Key Takeaways
- The UK Health Security Agency is exploring AI to analyze restaurant reviews for food poisoning sources.
- Researchers assessed large language models’ effectiveness in identifying gastrointestinal illness symptoms from thousands of reviews.
- Challenges included accessing real-time data and accurately determining illness sources due to language variations and misattributions.
AI and Foodborne Illness Detection
The UK Health Security Agency (UKHSA) is investigating the potential of artificial intelligence (AI) to analyze online restaurant reviews for identifying sources of food poisoning. In a recent study, researchers focused on the capabilities of various large language models (LLMs) to detect symptoms related to gastrointestinal (GI) illnesses by scrutinizing thousands of online reviews. The aim is to enhance disease surveillance effectiveness and improve the response to foodborne illness outbreaks.
During the study, researchers evaluated the LLMs on their ability to recognize symptoms such as abdominal pain, diarrhea, and vomiting, along with the types of food consumed by restaurant guests. UKHSA believes that this innovative approach could significantly augment the existing data on GI illnesses, which current systems often overlook, ultimately aiding in pinpointing outbreak sources. Steven Riley, UKHSA’s chief data officer, emphasized, “We are constantly looking for new and effective ways to enhance our disease surveillance.”
This initiative is part of UKHSA’s broader exploration of AI applications in public health. Previous efforts have focused on identifying illness outbreaks using online reviews, but the latest study involved a more thorough analysis of symptom-related language and terminology to fine-tune the identification of food poisoning sources.
To conduct the research, epidemiologists manually labeled over 3,000 restaurant reviews after filtering them based on a comprehensive list of GI-related keywords. This filtering process was aimed at isolating relevant symptoms while excluding general ones like headache and fever, which are not specific to GI disorders.
Despite the promise of AI, researchers faced key challenges, particularly in accessing real-time data. While LLMs provided valuable insights into the types of food that could be linked to illness, determining specific ingredients involved proved difficult. Additional challenges arose from variations in spelling, the use of slang, and the common issue of patrons misattributing their illnesses to particular meals or establishments.
Riley acknowledged that further refinement is essential before these AI techniques can be integrated into routine foodborne illness outbreak responses. This highlights the need for continued research and development to enhance the effectiveness of these innovative methods.
In related news, NHS England announced in March 2025 the rollout of AI software by Cera, designed to detect symptoms of prevalent winter illnesses, including Covid, flu, RSV, and norovirus, marking a significant advancement in leveraging AI for public health diagnostics.
The content above is a summary. For more details, see the source article.
