Will AI mimicry ruin online user-generated reviews?
Researchers at the University of Chicago have trained a neural network, or artificial intelligence (AI) system, to write fake reviews on Yelp and have found the AI-generated reviews to be virtually the same as those written by humans.
The AI-software learned how to mimic writing a review from publicly-available Yelp restaurant reviews. A customization process, which included feeding in details on specific restaurant dishes, tailored the review for a specific restaurant.
Among the five-star reviews the AI-system came up with for a New York City buffet restaurant:
- “My family and I are huge fans of this place. The staff is super nice and the food is great. The chicken is very good and the garlic sauce is perfect. Ice cream topped with fruit is delicious too. Highly recommended!”
- “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!”
The AI-generated reviews were found to be “effectively indistinguishable from those produced by humans” by test subjects and were rarely identified by plagiarism detection software. On usefulness, the test subjects gave an average score of 3.15 for AI reviews versus 3.28 for genuine reviews.
The researchers said AI-generated reviews will only become more sophisticated and are a bigger threat than those written by humans because automated reviews can be done rapidly and don’t require monetary-compensation.
Beyond reviews, the researchers said the expansion of AI-generated content will increasingly cause society to question what’s real or fake on platforms such as Twitter and online discussion forums. The researchers said in the study, “We hope these results will bring attention to the problem and encourage further analysis and development of new defenses.”
In a statement to The Verge, Yelp discounted the findings because the study focused on text. Yelp stated, “Yelp’s recommendation software employs a more holistic approach. It uses many signals beyond text-content alone to determine whether to recommend a review.”
- Automated Crowdturfing Attacks and Defenses in Online Review Systems – University of Chicago
- Researchers taught AI to write totally believable fake reviews, and the implications are terrifying – Business Insider
- AI trained on Yelp data writes fake restaurant reviews ‘indistinguishable’ from real deal – The Verge
DISCUSSION QUESTIONS: Are AI-generated reviews a bigger threat to the credibility and overall value of online user-generated reviews than human-written fake reviews? Will plagiarism detection software and other defensive measures be enough to offset increasingly sophisticated AI-reviews?