How A.I. technologies could help resolve food insecurity

Because food insecurity is so wide-reaching and triggered by countless circumstances, data is critical in streamlining responses. This is where A.I. is helping to mitigate hunger stemming from the complex circumstances that lead to food insecurity in its many forms. For certain instances, A.I. and machine learning lets organizations quickly gather and interpret large amounts of data to evaluate areas of need: predicting where and why hunger occurs and efficient food distribution.

When humanitarians lack data, structured and timely decision-making is hindered.

Forecasting hunger crises

When it comes to food aid, timing is vital. Availability is often thwarted when crises occur, like climate and economic shocks, or conflict and migration. Accurately forecasting helps relief arrive to the right areas on time. A study published in January 2022 by the University of Illinois details how machine learning models can help facilitate better informed and quick decision-making in the complex, ever-changing environment of food insecurity.

Yet, early warning can be difficult. Collecting data is costly, and critical data can be unattainable in locations hard to reach or where technology is absent. Comprehensive forecasting requires statistical models and readily available data on prices, weather, and demographics, which allows responders on the ground to conduct proficient intervention.

“People are doing really heroic work in data-scarce environments. But some places could benefit from incorporating algorithmically-based investments,” says Hope Michelson, co-author of the study and associate professor in the Department of Agricultural and Consumer Economics at the University of Illinois. She adds that humans are necessary for generating predictions about food crises in places with political complexity and very little data. “We are not proposing that any A.I. or machine learning-based methods should replace that system. But the idea is that you could have a multi-pronged process.”

Mitigating food insecurity involves thoroughly analyzing information across different data sources and platforms. The United Nations World Food Programme (WFP) aids nearly 128 million people in more than 80 countries. To achieve this, workers must balance the constantly evolving needs for location-specific realities: some countries are landlocked, some have access to ports, some prone to floods, or some experience something unexpected. The Frontier Innovations team at WFP assesses how nascent technologies like blockchain, A.I., or robotics can help operational teams when variables are distinct or constantly changing.

“In the past, there was data collection, but in a different way. You’d send surveyors to communities to collect data and then use different technology platforms. But, the data did not come in an aggregated manner or couldn’t be cross-pollinated as quickly as A.I.,” says Hila Cohen, head of business development and chief of staff for the WFP Innovation Accelerator. In addition, she noted it wasn’t as predictive as it is currently. “We [used to] receive a weekly or monthly report then assess a certain trend. A.I. gives you much more data points and speed.”

One such technology is WFP’s HungerMap LIVE, which tracks and predicts food security in almost real-time. By combining critical variables like weather, disease, population, war, nutrition, and macroeconomic data, WFP can display insights on an interactive map that is accessible and free to use.

These details guide decisions on where and when to place food before a crisis hits, which is often too late. Using its Optimus technology, WFP can optimize interventions across various situations by aggregating data to determine what to donate, how to get it, and how to distribute it in a timely and cost-effective way. According to WFP, Optimus alone has saved more than $50 million since its implementation.

Distributing more efficiently

According to the nonprofit ReFED, dedicated to ending food loss and waste, an estimated 35% of 229 million tons of consumable food in the U.S. went unsold or uneaten in 2019. That’s the equivalent of almost 90 billion meals. The problem with food waste is two-fold. First, if it ends up in a landfill, it rots and releases toxic methane gas, which in 2020 led to 14.5% of total methane emissions. Second, it leaves a lot of perfectly edible food unavailable for people who need it.

Because food comes from many origins along the supply chain, streamlining data helps reallocate excess food or waste and diverts is away from landfills. However, Blake Harris, technical director of the Global Food Traceability Center at the Institute of Food Technologists, asserts that it is challenging to teach an A.I. program without data standards (i.e., everyone collecting the same data in the same format). “Once consistent data is available, algorithms could be ‘taught’ to quickly identify extra supply and divert it to locations in need,” he wrote in an email. “Supply chains that can organize and share data between partners can better coordinate between growers, processors, distributors, and retail/food service to be more efficient and reduce waste and thus reducing the environmental burden of food production.”

