Protecting families from contaminated food and water
All fields are required
Protecting families from contaminated food and water
Posted in Our Blog on June 1, 2025
UK health agencies may have a new way to detect foodborne illness outbreaks.
That’s right. AI.
A novel use for a complex technology.
Most cases of foodborne illness go unreported. If the sickened person has a mild illness, they likely will not seek medical treatment.
When that person does not seek medical treatment, they are not diagnosed, and their sample is never tested for the major foodborne pathogens.
The illness is not reported.
Unfortunately for those who are more likely to experience more severe illness, it may take longer for the problem to be uncovered. Leaving them vulnerable to an infection that may be mild for some but life-threatening for others.
Whether you leave a Google review, Yelp review, or a post on your neighborhood social media page, there are countless places people go to for advice on restaurants.
While these are helpful to steer you toward a dining experience you will more likely enjoy, it can also be harnessed as data for the AI machine.
Places like Reddit, Facebook, and Instagram are also places where AI routinely data mines for many different purposes.
Mostly for advertising.
Why not train AI to help detect foodborne illness outbreaks?
That is exactly what some UK scientists are working towards. Starting with the Yelp platform.
Common foodborne illnesses like listeriosis, salmonellosis, E. coli infection, and others often have overlapping symptoms. Some of the more common symptoms of foodborne illness were used in the data set.
Symptoms like diarrhea, vomiting, and abdominal pain were included in the model.
Some symptoms were purposefully excluded, as they are not “sufficiently specific” to gastrointestinal illness. Symptoms like headache, fever, and respiratory symptoms were omitted.
So, how exactly do you train AI to detect foodborne illness outbreaks?
First, a human has to do the job.
Scientists collected over three thousand reviews containing a list of possible gastrointestinal related key words. Then, epidemiologists manually annotated them.
Yes.
Manually.
This human task is important to constructing the language model that the AI must learn to achieve this goal.
In addition to the initial learning curve, scientists anticipate certain initial challenges to this new AI technology.
The first being access to real-time data.
Data mining has been around for quite some time. Review sites, like Yelp, have been online for almost as long as people have been routinely accessing the world wide web. So, the data will be there.
But how quickly is the data available?
Some people falling sick after eating a meal or having a reaction during their dining experience may write a review right away.
Others may stew on it for a bit. Delaying information availability. If there is a subsequent delay in AI finding this data, a good amount of time may pass.
The data is still relevant in determining if a foodborne illness outbreak has occurred. But discovering this after-the-fact does nothing to help prevent additional illnesses from happening during the event.
General information can be found using this AI tool. But certain key epidemiological details necessary to track down a source are likely not possible using this tool.
For example, general information on the types of food people have eaten may be included. But the specific ingredients or other factors may not be captured.
Variation in spelling, misspelling, and the use of slang have also been identified as potential challenges. The grammar and spelling police do not hold much sway on these review platforms.
And let’s face it. Diarrhea is a difficult word to spell. Likely even more so if you are sick at the time of writing the review.
When people get sick, they often think it was caused by the last thing they ate.
In certain cases, with fast acting germs, this could be true. Some will cause symptoms within a few hours. But many cases of foodborne illness do not become symptomatic for a day or so after eating it. In some cases, several days.
For this reason, many people may blame their symptoms on a particular meal, when it was a different source entirely.
Inaccurate reporting can complicate attributing illnesses to correct sources.
UK Health Security Agency scientists believe that this type of AI information gathering could one day become routine and provide vital clue from uncaptured official data about possible sources and causes of foodborne illness outbreaks.
“We are constantly looking for new and effective ways to enhance our disease surveillance,” said Professor Steven Riley, Chief Data Officer, UKHSA. “Using AI in this way could soon help us identify the likely source of more foodborne illness outbreaks in combination with traditional epidemiological methods, to prevent more people from becoming sick.”
If this system works in the UK, this model (with key changes in language) could be adapted to any other country. Including the United States.
How do you feel about seeing AI detect foodborne illness outbreaks in the U.S.?
Beneficial?
Overstepping?
Prone to misinterpretation?
Information is simply that. Information. If foodborne illness investigators use it as a conversation starter with food businesses, as opposed to a strong arm, this technology could prove useful. Surprise inspections, after all, are a part of most health department jurisdictions.
If you’d like to know more about food safety topics in the news, like “Could AI Help Detect Foodborne Illness Outbreaks?,” check out the Make Food Safe Blog. We regularly update trending topics, foodborne infections in the news, recalls, and more! Stay tuned for quality information to help keep your family safe, while The Lange Law Firm, PLLC strives to Make Food Safe!
By: Heather Van Tassell (contributing writer, non-lawyer)