GoKawiilSign up to see the future, today Can’t-miss innovations from the bleeding edge of science and tech Email address Sign Up Thank you!
Planning your next getaway? You’d be wise not to trust the advice of an AI chatbot — even if they’re explicitly designed to draw on the testimonies of real travelers.
As The Guardian reports, an AI tool from Tripadvisor designed to summarize the reviews of hotels and other locations on its platforms routinely glosses over some of the horrifying complaints that guests have made about their stays.
The AI described one hotel in Cape Verde as “spotless,” failing to mention that it was being sued for mass food poisoning, according to the findings of the consumer campaign organization Which?, reported in the newspaper.
And for a resort in Turkey facing multiple complaints of sexual harassment, the AI waxed enthusiastic about its “friendly service,” merely alluding to those serious allegations as “lapses [in service] noted by a few.”
“The platform has a responsibility to revisit the accuracy of its AI summaries and AI chatbot,” Rory Boland, the editor of Which? Travel, said in a statement, per The Guardian. “In the meantime, users should scroll past these summaries and look at guest reviews, particularly one-star ratings, and at reviews on other sites, to make sure their next stay is a safe one.”
It’s another example of how AI summaries, when they’re not outright spewing misinformation, frequently fail to give the complete picture. And it’s more egregious when you consider that the Tripadvisor tool’s purpose is to give a roundup of what other guests have said, as opposed to a general purpose chatbot that may draw from more generic information on the broader web, a lot of which may be marketing copy from the hotels and resorts in question.
In other words, you’d hope Tripadvisor’s AI would cut through the BS, which is the whole appeal of the customer reviews it’s trained on in the first place. We all jump to the one star ratings to see how bad a place can get, and make a judgment call on whether they’re truly representative of the average person’s experience.
But it seems the chatbot isn’t doing that, which is symptomatic of a problem with the tech at large. According to Duncan Brumy, a professor of human-computer interaction at the University College London, AI tends to “sanitize and rub off the edges” of harsher viewpoints, which is likely a consequence of how most of its training data contains bland observations, he said.
“Here you have guests describing a really negative experience, but the AI has decided to tone it down,” Brumby told The Guardian. “It’s as if it’s being polite.”
... continue reading