Most of the time, when you talk to an Amazon Echo device, only Amazon’s voice-recognition software is listening. But sometimes, Bloomberg reports, a copy of the audio is sent to a human reviewer at one of several Amazon offices around the world. The human listens to the audio clip, transcribes it, and adds annotations to help Amazon’s algorithms get better.
“We take the security and privacy of our customers’ personal information seriously,” an Amazon spokesman said in a statement emailed to Bloomberg. “We only annotate an extremely small sample of Alexa voice recordings in order [to] improve the customer experience.”
Bloomberg hints at a significant workforce doing this kind of work. Bloomberg says Amazon has employees listening to audio clips in offices in Boston, Costa Rica, India, and Romania. Employees interpret as many as 1,000 audio clips in a 9-hour shift.
It’s not hard to see why Amazon would want to have some audio clips reviewed by human beings. There’s probably no substitute for having a human being listen to clips and verify that the software is interpreting them correctly.
But Amazon could have been more explicit about the role of human reviewers. “We use your requests to Alexa to train our speech recognition and natural language understanding systems,” Amazon says in its frequently-asked questions page for Alexa—without mentioning the human element.
Amazon told Bloomberg that it has strict privacy safeguards in place to prevent misuse of the system. “Employees do not have direct access to information that can identify the person or account as part of this workflow,” the spokesman said. “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.”
Employees occasionally hear unexpected or extreme audio, Bloomberg reports. In one instance, two workers heard what sounded like a possible sexual assault but were told that it wouldn’t be appropriate to intervene. Employees can discuss what they hear with other employees in an internal chat service, and they can share clips they have trouble interpreting—though the report also mentions files being shared simply because they are “amusing.”
Bloomberg says Apple and Google also have human reviewers
Bloomberg says that Apple’s Siri “also has human helpers.” The company points to an Apple privacy white paper that describes how Apple uses audio captured from customer devices.
“User voice recordings are saved for a six-month period so that the recognition system can utilize them to better understand the user’s voice,” the white paper says. “After six months, another copy is saved, without its identifier, for use by Apple in improving and developing Siri for up to two years.”
What about Google’s virtual assistant? “Some reviewers can access some audio snippets from its Assistant to help train and improve the product,” Bloomberg reports. “But it’s not associated with any personally identifiable information and the audio is distorted, the company says.”