Health officials in Illinois turned to an AI chatbot to try to solve a puzzling outbreak linked to a county fair. But whether it was actually helpful or not remains unclear.
According to a report this week in the Morbidity and Mortality Weekly Report, officials in Brown County got the first hint of an outbreak from the county sheriff, who noted on August 5, 2024 that a remarkable number of potential jurors for an upcoming trial said they had a stomach bug. Then, on August 12, the state health department notified the county of a case of Salmonella enterica serotype Agbeni.
With those two tips, county health officials opened an investigation and were able to identify 13 cases—seven laboratory-confirmed cases of S. enterica Agbeni and six probable cases that were in close contact with confirmed cases. The cases spanned five counties, but they all had one thing in common: everyone had gone to the Brown County fair.
The annual event takes place in a close-knit rural community of about 4,200 residents. But it’s popular, drawing an estimated 36,000 people from surrounding areas each year. The 2024 fair spanned July 30 to August 4, so it was packed up and gone by the time investigators figured out the connection to the illnesses. Still, they wanted to find the source and started with the hypothesis that it was a food vendor—Salmonella species are generally known for causing food poisoning. The bacteria dwell in the innards of wild and domestic animals, including poultry and cattle, and can contaminate food in a variety of ways, causing outbreaks of gastrointestinal illnesses (vomiting, diarrhea, stomach cramps).
While nine of the 13 affected individuals said they ate food from at least one vendor at the fair, the remaining four had not eaten at all—which seemed to rule out food. The investigators noted that there were obvious hygiene concerns at the fair, which had portable bathrooms and a “limited” number of portable handwashing stations. Ten of the sickened people admitted to not washing their hands. But, the only thing they all had in common was that they all drank a cold, canned beer from the fair’s sole beer tent.
According to the MMWR report, the fair’s beer supply in the single tent was kept cold in a large makeshift cooler, described as being made of “a 10-ft length of non-food-grade corrugated black plastic farm drainage tile with four internal compartments.” It was reportedly hosed off at the start of the fair, but then never fully drained or cleaned again. It was simply refilled daily with ice—made from municipal tap water—as melting occurred and it was partially drained. The people who worked the beer tent handled the beer and ice with their bare hands.
An AI assist?
The author of the MMWR report, county health official Katherine Houser, noted that the beer-tent workers were hesitant to give details because they didn’t want to get any of their community members in trouble. But one let slip that someone had put leftover food in the cooler overnight at the start of the fair.
The county health officials hypothesized that the cooler had become contaminated with Salmonella that spread to beer cans from which people then drank, allowing for infection. But with the makeshift cooler gone, it would remain only a hypothesis. So, the health investigators then turned to ChatGPT for assurances.
After providing the chatbot with details of the outbreak, health investigators asked it several questions, including: “Will S. Agbeni grow in an improperly drained cooler?”; “Are any other sources, other than ice, likely if only canned beverages and no foods were available at this location?’ ; and “What examples of similar outbreaks have been documented in scientific literature?”
Some of the questions are easy enough to answer without a chatbot. A simple search on PubMed, a federal database of scientific literature, quickly pulls up examples of Salmonella being found in ice, for example. But, the chatbot assured the officials that the cooler was a “credible and likely” source of the outbreak and they stuck with the hypothesis.
In the end, the officials required new cooler sanitation protocols—and concluded that the AI assistance was helpful. “AI was effective in this rural setting for rapid situational awareness,” Houser wrote. However, she also acknowledged the potential concerns of using AI for outbreak investigations: “Given the inherent limitations of generative AI tools, including potential inaccuracies and lack of source transparency, all AI-generated summaries were critically reviewed and validated against primary literature before incorporation,” she wrote.
Overall, the case report has a murky ending. It’s unclear how helpful the chatbot actually was in this case. Critically reviewing AI-generated answers can easily take as much time as simply researching the answer on one’s own. And of course, we’ll never know for certain what was really going on in that makeshift beer cooler—though the new cooler sanitation protocols seem like a good idea, regardless.
