AI hallucinations are reshaping day-to-day interactions in restaurants, shops and other customer-facing roles, as more people arrive armed with information generated by chatbots that is simply incorrect. Workers across multiple industries report a growing tension between what AI tools tell customers and what staff know to be true.
Madison, a server in New York City, routinely asks diners about allergies before taking orders. She describes close calls in which guests with a declared shellfish allergy nonetheless attempt to order dishes containing shellfish, confident that ChatGPT has assured them the meal is safe. When she warns them that a dish contains an allergen, some push back, insisting the chatbot disagrees. “You’re allergic to shellfish, and ChatGPT says there’s no shellfish in this, and we are telling you, yes there is,” she explains.
Wine Lists and Invented Bottles
At the Michelin-recommended Italian restaurant where she recently worked, Madison noticed increasing reliance on generative AI among patrons. Offers to fetch the sommelier are frequently declined in favour of consulting ChatGPT, and sommeliers have at times been dismissed from tables entirely. Some guests request bottles the restaurant does not stock, or wines that do not exist at all.
Madison says these AI hallucinations feed existing anxieties about wine, including the perception that sommeliers push customers towards pricier bottles. To avoid confrontation, she and colleagues soften the message, telling guests the venue carries a hyper-specific list that may not include a French wine the chatbot suggested, rather than stating outright that the recommendation was fabricated.
A Pattern Across Industries
The phenomenon appears widespread in customer service. The launch of Muse, Meta’s new AI agent, is expected to add to the problem and to the challenges facing staff who must handle unusual requests. Workers including cinema staff, baristas and Apple Store employees have all encountered customers asking for products or features that do not exist. On Reddit and X, customer service workers describe patrons who treat ChatGPT’s responses as definitive.
Jason, a river ranger working on federal lands in the Mountain West, reports a rise in visitors arriving with nonsensical AI-generated itineraries. He says colleagues have had to explain that generative AI had invented campsites or scheduled activities out of geographical order, leaving groups expecting to row upstream on some days and float with the current on others. Even with a ranger present to explain the errors, some visitors remain reluctant to abandon their AI-generated plans.
Source
Image: theverge.com
