WhatsApp regularly receives new features that either enhance or hinder the messenger. Often, these are smaller changes that can have a significant impact. In the era of AI hype, Meta also integrates many features related to Artificial Intelligence. This is precisely the case with AI stickers that can be created by users. However, it seems there are significant problems with this. WhatsApp AI sticker generator can create stickers from user inputs, but in some cases, it may generate biased stickers without user input, raising concerns.
AI Stickers on WhatsApp with Biases
Artificial Intelligence is on everyone’s lips. Virtually every company wants to offer AI features, even if they are often not yet fully developed. Meta is at the forefront in this area and keeps releasing one AI feature after another. In WhatsApp, an AI sticker generator is integrated. However, this one seems to have gone somewhat out of control – as discovered by The Guardian.
WhatsApp’s AI sticker generator creates stickers based on the inputs made by users. In certain areas, however, the AI goes too far and generates stickers based on biases that were not inputted so explicitly. For example, if a Muslim child from Palestine is supposed to be generated, the AI creates a child with a weapon. Even when “Palestinian” is inputted in the AI sticker generator, people with weapons are generated. This doesn’t happen when the same terms are inputted with “Israel.”
According to Meta, the operator of WhatsApp, this is said to be a mistake for which they apologize.
How Many Biases Does The AI Still have?
This is just one example that was discovered due to the current Middle East conflict. It is currently impossible to estimate how many biases Meta’s AI generator still has. Artificial Intelligence is trained based on a wide range of content, and these are often not neutral. Accordingly, more biases could emerge over time that WhatsApp’s AI sticker generator brings to light.