From: iTopnews.de
Published in cooperation with iTopnews.de
The smart Ray-Ban Meta smart glasses have been criticized because of possible data protection problems. According to a report in the Swedish newspaper “Svenska Dagbladet”, video data from the glasses is partly evaluated by human data annotators, including apparently very private content.
The report According to the report, the auditors work for an external service provider in Kenya. Their job is to analyze video clips and mark objects in the clips so that Meta’s AI systems can be better trained.
The AI function sends the video for analysis
The smart glasses can either record videos manually or use them for AI functions. For example, users can ask a question about an object they are currently viewing, such as a product, a building, or a vehicle. So that the AI can respond, the images are transferred to meta servers and evaluated there.
A report now suggests that some of these recordings will also be reviewed by humans to improve the AI systems.
Employees report very private scenes
According to the report, several auditors say that extremely sensitive content appears in the clips. These include photos from bathrooms, intimate situations or videos in which personal items such as credit cards can be seen.
An employee explained that some of the recordings apparently come from private homes and could have been taken in everyday life – for example when users wear the glasses permanently and accidentally record very personal moments.
Meta does not reveal any specific information
It remains unclear when exactly and for how long video recordings will be transferred to Meta. For example, if users ask the AI about an item, it is not transparent how long the system will continue to transmit or store images.
According to the report, when asked, Meta simply referred to its terms of use. It generally states that interactions with AI systems can be automated or checked manually by humans.
Former employees point out that sensitive content should actually be filtered out. However, the automatic systems do not always seem to recognize these reliably.
About the Author
