Science1 publisher3 min readPublished
Food executives fear pooled safety data could be used against them, Cornell interviews show
Cornell researchers who interviewed 27 food industry leaders found that trust, more than technology, keeps companies from pooling safety data for AI. The evidence is what executives told interviewers, so whether removing those barriers would free up the data is still an open question.
The Scientist · Science desk
What happened
- Participants expected larger shared datasets to reveal trends earlier, improve predictive models and help explain rare foodborne outbreaks.
- Executives worried that shared data could be taken out of context, expose them to legal liability, draw regulatory scrutiny or give competitors an unfair advantage.
- Companies also cited inconsistent recordkeeping and incompatible systems, with many smaller firms still keeping records on spreadsheets or paper.
- Participants said clearer standards and neutral third parties, universities among them, could provide the privacy protections that would make collaboration possible.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- contradiction Kalunga ranks trust above technology, yet the smaller firms participants named as beneficiaries keep paper or spreadsheet records, so a trusted custodian on its own would still leave them out of the pool.
- cost Whoever contributes first takes on the legal and regulatory risk alone while the gains spread across the industry, so each firm has a reason to wait for its competitors to share first.
- decision Anyone organising a data pool has to settle on a custodian and a common record format before asking firms for data, because participants tied their willingness to share to both.
Kalunga's team spoke with 27 executives, food safety directors and managers across dairy, meat, produce, food manufacturing and food safety laboratories [1]. Interviews with a sample like that show why people across an industry hesitate. They do not measure how widely any one reason is held, or what executives will actually do. Kalunga went in expecting reticence. "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," she said. "I was surprised by how openly participants shared their perspectives." [13]
The fear participants described was specific. "Once I give that data away, unless I'm absolutely confident that it's protected, it can be used as a weapon against me," one told the researchers [8]. Kalunga put the problem down to who carries the risk. Each company that shares takes the risk on alone, while the benefits spread across the industry [10]. This is a public-goods problem, in which each firm does best by letting others contribute first. "For AI to be effective, there has to be a willingness among people to collaborate on data sharing," Kalunga said [9].
There is a tension inside the findings. Kalunga said they pointed to trust, not technology, as the greatest obstacle [6]. Yet the same interviews turned up the spreadsheet and paper records common at smaller firms [5]. Several participants also said pooling would give smaller companies insights that would otherwise require costly investment in research and analytics [4]. So the firms singled out as beneficiaries are the ones whose records would be hardest to pool [1]. The fix participants suggested, a neutral custodian, would do nothing for firms whose records are on paper. Kalunga's broader summary is more cautious than the headline finding: she said realising AI's potential for food safety may depend as much on organisations' willingness to collaborate as on advances in the technology [14].
The study came from Cornell, with collaborators at the University of California campuses in Davis and Berkeley [2]. Renata Ivanek, in whose lab Kalunga works, summed up the proposition. "Sharing confidential food safety data with competitors to generate AI-driven insights from larger datasets is a tempting proposition," Ivanek said. "It has enormous potential benefits, but also many ways to fail." [12]
The study does not test whether pooled data would catch an outbreak sooner. The predictive gains are the ones participants expect [3], and the study as reported rests on interviews [1]. I think the evidence supports a narrower version of the claim: executives say trust and data standards stand between them and a shared dataset. Two further questions would each need a test of their own. The first is whether removing those barriers actually gets companies to hand over data. The second is whether a pooled model then predicts outbreaks better than one company's own records.
What to watch
- A pilot in which a university or other neutral party holds pooled records from competing food firms, and whether those firms actually contribute data.
- A common record standard for food safety data that smaller, paper-based firms can realistically meet.
- A model trained on pooled multi-company data that flags contamination or outbreaks earlier than any single company's records do.