Cornell study finds trust is holding back AI-powered food safety

Artificial intelligence could help food manufacturers spot safety risks earlier and identify patterns that individual companies would struggle to detect alone, but new research suggests the biggest obstacle may have little to do with the technology itself.

• Cornell-led research found food companies recognize the potential benefits of pooling confidential food safety data for AI
• Interviews with 27 industry leaders identified trust, incompatible systems and inconsistent recordkeeping as major barriers
• Researchers say neutral third parties and clearer standards could make data sharing more practical

A Cornell University-led study found that food companies recognize the potential value of pooling confidential food safety data, particularly as larger datasets could improve predictive models and reveal rare patterns associated with foodborne outbreaks. Concerns over how that information might subsequently be used, however, continue to limit companies' willingness to share it.

The research, published in npj Science of Food, was led by Cornell doctoral candidate Linda Kalunga and involved collaborators from Cornell, the University of California, Davis and the University of California, Berkeley.

Researchers interviewed 27 executives, food safety directors and managers from the dairy, meat, produce, food manufacturing and food safety laboratory sectors. Participants broadly saw advantages in combining data, but raised concerns ranging from legal liability and regulatory scrutiny to competitive risk and incompatible digital systems.

“Before I began this research, I expected that companies would be hesitant to discuss sharing their food safety data, especially when it came to collaboration with competitors,” Kalunga said. “I was surprised by how openly participants shared their perspectives.”

Pooling data could give AI systems access to incidents and patterns too uncommon to appear frequently within the records of a single manufacturer. According to the study, participants believed this could help companies identify trends earlier, strengthen predictive models and improve their understanding of rare foodborne outbreaks.

There could also be benefits for smaller manufacturers that lack the resources to invest heavily in their own research and data analytics. Access to insights generated from a wider industry dataset could potentially help those businesses identify problems earlier without requiring the same internal infrastructure as larger food companies.

But the research found significant differences in how food safety information is currently collected and stored. Larger companies may operate sophisticated digital platforms, while some smaller businesses continue to rely on spreadsheets or paper records. Incompatible systems and inconsistent recordkeeping make combining information across businesses considerably harder.

“Sharing confidential food safety data with competitors to generate AI-driven insights from larger datasets is a tempting proposition,” said Renata Ivanek, professor in Cornell's College of Veterinary Medicine and co-director of the Cornell Institute for Digital Agriculture. “It has enormous potential benefits, but also many ways to fail.”

The more difficult problem appears to be what happens to information after a company releases it.

Interviewees were concerned that confidential data could be taken out of context, attract additional regulatory attention, create legal exposure or provide competitors with commercially useful information. One participant told the researchers that data could be “used as a weapon against me” unless its protection could be assured.

That creates a difficult trade-off for manufacturers. The benefits of a larger dataset could be distributed across the food industry, while an individual company sharing its information may carry much of the immediate risk.

“For AI to be effective, there has to be a willingness among people to collaborate on data sharing,” Kalunga said.

The findings come as AI and advanced data tools take on a larger role across food production, from product development and regulatory work to process monitoring and manufacturing. For food safety applications, however, the quality, quantity and diversity of the information available to train and operate models can determine how useful those systems become.

The researchers identified neutral organizations as one possible route around the impasse. Participants suggested universities and other independent third parties could help manage shared information, establish privacy protections and create clearer standards governing how data can be accessed and used.

Rather than pointing to another technical breakthrough as the next requirement for AI-based food safety, the study suggests companies first need a framework in which they are comfortable contributing the information those systems require.

“This is one of those systematic issues where understanding the whole ecosystem is necessary to solve the problem,” Ivanek said. “Each discipline on our team contributed a different piece of the solution, from designing interview questions to identifying the practical requirements for making data sharing work.”

If you liked this, check these out...

• The intelligence layer: Why food companies are racing to capture what they already know

• University of Nottingham launches BioSee AI to bring real-time visibility inside bioreactors

• Merck, Vow, MilliporeSigma and Solar Foods lift the lid on keeping alternative proteins safe