Google, Harvard use AI to pinpoint foodborne illness sources in real time
- Google and Harvard have developed a machine-learning tool to identify "potentially unsafe" restaurants through anonymous user search data. FINDER, short for Foodborne Illness Detector in Real-time, aggregates queries indicative of food poisoning and connects them to a user's recent search for restaurants, according to results published in the journal Nature.
- The team tested the tool with the health departments of Las Vegas and Chicago, which used the data to initiate inspections. Engadget reported that more than half of all restaurants inspected by FINDER were deemed unsafe, compared to about 23% of those visited after traditional complaints. Chicago's existing system already mines Twitter for similar user complaints through a platform called nEmesis, but this tool outperformed it by 68%.
- "The FINDER approach is more robust than individual customer complaints, as it aggregates information from numerous people who visited the venue," the researchers noted in Nature. Their goal is to ultimately harness the technology to pinpoint foodborne illness faster and more accurately.
Google and Harvard are testing artificial intelligence (AI) to battle the growing numbers of foodborne illnesses as city governments — strapped with limited resources — struggle to keep up with health inspections. Advice to avoid such illnesses leans on home food preparation, but when going out, consumers have to trust that restaurants take all necessary precautions. That's where tech can step in.
Though still in its early stages, FINDER tries to bridge another gap between government oversight and the vast troves of information available online. Yelp, for example, has blossomed into a trove of user-generated data on where and when customers feel ill after eating out.
Currently consumers tend to blame the last restaurant they visited, who in turn file complaints for a venue that inspectors check to no avail. The Google and Harvard researchers call that system inefficient because in nearly 40% of reported illness cases, the second-to-last restaurant caused illness, not the most recent.
FINDER scoured Google searches in the two test cities, first for terms related to foodborne illnesses and then for location data related to foodservice establishments. Researchers note that the sample size remains small, but the results of this initial experiment indicate that restaurants of any risk level could be a culprit.
This technology should put restaurants of all shapes and sizes on guard. It gets smarter with time, with searches through any Google product. The more customers that visit a restaurant and subsequently search phrases such as "how to relieve stomach pain" or "food poisoning symptoms," the more likely it is that the health department could be on its way.
Inspectors in Las Vegas and Chicago looked at data from both systems blind, meaning they didn't know whether FINDER or a consumer complaint flagged the restaurant. Yet FINDER restaurants were more than three times as likely to be cited for safety violations, especially more critical ones. In the case of Chipotle's most recent outbreak in Ohio, for instance, FINDER could have sent inspectors to that precise outlet within days rather than weeks.
Health departments could likewise build more detailed reports of restaurants cited by potentially dozens of customers — not everyone bothers or even knows to call local officials about a related illness. In the report, researchers emphasize that FINDER could augment existing protocols in city governments, but add that the long-term cost benefits would likely outweigh high false-positive rates.
The ultimate goal: more accurate reporting that results in more worthwhile inspections. As this and other technology improves, restaurants should show extreme caution in handling even minor customer complaints that could link them to an outbreak. Of course, in many cases the issue originated earlier in the supply chain. FINDER doesn't seem to address that possibility, nor the chance that consumers could get their hands on this rich data mine, which could harm a restaurant's reputation without due process. The recent report, however, indicates that only governments would have access to the aggregated data.