Meta's AI glasses send sensitive video footage to experts in Kenya

Meta Ray-Ban smart glasses are at the center of a large-scale ethical scandal. It turned out that the corporation transfers fragments of user video recordings to third-party contractors in Kenya for “manual” training of AI algorithms. Despite promises of a high level of protection, hundreds of hired moderators have gained access to the private lives of gadget owners.

How does “training” of a neural network work

class=»notranslate»>__GTAG7__According to internal reports made available to specialized media, writes xrust, the Meta artificial intelligence system is not always able to independently interpret complex visual scenes. To “label” data, the company uses the Sama platform, whose offices are located in Nairobi.

Kenyan experts watch thousands of short videos, tagging objects and actions. The main problem is that the video often captures:

• faces of random passers-by;
• interiors of private apartments;
• computer screens with passwords and personal correspondence.

Expert reaction and privacy issues

Human rights activists from the organization Privacy International have already expressed concern. In their opinion, users were not properly informed that real people would be watching through their “eyes.” Meta representatives say that the data is anonymized and moderators are strictly selected, but cybersecurity experts remind about the human factor and the risks of leaks.

Summary

This is not the first time that tech giants have used cheap labor from African countries to train AI. Previously, OpenAI (the creator of ChatGPT) was criticized for using Kenyan moderators to filter out toxic content. The current Meta incident is complicated by the fact that data collection occurs through a wearable device that records almost continuously in everyday life.

Xrust Meta AI glasses send confidential video footage to experts in Kenya