Researchers Develop an AI That Can Detect Unsafe Food on Amazon Quicker than FDA
Artificial Intelligence is evolving at a rapid pace and finding applications in a lot of fields for providing innovative solutions or improving the existing solutions with increased efficiency. This time, a team of researchers at the Boston University School of Public Health have come up with an AI that can detect food which is not fit for consumption.
The AI does not magically determine the quality of food, it does so by going through reviews posted by customers on Amazon. The researchers have trained deep learning AI called Bidirectional Encoder Representation from Transformations (BERT) for detecting negative reviews.
The study was published earlier this week in the Journal of the American Medical Informatics Association where the researchers explained their approach for training BERT. According to the journal, the researchers collected 1,297,156 reviews of food products which included 5,149 of the food products that had been banned by the Food and Drug Administration during 2012 – 2014.
The researchers then sorted the reviews and labeled them into categories like “sick”, “rotten”, “label” – the same terminologies used by the FDA. With all these details, BERT was able to identify the recalled products with 74 percent accuracy. In addition, the AI also identified over 20,000 potential unsafe products which have not been recalled by the FDA so far.
I do have some doubts regarding the reliability of this AI as competitors could conduct campaigns by hiring people to post negative reviews which in turn will take a hit on the accuracy of the AI. Also, the AI lacks emotional intelligence and hence it can’t differentiate if the product review was actually genuine or posted out of frustration.
While this AI cannot be completely depended on recalling food products, it will definitely help government organizations like the FDA to improve their efficiency as they will have a list of flagged products which they need to review with more priority rather than approaching all the products with traditional methods.
So, what do you think of this project? Share your thoughts with us in the comments.