Development
Landscape Environmental Strategy
Nuance and North Stars: Navigating Environmentalism and AI
A qualitative research project involving government agencies, environmental non-profits, and environmental technology firms to map out the status quo and define an ecosocially productive path forward.
Providing partners and groups with recommendations on de-risking the use of AI in their work while identifying policy or process barriers to AI use, particularly on NEPA-related topics.
Talent Development and Support
Land Manager AI Skills Taxonomy
We’re mapping opportunities with partners across federal, state, Tribal, local, nonprofit, and industry settings; clarifying the skills that modern land management requires; and designing practical pathways—training, credentials, and HR guidance—that will help organizations staff and upskill to meet the moment.
Looking Ahead Through AI Literacy
Our work orients best practices in AI literacy towards the environmental domain, giving decision-makers the knowledge to understand what different AI approaches are good at and how to weigh their governance, ethical, and ecological impacts—laying a stronger foundation for responsible innovation.
Tool Development
We collaborated with Atlas Public Policy to use AI to extract data from the U.S. Army Corps of Engineers’ public notices to make it easier for agencies, advocates, and developers to see where projects were being sited and more readily engage with the planning process.
Drinking Water Service Area Boundaries
Working with SimpleLab and the Internet of Water Coalition, we engaged over 120 academic, nonprofit, industry, and government partners to prototype the first national dataset of drinking water service area boundaries (SABs).
As the program expands we are focused on:
Improving transparency into the impacts of hyperscale data centers by uplifting and unifying existing research, contributing new information, and engaging directly with communities and their advocates.
Gathering and coordinating researchers and practitioners to develop and promote environmental ethics in the science of model alignment.
Improved data stewardship, standardization, and sustainment in an increasingly large, fragmented, and uncertain time for environmental monitoring and data.
All of that work can only get us partway to the desired impact. The final - and most important - mile depends on putting these tools in the hands best-suited for the mission with the right support network.
Tool Development – We partner with other environmental technology organizations, philanthropies, and governments to produce, pilot, and deploy tools. We engage frontline practitioners with seasoned technologists and organize their efforts within a sector-wide strategy to bring the full weight of the environmental movement to bear. Interweaving the strengths of each sector to shore up the weaknesses of their coalition partners creates a stronger fabric for ideating, building, and iterating on critical technologies.
Innovating towards an Environmental Ethic – We partner with communities, practitioners, and researchers to drive the digital and physical infrastructures at AI’s heart towards ecosocial flourishing. Practitioners, communities, and their advocates are experts in their experience: we rely on their insights to understand what tools need building and how computing infrastructures are causing harm. We partner with fellow researchers to synthesize those insights, surface potential solutions, and communicate better paths forward.
Propagating Policy – We develop collaborations on a policy-by-policy basis to focus our response on the most impacted/implicated parties. With those coalitions built, we collaborate with governments to promote and incorporate recommendations that drive the AI regulatory environment towards ecosocial flourishing.
Throughout our research and engagements we’ve found the environmental sector packed with people whose excitement and imagination are matched only by the barriers they face. Walls - technical and otherwise - that rise and twist into a maze that makes first steps feel like wrong steps. Maneuvering this maze requires complementary strategies among organizations to read and react to the pace of change. Ongoing partnerships increase trust and coordination to improve the speed, accuracy, and sustainability of the decisions we make with and about AI technologies.
Conclusion
The launch of EPIC’s AI program is less about shooting off fireworks and un-tarping a grand contraption than it is about introducing the philosophies, projects, and people that we dedicate to bending the arc of these technologies towards environmental progress. So I’d like to close by describing our people a little bit. Contributors to this work include: an environmental justice data jockey, a UX research and design shenaniganist, a futurist forest ranger, a seal wrangler turned data wrangler, a theater director turned data science ethicist, and a political scientist working tech workforce wizardry. We’re…eclectic, bringing a variety of philosophies, perspectives, and experiences into play. This creates healthy internal tension where nothing gets taken for granted and ideas and assumptions are tweaked and tuned until they resonate just right. It also makes us flexible: we can apply a wide range of skills and expertises depending on specific project needs.
So, if you read this and thought “wow, exciting!” instead of “wait…telescopes?” we hope you’ll reach out. Let’s chat about building tools, mitigating infrastructural harms, stimulating organizational readiness, and distributing benefits to people and our shared planet. After all: the fuel for digital technologies comes from the planet and the power from the people, it’s only right to reinvest its benefits into those sources.