A shocking report by a news agency in Belgium claims that contractors paid by Google
to transcribe audio recordings uploaded by Google Home
smart speakers have also been able to hear private conversations that should have never been recorded in the first place. The recordings run the gamut from bedroom conversations to professional phone calls containing private information.
We already know that Google thrives on data collection. That is how the company is able to tailor its services to users, including obviously its search engine, but also its AI assistant. In a related FAQ, Google explains that it "collects data that's meant to make our services faster, smarter, and more relevant, and more useful to you."
"Google Home learns over time to provide better and more personalized suggestions and answers," Google adds
, but you will not find is that its smart home speakers and accompanying smartphone app can lead to subcontractors listening to things you never intended to have recorded. Like having sex. There's no mention of subcontractors having access to that data at all, whether the recordings are intentional or not.
According to the VRT News report, most of the recordings subcontractors listen to are ones that are made consciously by Google Home users. However, some of the recordings were taken when the wake phrase "Okay Google" was clearly not given. This can happen if a smart speaker detects a phrase that sounds similar.
This leads to unintended recordings. VRT News examined 1,000 recordings shared by a subcontractor, 153 of which were accidental audio captures. Google anonymizes the data by deleting the user name and replacing it with an anonymous serial number, but the actual recorded conversations sometimes reveal identifying information.
According to the report, Google acknowledged that it works with subcontractors to transcribe conversations, which is intended to improve its speech technology (speech recognition automatically generates scripts, and the subcontractor then double checks for accuracy, correcting where necessary). However, Google claims they only transcribe "about 0.2 percent of all audio fragments."