For several days, US media has been having fun testing out one simple trick – typing in “As an AI language model” into search engines. This expression is frequently used by ChatGPT, a popular generative artificial intelligence, when it is unable to answer questions, and with it, you can easily find examples of fake comments or messages on the web. But, upon closer inspection, these cases are few and far between. Despite some attempts in French, like an Amazon review praising a connected watch, the threat of a fake comment invasion on major e-commerce platforms appears to have been overstated.
In order to test how far generative AI has come in content creation, innovation manager Ari Kouts of the consulting firm Viseo conducted a full-scale experiment, using ChatGPT to create two completely automated websites. One website was dedicated to recipes and the other relaying technology news from major US outlets. Surprisingly, the websites generated hundreds of articles that even included subtle nuances and those which seemingly wrote with their own style, mimicking a chef or journalist’s personality.
Though it is unknowable if similar developments have been made privately and not shared, Kouts confirmed that it is possible to generate entire blogs for a minor cost and in a few hours. That being said, generative AIs have their limitations, as exemplified in a difficulty when it comes to reviewing new products, audiences that require more research and factual component than a generative AI can provide.
When it comes to those producing fake reviews, Brett Hollenbeck, an assistant professor and author at the University of Los Angeles School of Management, shares that generative AI has little use. The value of a fake review lies in the review being posted by a real customer, and this is something that AI lacks.
It is clear that while ChatGPT and other content creation programs are real, they have limited applications of use to malicious third parties. The occasional prank or harmless user may find some amusement in these programs, but as of now, there is unlikely to be a flood of “AI-generated” posts on the web.
