Transforming Global Food Systems with AI

Radhika Krishnaswamy, Senior Vice President, Findability Sciences

Recently, I was watching a documentary with my children about leafcutter ant colonies when my ears perked up as the narrator described the ants as ‘farmers’. Leafcutter ants harvest leaves they do not eat directly. Instead, they cultivate the leaves into fungus, which becomes their primary food source. Different ants specialize in different roles, with physical characteristics adapted to the tasks they perform. 

I had been thinking about the global food system for this article, and given a subject area of this magnitude where outcomes are driven by more than individual participants’ actions, the leafcutter ant colony mechanisms proved thought provoking. To feed 10 billion mouths by 2050, we know that we can do more with the resources we currently have – and so improve productivity, build resilience to handle climate change, and manage costs. But to do so in a way where access to nutritious food is equitable, food safety is strong, and food that is produced is sustainable for the long term, will require collaboration beyond optimising the siloes. 

Can AI support both goals? 

To explore this question, I’ll borrow a few analogies from the ant world (ironically, from an agricultural pest!) to look at two ways AI can support near-term and long-term food-system outcomes: 

1. Building resilience within individual parts of the system using data we already have — a problem AI is increasingly well suited to address. 
2. Building resilience at the ecosystem level, where coordination across participants matters as much as optimization within any single stage. 

Goal 1: Building resilience through better decision-making with “wide” data 

Within its specific role, the leafcutter ant is highly optimized. Foragers cut leaves into shapes and sizes that are easier to carry over long distances, adjust the types of leaves they collect based on feedback from the fungus garden, and increase or reduce foraging effort depending on the colony’s needs. These ants respond continuously to multiple signals, allowing their part of the system to remain resilient under changing conditions. 

In a similar way, for our global food systems, “wide” data signals collected at the farming level can help build resilience to improve farming results. Satellites passing over my country every day? Let’s grab the images for my farms. Deep historical knowledge collected by farmers over years? Let’s develop a model that “learns” from this information to guide actions for the current season. Thousands of sensors on the ground for more granular information? This becomes an additional data feed. AI is very well suited to handle data like this and keeps improving as the data quality improves. Combining this capability with experienced domain knowledge has resulted in models to improve farming productivity. Of course, model development is not a new concept – what’s different now is advancements in AI and cloud computing  are enabling modeling with vast amounts of data.  

Every stage in the food value chain can follow a similar process of resilience building. In agro-industrial settings such as sugar mills, artificial intelligence can be applied to real-time operational data spanning milling, clarification, evaporation, crystallization, and centrifugation. Predictive models and statistical optimization techniques help detect anomalies, recommend operating set-points, improve throughput, and reduce energy consumption. Many of these techniques have existed in theory for decades, but their practical deployment has accelerated due to improved data availability, cloud-ready infrastructure, and advances in statistical and machine-learning methods. 

Goal 2: Ecosystem-level problem solving 

In our Leafcutter ant colony, each stage performs its actions well, but if they were not working towards a larger goal, the colony would fall apart. That larger goal guides the “institution” that orchestrates actions. And in the ant colony, the fungus-monitoring ants form the governing institution! Individual ants can survive based on the actions they perform, collecting leaves, making and eating delicious fungus, but if it does not support the entire colony and future offspring, then over time the colony suffers and dwindles in size. 

In human global food systems, the complexity is multitudes higher of course. There are several larger goals, with varied institutions focusing on different problem statements.  

Some characteristics of messy problems: they require a lot of coordination, are multi-domain in nature, and rewards are not immediate.  

Here are some multi-disciplinary food ecosystem problems and the role that AI could play with human collaboration and creativity and generating patterns across seemingly disconnected events: 

• Food safety: Stakeholders across the supply chain align on standards, share data, and use AI to detect deviations from normal patterns before failures occur. 
• Famine detection: Combining geopolitical signals, climate data, and market indicators to identify emerging risks and trigger earlier intervention. 
• Food security and nutrition: Integrating public health, income, and demographic data to identify vulnerable populations and target support more effectively. 

What can AI do when rewards are not immediate? Researchers are exploring this and many other thorny human-centered topics. Here is one example on the application of AI to enable greater openness and collaboration.  Notice how the application of AI often begins with data and models, but succeeds only when shaped by human collaboration and domain expertise — especially for complex, system-level challenges. 

Ultimately, ecosystem level problems are thorny ones to solve and I am curious whether we have gained the ability to do things differently with AI.