Pink cheeks, a hat on his head and… an assault rifle in his hands. This is the representation offered by the WhatsApp “sticker” function – an option that allows you to create drawings to integrate into a conversation – when the words “Palestinian”, “Palestinian” or even “Palestinian Muslim child” are mentioned in the search bar. . Results verified by the British newspaper The Guardian.
Powered by generative artificial intelligence created by Whatsapp, the new “sticker” function aims to create “tailored” designs, based on Internet users’ requests. Currently being implemented, it is not accessible to all French users. It was subject to abuse since its implementation in early October.
The Guardian details its own evidence, with the words “Palestinian Muslim boy.” Four AI-generated stickers appeared; one showed a boy with an AK-47.
A similar result was observed with the word “Palestinian”: two women appear with a hijab – a veil worn by some Muslim women – next to a man in a white shirt. Along with them, WhatsApp offers the representation of an armed, child-like soldier, whose suit (green and brown) recalls the colors of Hamas terrorists.
Another observation from The Guardian: when the “Israeli army” was mentioned, no weapons appeared in the four search results proposed by AI. Drawings of soldiers with helmets were thus offered.
Meta employees took the problem to their superiors, The Guardian reports. Presented at the Meta Connect 2023 conference on September 27, Emu (by Expressive Media Universe) is the generative artificial intelligence developed by the Meta group, owner of Whatsapp.
Asked by The Guardian about the issue, Kevin McAlister, spokesperson for Meta, assures that the company is “aware of the problem” and that it is “trying to resolve it.” “We mentioned it when we launched the feature: the search results provided by AI may be inaccurate or inappropriate, as is the case with all generative AI systems,” the company defends.
Source: BFM TV
