The robots are here, and they’ve got strong opinions about food. In a paper to be presented at the Association for Computing Machinery Conference on Computer and Communications Security on November 1, UChicago computer scientists showed that artificial intelligence can be used to generate fake Yelp reviews so convincing that users found them to be indistinguishable from, and just as useful as, human-written reviews. Even plagiarism detection software usually couldn’t spot the difference.

Using Recurrent Neural Networks (a machine learning technique) and a diet of real Yelp reviews, the team—graduate students Yuanshun Yao and Jenna Cryan and Neubauer Professors Haitao (Heather) Zheng and Ben Y. Zhao—“trained” a computer program to generate convincing fakes and swap in the appropriate menu items for each restaurant. For instance: “The food here is freaking amazing, the portions are giant. The cheese bagel was cooked to perfection and well prepared, fresh & delicious! The service was fast. Our favorite spot for sure! We will be back!”

While it’s nice to know that artificial intelligence enjoys a cheese bagel as much as the next guy, this type of software poses a serious threat to companies that rely on online reviews to attract and retain their customers—and wouldn’t take much time or expertise for a savvy cyber attacker to develop. “I want people to pay attention to this type of attack vector as [a] very real and immediate threat,” coauthor Zhao told Business Insider. Fortunately, the team discovered a telltale statistical sign of reviews created using Recurrent Neural Networks and developed an algorithm to detect it.

This story appears on p. 19 of the Fall/17 print edition with the headline "Everyone's a critic."