AI set to flood web with fake business reviews
Machine-written reviews could make online comments untrustworthy, researchers show.
You’re looking for a restaurant for dinner, so turn to Yelp for help choosing. Here’s the first review: “I love this place. I have been going here for years and it is a great place to hang out with friends and family. I love the food and service. I have never had a bad experience when I am there.”
Sounds ideal, but that review was written by a machine, not a human customer. Researchers from the University of Chicago teamed up with Yelp to train their neural network on the site’s existing, human-written reviews — and the results suggest artificial intelligence could easily write content that can fool us. “As a tool, everything comes with a positive side and a negative side,” Ben Zhao, one of the authors and a professor of computer science, told PC Pro.
The paper ( pcpro.link/278yelp) raises issues of trust online – although there are plenty of fake news and dishonest reviews written and posted by humans. Switching away from hiring low-paid people to churn out reviews to all-but-free neural networks could spread the practice even further, said Zhao. “Sites like this already have lots of jobs where you can create fake reviews, handwrite them, but they tend to be quite expensive,” he said.
“There’s also a timing component,” he explained. Post an ad on a site such as Mechanical Turk asking its jobby-job workers to churn out fake reviews for you – either positive about your business, or negative about a rival’s – and they’ll jump on it immediately, suddenly swamping a business with dozens of comments all at once, where they might have had only a handful before. “That’s about the only feature that’s usable to detect these attacks,” Zhao said, adding that an automated system could dole out the reviews in a more subtle manner, making fake reviews even harder to defend against.
It’s possible to spot neural network reviews, he notes, as the language does differ from how humans write. “We’d look for the tell-tale signs that something has been disturbed, that something’s anomalous compared to normal English language,” Zhao explained, although as neural networks improve that will be harder to detect.
What does that mean for trust online? Chris Brauer, director of innovation at the Institute of Management Studies at Goldsmiths College, University of London, said this could further undermine trust in crowdsourced content, already eroded by those human-written fake reviews. “It’s a challenge for those platforms, in terms of their value in the market,” he told PC Pro, adding that taken a step further it’s clear such systems could write fake news or comments, further muddying our already confused online discourse.
Fake reviews could be particularly difficult for small businesses that depend on user recommendations rather than buying advertising to get new customers. “That puts you in a position where you’re highly susceptible to being undermined by one of these attacks,” Brauer said.