Ever since the controversy arose surrounding Google training its voice recognition technology using human contractors, barely a week goes by without another company admitting to the same. While the practice doesn't come as a surprise to many familiar with the underlying technology, no company thought of clearly stating what they were doing, which is not a good idea considering rising privacy awareness among customers and regulators. Bloomberg found out that the same is true for Facebook Messenger, affecting users who turned on voice-to-text for audio messages.
Following an investigation by Bloomberg, Facebook admitted that it's employing contractors to manually transcribe some of its voice-to-text messages. Only people who explicitly turned on this feature are affected. The problem is, Facebook doesn't state anywhere that humans are involved in the transcriptions. When users opt in to the feature, a prompt only warns them that machine learning is used, without mentioning human listeners anywhere – leading them to believe that their spoken conversations stay private (well, as private as they can be without end-to-end encryption).
Experts say that it's reasonable to expect that humans are still helping machines to understand us better. The issue is that almost no company clearly discloses it's happening. US lawmakers are already calling for more regulation, with some lawyers hinting that the lack of communication might come into conflict with a settlement between Facebook and the FTC, which covers Facebook's usage and sharing of private data.