Applying Decision Intelligence to Conservation's Toughest Challenges

Feb 7

Imagine you’re an ecologist on the West Coast of the United States. In recent years you have taken great efforts to protect sea otters, which are endangered over much of their former range. In some places sea otters have returned, begun breeding, and are eating one of their favorite foods: sea urchins. Along with this great news, the resulting decrease in sea urchins has allowed the recovery of kelp forests, which is good habitat for fish. More fish has helped the economies of these states.

Everything’s great, right? Not quite: sea otters also eat other animals, including lots of butter clams, geoduck clams, and crabs, which has cost these fisheries millions of dollars.[1] In addition, some sea otters have died from pathogens that are suspected of coming from domestic animals on land, a problem that will only worsen as we increase the number of both people and sea otters in the same areas.[2] Benefits for some creates a large cost for others, but reversing who benefits and who suffers does nothing to improve outcomes overall.

This is what experts call a "wicked problem," and it's exactly the kind of challenge that decision intelligence was designed to address.

Decision intelligence (DI) is an engineering approach to decision making within complex situations that combines people, process, and technology to create decision systems that learn over time.[3] What makes DI particularly powerful for conservation is its use of causal diagrams and simulations, tools that map out unintended feedback loops before you stumble into them. Over this monthly series, I'll explore how DI techniques can tackle complex conservation challenges and other public policy problems that share the same thorny characteristics.

What Makes a Problem "Wicked"?

Many conservation and public policy problems are "wicked problems", complex social system problems where information is confusing, stakeholders have conflicting values, and the ramifications ripple unpredictably through entire systems.[4] In their foundational work, Rittel and Webber identified ten characteristics that define these problems.[5] Four stand out as particularly relevant to conservation:

There's no definitive formulation of the problem. Ask five stakeholders what "the homelessness crisis" actually is, and you'll get five different answers: mental health failure, housing shortage, addiction epidemic, wage stagnation, or family breakdown. Each framing points toward completely different solutions.

Solutions are better or worse, never right or wrong. You can't prove a wicked problem is "solved." Did that kelp forest intervention work? This depends whether you ask marine biologists, fishermen, or otter advocates.

Every attempt counts significantly. Unlike laboratory experiments, you can't run controlled trials on homelessness policy or forest management. Each intervention changes the system, potentially irreversibly, making traditional trial-and-error learning impossible.

Every wicked problem is essentially unique. Cookie-cutter solutions fail because each situation involves different stakeholders, different ecosystems, and different feedback loops. What worked in Seattle's homelessness programs may fail in Phoenix.

These problems are wicked because they involve complex systems, which as Donella Meadows describes, exhibit resilience, self-organization, and hierarchy [6]. They're nonlinear, full of feedback loops that vary in timing and magnitude. Push the system too hard in one direction and it might flip to an entirely different stable state, perhaps irreversibly. Ecologists have studied this behavior for decades, watching lakes flip from clear to turbid, forests shift to grasslands, and fisheries collapse and never recover.[7]

Wicked Problems in the Wild

Consider homelessness. This problem sits at the intersection of mental health, addiction, housing markets, labor economics, family breakdown, and criminal justice. Each stakeholder (local governments, landlords, social service agencies, businesses, housed residents, and unhoused individuals themselves) has different interests and agency. The feedback loops operate on wildly different timescales: housing takes years to build, but someone can lose their apartment in weeks. Healthcare system changes take even longer, creating loops that stretch across decades while people suffer today.[8]

Or consider a problem closer to conservation's heart:

Drug-resistant bacteria represent what researchers call a "super-wicked problem"—a wicked problem where time is running out, those causing the problem must provide the solution, no central authority exists to coordinate action, and political will focuses on short-term thinking.[9] Pharmaceutical companies, hospitals, physicians, patients, agricultural producers, and regulators across multiple countries each have agency and competing interests. Resistance evolves over decades, but political and economic pressures operate quarterly, systematically favoring decisions that accelerate the crisis.

In conservation, we see this pattern everywhere. Protecting an endangered salmon run requires coordinating dam operators, farmers, tribal fishing rights, recreational anglers, hydropower customers, and the salmon themselves (yes, non-human species are stakeholders too, with their own agency, fitting into complex systems at different hierarchical levels, often self-organizing into family groups). Change water flow to help spawning? You've just affected agricultural irrigation. Restrict fishing? You've ignited conflicts over treaty rights and economic survival. Every solution creates new problems.

Enter Decision Intelligence

Given this complexity, it's unsurprising that policymakers struggle. The time frames don't align, information arrives incomplete or too late, and confounding variables make it nearly impossible to measure whether interventions actually work. This is exacerbated by the fact that machine learning models that do a great job at showing correlation say nothing about causation, which is what we need to understand to move forward.

Historically, policy analysts, ecologists, and government agencies have attempted to tackle wicked problems with systems modeling. Tools like Stella from isee systems and Analytica by Lumina Decision Systems allow you to draw the complex connections between these elements. However, systems modeling is not widespread. Why: because it requires specialized expertise to create and read systems diagrams. They are, simply put, too hard to use.

Decision intelligence offers a different approach. DI brings together front-line stakeholders, decision-makers, researchers, and technologists to build decision-support systems grounded in causal thinking. Unlike other approaches, DI works one decision at a time. This keeps decision mapping much simpler than attempting to map an entire system at once. In DI, we explicitly draw maps causal dependencies in complex systems and links specific interventions available to specific roles. We can do this in a simple diagram – which has a surprising amount of value on its own – and/or add simulation so as to link actions to desired outcomes using machine learning, simulation, and optimization.

Here's how it works in practice: The process begins with creating a causal decision diagram (CDD), a visual map of what causes what in your system. For the sea otter problem, your CDD might show:

otter return → urchin removal → kelp recovery.

But the same model might also show:

otter return → crab fishery declines → political pressure → otter removal → urchin population recovery → kelp decline.

Opposite outcomes for the same action! There are likely also dozens of other causal pathways: how water temperature affects kelp, how otter predation affects urchin genetics, or how fishing pressure affects the broader ecosystem. This can become quickly overwhelming, so DI keeps it simple by working one decision maker at a time, which means we don’t end up with systems diagrams that nobody can use.

This diagram becomes both a shared language for all stakeholders and an experimental framework. As the decision maker implement experiments with decisions and observes outcomes, they test your causal hypotheses in the decision model. Did kelp respond as predicted? Did otters behave as expected? Each result refines the model, improving future decisions. The system learns.

This is why DI focuses on human-in-the-loop decision-making: high-complexity, high-stakes decisions are often a collaboration between people and computers and don’t make sense to fully automate. The human role in the collaboration comes from the fact that we are natural causal thinkers who run mental models of how our actions lead to outcomes every day. When combined with modern data analytics, this creates systems that improve over time rather than repeating the same mistakes.

What's Next

In my next post, I'll dive deeper into decision intelligence itself: where it came from, how it's being applied today, and why its combination of qualitative and quantitative knowledge makes it uniquely suited for conservation challenges. I'll walk through a more detailed causal decision diagram and show you how practitioners use these tools to navigate exactly the kinds of impossible trade-offs we've discussed here.

Because here's the thing about wicked problems: you can't solve them, but you can get better at navigating them. And sometimes, getting better is all we need.