AI and Food Security: Promises and Challenges

Willette M. Crawford, a food safety consultant and co-founder of Katalyst Agricultural Solutions, describes how AI fits into outbreak management and recalls. She explains that the tool accelerates the analysis of massive data sets without detecting pathogens itself, while highlighting persistent obstacles related to costs, data, and time. She believes that the apparent increase in recalls primarily reflects better visibility of risk and recalls her contribution to the Food Safety Modernization Act, which directed the FDA towards prevention.

Incomplete Governance and Fragmented Global Supply Chain Limit AI

There is no comprehensive regulatory framework specific to food for the use of AI in food safety. The food supply chain operates as a complex network of siloed operations, intersecting during processing, transportation, or changes of hands of products. The entities involved do not always share the same standards, priorities, or regulatory requirements, and each generates large amounts of records and information. This chain is global, necessitating navigation through different regulatory systems, record-keeping practices, and operational standards, even if it allows for increased product availability. When the cost of AI is prohibitive for many players, even the best-equipped companies must consider the limitations of their partners.

Digitization and Data Quality: Conditions for Real Gains

Companies that succeed in using AI in food safety have first digitized their information. In contrast, those deploying tools without this step struggle to see a return on investment. Those investing in digitization, even though it requires significant effort, are better positioned. This work involves staff dedicating time to system architecture, which is crucial from the outset. Incomplete, inconsistent, or poorly maintained records limit the benefits of an AI system. AI allows for linking scattered information and automating processes, freeing up time for interpretation and investigation, but the main obstacle remains the quality and availability of data, more so than the mathematical aspects.

Current Uses: Sequencing, System Links, and Accelerated Recalls

In the regulatory field, dedicated networks for whole genome sequencing are being established to enhance outbreak detection. AI solutions do not directly detect pathogens, but they facilitate their identification within the system, connect different systems and states, and better assess the extent of an outbreak or the necessity for a recall quickly. Artificial intelligence is seen as a support for faster and more targeted methods, due to its ability to continuously integrate larger and more varied data volumes, reveal non-explicitly programmed links, and determine action priorities with greater precision. Willette M. Crawford, who has worked for two decades in the field and contributed to the Food Safety Modernization Act that shifted the FDA towards prevention, situates these uses within a more proactive dynamic.

Early Adoption, Costs, and Time Weigh on Stakeholders

The adoption of AI in food safety remains at an early stage, primarily among certain large companies. For smaller entities or producers, cost is often a significant barrier. The time required to configure models, validate them against existing systems, and designate a responsible party for their long-term governance constitutes another obstacle. High costs and non-uniform data add to the list of challenges faced.

More Visible Recalls, Risk Not Necessarily Increased

After a series of recalls this summer involving lettuce, jalapeños, and blueberries, Willette M. Crawford believes that the perception of increased risk among consumers is not necessarily justified. According to her, the food supply is not riskier, but the ability to detect risk has improved. As a co-founder of Katalyst Agricultural Solutions, she emphasizes that AI helps teams sift through enormous data sets and manage increasingly large volumes of information.