But because A.I. technology and cooperation like that isn’t scaled yet, so private companies are using A.I. to reallocate recovered food donations efficiently. Donating excess food isn’t new, of course. What’s new is how distribution is becoming more effective.

For example, San Francisco-based Replate collects surplus food from vendors and delivers it to nonprofits in a strategic, data-driven format. “Instead of just moving food around, we took a step back to understand how to match the right food to the right nonprofit,” says Replate founder and CEO Maen Mahfoud, who grew up in Syria where his mom encouraged him and his brother to offer part of their meals to hungry neighbors before they ate.

Replate’s vendors range from San Francisco International Airport, grocers, and restaurants to companies that provide lunches to employees. With Replate, they input information about quantity and type, request pickups, and track donations over time. The latter empowers them to learn from their waste and order more efficiently in the future. For example, if a donor sees that they’re constantly donating excess beans, Mahfoud hopes they reduce that source from the onset.

“Ideally, the donor will benefit from that by being a bit more informed. Stop ordering beans, for example,” says Mahfoud, adding some companies leave Replate once they’ve refined their procurement habits and eliminated excess completely.

For recipients, helps determine what donations they actually need. Charities regularly face challenges with storing and distributing food. Additionally, they’re left with items they don’t have the capacity or need to use. When this occurs, the nonprofit absorbs the costs of storing or disposing; sometimes leading to waste.

To alleviate these burdens, Replate collects demographic information of the charities, like the site’s capacity to sort and store, the number of people it serves, when it needs food, and what type of food it requires (including preferences, like Halal, Kosher, and more). Replate also considers nutritional factors, so donations add value to people’s diets. This type of data helps donors become more thoughtful with their contributions.

“From a systemic standpoint, there are two things: you’re helping people experiencing food insecurity, but the question is, are you really helping them,” Mahfoud explains regarding donations that don’t necessarily align with the charity or recipient’s requirements. “As a company without enough data, you might think you’re doing good things, but you might be doing the wrong things.”

Replate addresses this by purposefully connecting donations with specific needs.

Data collection for humanitarian efforts

As with much A.I. integration, balancing good intentions with ethics is prevalent, especially in food insecurity. Context matters; on a global scale, one format or data set doesn’t fit all. Problem-solving at scale requires massive amounts of data from a multitude of sources.

“That’s one of the things the community is trying to solve with its pre-standardization efforts,” says Frederic Werner in an interview. As head of strategic engagement for International Telecommunications Union’s standardization bureau of A.I. for Good—a United Nations platform for A.I. dialogue—Werner is eager to watch A.I. evolve in the humanitarian space in a mindful and collaborative way. “The mechanism under which you can share data needs to be worked out, because that’s what is missing to enable problem-solving at scale using A.I.”

In a dream world, all data will be searchable, discoverable, and labeled to distinguish free data, licensed data, and so on. But mostly, data will remain protected. “When this much data is being shared and utilized, it needs to be done in a thoughtful way and interpreted correctly in conjunction with policymakers from the start,” Michelson urges when discussing vulnerable and underserved communities.

Governments and policymakers can be significant drivers in this. Recently, the Biden Administration also published guidelines to ensure “underserved populations are empowered by and benefit from federal data, surveys, and equity assessments.” Additionally, another set of guidelines created by WFP’s accelerator team members illustrates how access to data could become “the largest practical bottleneck to developing humanitarian A.I. applications, which require a continuous flow of annotated data to train new models and update old ones.”

“WFP serves people in in very, very vulnerable situation,” stresses Cohen. “Data privacy for us is crucial. So, we have to make sure that the data that we put in doesn’t, in any way, breach the security and privacy of people.”

Currently, those who hold valuable and relevant data streams hesitate to share it widely. Without distinct regulations to protect data and use it thoughtfully, the bottleneck may continue.