There are a lot of people all over the globe who employ the use of smart speakers in their homes and Amazon’s Alexa is by far a favorite, even in spite of numerous other people raising their voices about privacy concerns. And by the looks of it, they are not wrong – someone, sometimes is, indeed, listening.
A report from Bloomberg states that Amazon does employ people all around the world, from Costa Rice to India and Romania, whose job is to improve the digital assistant by listening in to voice recordings that have been captured in both the Alexa owner’s offices as well as in their own homes.
These recordings are not only listened to but also transcribed and fed back into the software in order to get Alexa to better understand human speech and respond to commands more efficiently.
The employees who take on this job signed non-disclosure agreements who do not allow them to speak freely about the program and usually work in buildings that do not advertise the Amazon logo in any way, shape or form.
The 9-hour work day is repetitive and every reviewer goes through around 1,000 audio clips on every shift. One reviewer from Boston, according to Bloomberg, has spent a few good solid hours gathering up voice data for the specific ways people pronounce Taylor Swift’s name and annotated them to let the algorithms know that the user meant to search for the artist.
But the reviewers often listen in to things the Alexa owners would rather keep private and it’s worth noting that, according to the report, the teams make use of internal chat room to share the audio files they might find amusing.
There’s also cases where the recordings are upsetting or criminal in nature and, according to two Romanian reviewers, when happened to come across these sort of materials and reported them, asking for guidance, Amazon’s reply was that it was not the company’s job ‘to interfere’.
“We take the security and privacy of our customers’ personal information seriously,” an Amazon spokesperson told Bloomberg in an emailed statement “We only annotate an extremely small sample of Alexa voice recordings in order [to] improve the customer experience. For example, this information helps us train our speech recognition and natural language understanding systems, so Alexa can better understand your requests, and ensure the service works well for everyone.” he continued to say that they have “a zero tolerance policy for the abuse of our system. Employees do not have direct access to information that can identify the person or account as part of this workflow. All information is treated with high confidentiality and we use multi-factor authentication to restrict access, service encryption and audits of our control environment to protect it.”
Bloomberg states that, according to a screenshot they’ve seen, the recordings the reviewers go through do not provide the user’s full name or address but they do show the account number, the user’s first name and the device’s serial number – which should be more than enough to raise a few concerned eyebrows.
Some Alexa reviewers simply transcribe the commands they hear and compare the interaction between the user and the software but others hear everything that is being spoken, including sensitive data, such as bank details. When that happens, the reviewers tick a box that marks the recording as ‘critical data’ and then continue to the next clip.
The reviewers amusingly note that, sometimes, they receive recordings of people asking Alexa “Do you work for the NSA?” or “Alexa, is someone else listening to us?”
One of the reviewers has stated that they transcribe around 100 recordings of conversations daily that happened even when Alexa did not receive its wake command or was somehow turned on accidentally.
Amazon states that the device does not store any audio unless it is turned on by pressing a button or hears the wake word but by all accounts, Alexa starts recording at random anyway, according to the reviewers interviewed by Bloomberg – they might receive a clip of a TV blaring out for example or just noise that doesn’t make sense. In that case, they mark it as such and continue to the next recording.