What Current Models Miss When Assessing Our Food Future (and How to Fix Them)

Tweet

Authors: By Raymundo Marcos Martinez, Javier Navarro-Garcia, Michael Battaglia, Adam Charette-Castonguay, Enayat Moallemi

In Salman Rushdie’s novel Midnight’s Children, the protagonist’s grandfather, a doctor, is asked to treat a young woman named Naseem. However, her protective father forbids him from seeing her. Instead, the doctor must examine her through a sheet with a single hole in it, viewing her body only in isolated fragments—a stomach here, a wrist there.

Over time, the doctor falls in love with these glimpses, convincing himself that by piecing together these parts, he knows the whole woman.

This story carries a profound warning for the scientists and policymakers trying to fix our global food system today. To figure out how to feed 10 billion people without destroying the planet, we rely on complex computer models. We peer at the problem through the "hole in the sheet"—using one model to look at economics, another to look at crop yields, and a third to look at calories. We fall in love with these fragments. We trust the data. But a groundbreaking review published this month in Nature Food warns that we are mistaking the parts for the whole.

Led by researchers from Australia’s science agency, the Commonwealth Scientific and Industrial Research Organisation, and international partners, the study evaluated 20 of the world’s most prominent food system models. Their conclusion? Like the doctor in the story, our current tools are missing the living, breathing, and messy reality that connects it all together.

The "Rational" Trap: We Model Markets, Not People

The latest EAT–Lancet Commission report makes it clear: we need to transform the food system to save 40,000 lives a day and meet climate targets. But our current models—our tools for navigation—may not be fit for purpose.

For the last two decades, tools like Integrated Assessment and Computable General Equilibrium models have focused on perfecting the economics. They are excellent at calculating how a carbon tax might shift global trade or how much land is required for bioenergy. However, the study highlights a glaring blind spot: these models largely assume humans are perfectly rational economic actors. They presume that if a policy makes economic sense, people will adopt it.

Real life is rarely so logical. For instance, a model might show that taxing meat is the most cost-effective way to lower emissions. In the real world, such policies could trigger massive social and political backlash, rendering "optimal" solutions impossible to implement. Critical factors like shifting social norms, cultural preferences, and political lobbying ("vested interests") are often treated as external assumptions rather than dynamic parts of the system. By ignoring the human element that leads to “suboptimal” decisions, we risk designing policies that look perfect in a spreadsheet but fail in the street.

The Top-Down View

Current models tend to view the world from a high altitude, aggregating data into large regions. While this simplifies the math, it erases the critical differences on the ground. The review points out that this approach disproportionately favours industrial agriculture because data on large-scale systems is easier to find. Meanwhile, Indigenous and Local Knowledge is often invisible. Traditional farming practices, local food processing, and subsistence agriculture, which are vital for food security in many parts of the world, are frequently glossed over. This isn't just a data problem; it's an equity problem. If our models don't "see" smallholders or Indigenous communities, the policies derived from those models likely won't serve them.

Expecting the Unexpected

If the recent years of pandemics and geopolitical conflict have taught us anything, it is that the future is nonlinear. Yet, the review finds that most food system models struggle to handle acute shocks. Most current tools operate on the assumption that risks are predictable ("known knowns"), like the gradual impact of climate change on crop yields. They are ill-equipped to simulate "deep uncertainty", the sudden, unpredictable events that can shatter global supply chains. When our tools assume stability in an unstable world, we leave ourselves vulnerable.

Removing the Sheet: A Toolkit for Transformation

The authors do not suggest tossing these digital tools into the scrap heap. Instead, they envision a radical evolution, a shift from rigid prediction to dynamic exploration. This transformation begins by putting the "human" back into the machine—for instance, through the integration of Agent-Based Models. Unlike standard economic models that view us as rational calculators, ABMs track individual "agents" such as a sceptical farmer, a trend-conscious consumer, or a worried voter, allowing researchers to watch how messy, real-world behaviours ripple through a system. It is the difference between modelling what people should do mathematically and understanding what they might actually do under certain conditions. But better math is insufficient without grounded truth. The paper advocates for breaking down the walls of the laboratory through participatory modeling, inviting local stakeholders and Indigenous communities to co-design the scenarios. This approach ensures that global data does not override on-the-ground realities, validating that a solution is not just scientifically sound, but culturally feasible.

Scientists are also harnessing Artificial Intelligence (AI) to better account for the unpredictable nature of the future. Instead of relying on a handful of "likely" scenarios, AI and surrogate models allow researchers to run thousands of rapid-fire simulations in a fraction of the time. This enables them to explore a vast "uncertainty space" and search for robust strategies that hold up even when unforeseen events occur. However, speed cannot come at the cost of scrutiny and objectivity. We must remain critical of who decides what is represented in these algorithms; otherwise, we risk simply automating implicit bias in training datasets, using AI to potentially amplify existing issues faster than ever before.

Finally, the authors suggest abandoning the quest for one perfect, all-encompassing super-model in favour of a "plug-and-play" architecture. The future lies in model coupling, linking a global trade model with a local water simulation, or connecting climate data or food production and consumption data directly to public health metrics. By connecting these distinct tools, we can build a flexible network that, like the food system itself, is greater than the sum of its parts. However, realizing an agile and sustainable plug-and-play architecture presents challenges within the current funding landscape, which often prioritizes immediate project outcomes. While significant investment is made in developing scientific capability for the public good, bridging the gap between this expertise and the sustained operational support required to maintain robust food system models remains a critical priority.

From Fragments to Whole

The doctor in Midnight’s Children made the mistake of thinking that a collection of parts equals a person. We cannot make the same mistake with our planet.

Food system transformation is political, cultural, and deeply human. Success relies on agreement among diverse stakeholders. We need to move beyond the fragmented view of the "hole in the sheet" and bring scientists, policymakers, farmers, and businesses to the same table. Only by integrating these diverse perspectives can we build a shared foundation of evidence and turn visionary reports like the EAT-Lancet into roadmaps for concrete action.

This post is based on the review article "Complexity and uncertainty in future food system transformation modelling" published in Nature Food (2025).