“A New Era of Food Safety”: How the Food Industry Can Leverage AI for Safety

Rigorous environmental monitoring and preventive food safety procedures help mitigate the risk of a recall at your facilities. But what about a recall threat you don’t see coming, one that’s embedded in a far-flung supply chain, affecting products your company doesn’t even make? How do you stay proactive when you don’t even produce the product in question?

“It comes down to managing risk across supply chains when huge swaths of data are available. And they're not connected, they don't seem relevant, and a lot of them aren't even yours,” Seán Leighton, Senior Vice President of Food Safety, Quality, and Regulatory for Cargill, said last week.

The Cargill Case Study

Leighton’s remarks came during a panel at last week’s Food Safety Summit in Rosemont, Illinois, entitled Leveraging AI for Food Safety: From Strategy to Impact. The discussion brought together academics and food safety experts from McDonald's, Chick-fil-A, and Cargill to discuss how AI is shifting the industry from a reactive to a predictive, confident food safety culture.

Leighton described how the AI-powered tool that the multinational food production giant built has helped it avoid 41 potential global contamination events. The tool has done so by connecting massive amounts of data to suss out the meaning for food safety.

Among other methods, Leighton said the AI tool uses millions of lines of data on commodity pricing to predict safety issues. He described how, as commodities become more expensive, suppliers are more likely to bypass standard safety protocols or adulterate products to cut corners and save money.

The company doesn’t have to wait for a lab test to find a contaminant. Instead, it can monitor price fluctuations for red flags and take a proactive approach.

Leighton divided into a 2020 contamination event - the most difficult of his career - which involved ethylene oxide in sesame seeds - a product that Cargill doesn’t make. The investigation eventually reached locust bean gum made in Turkey that also contained trace amounts of ethylene oxide. Locust bean gum, a texturizer produced by Cargill for use in a wide variety of food products, has a shelf life of 5 years, while ice cream that contains the texturizer has a shelf life of 3 years. This means that recall mitigation efforts had to stretch back years to allocate and remove the contaminated products.

"How do you react to something that hasn't happened yet? In this day and age, AI is here to help us with all of that,” Leighton said.

Not to Replace Humans - But to Amplify Their Work

Cindy Jiang, who worked for McDonald’s for more than 30 years, including as Senior Director of Global Food Safety Risk Management, spoke about the role of AI in protecting food products amid increasingly large and complex supply chains.

“Supply chains are global and highly localized, and data is growing exponentially. But it's still not fully leveraged, and risks are evolving much faster than traditional systems can respond. So the question isn’t about commitment or expertise, it’s, can we continue to rely on yesterday’s approach to manage tomorrow’s risk? That’s where AI comes in.”

Jiang noted that the goal with AI is “to analyze data at a scale no human can do alone,” to “move from reactive excellence to predictive confidence, that is a shift from strategy to impact.”

It’s not a lack of data - but how to use it

Lisa Shelley, Ph.D., Director of Research, Safe Plates, North Carolina State University, said AI helps the industry parse and understand the massive amounts of data at its disposal.

“It’s more data than any one person or one team could actually use. This is where we see AI tools really starting to close the gap, not by adding new data, but by finally giving us the capacity to read what we already have.”

Shelley compared it to the academic world, where there is a surfeit of datasets from research projects, but it can be difficult to determine how to hone this data and identify trends and signals within the numbers.

“We already have the data, but we can't read and understand it all in real time. This is where I think AI is shining right now,” Shelley said.

Centralize your data - and bring in the data scientists

With the increased use of AI, companies should also pursue expertise from outside the world of food safety, Steven A. Lyon, the Director of Food Safety–Field Operations for Chick-fil-A Inc., said

“Scientists and food microbiologists can only do so much. When we brought in data engineers and data architects, we really saw the shift in AI go extremely, extremely quickly,” Lyon said.

Chick-fil-A is using AI to create risk modeling for its highest risk pathways - chicken, produce, and norovirus transmission, Lyon said.

For AI to be successful, Lyon said that companies must stop siloing data and centralize it so it can be better utilized for safety.

“You’ve got to centralize your data. That's the most important thing that we found to start off with. That may mean getting rid of some data that you really don't need anymore.”

By running historical incident data through AI models, Lyons said the company can get proactive on risks in a way they couldn’t before.

In a nod to many of the concerns in the industry about the rising use of AI, he closed with an emphasis on the crucial role humans play.

“We are only using AI to enhance our decision-making process, and the importance of humans does not go away with AI; if anything, it's more and more important. Humans play a big role in this, and AI isn’t replacing them.”

AI and the Scale of Safety

At the close of the talk, John (Chuck) Hassenplug, a Senior Policy Analyst for the U.S. Food and Drug Administration, discussed how the FDA’s Human Food Program is using AI to handle the massive scale of the issues facing the U.S. food supply, despite a shortage of resources.

Hassenplug described how the FDA is using AI for research and rapid tests to quickly respond to potential food safety incidents. They also use it to expedite communication about food safety events and to determine where to best prioritize their oversight work. He said that the FDA is able to find a 3 to 5 times higher violation rate for samples predicted by their AI model.

He advised the industry to use AI to find the early signals that can help spot an outbreak early, to quickly scan images and labels, and to go through satellite imagery to assess if facilities are operational.

But like Lyon, he also cautioned against the overreliance on AI, and highlighted the necessary role of humans.

“The human is always in the loop. These tools are guided and prompted by humans, and the output is closely looked at, and a human makes the final decision moving forward."