According to Bloomberg News Amazon employees around the world listened to recordings from smart speakers as part of improving its machine learning in assisting customers. Like Apple's Siri and the Google Home devices, Alexa is a digital assistant built into several Amazon devices. Amazon does make it clear that it uses data "to train our speech recognition and natural language understanding systems", with users given the chance to opt out. But it does not explicitly say that training involves workers in the United States, India, Costa Rica and other nations around the world.
The report said Alexa's auditors do not have access to customers' full names or address, but they do have serial numbers and account numbers associated with the device. "We take the security and privacy of our customers' personal information seriously," Amazon said in a 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. We have strict technical and operational safeguards and 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."
The company says on its website that Alexa devices only send recordings back to Amazon servers if the device hears a "wake word" like "Alexa", "Echo" or "computer". When the wake word is detected, the light ring at the top of the device turns blue, indicating it is recording and sending the command to the server. But devices can be fooled by ambient, local noise, like loud television sets, and regardless of whether the activation of the device is mistaken or not, the Amazon reviewers are required to transcribe the recording. When the reviewers hear private information that has been accidentally recorded, they are meant to flag it as critical data and move on to the next recording.
It is not the first time Alexa devices have copped flack over what it records after a family in the United States saw a private conversation mistakenly recorded and sent to a random contact. A family member said she received a call from an employee of her husband who urged them to unplug their Amazon Echo devices, which use the Alexa software. Amazon described the event as an "extremely rare occurrence", saying in a statement that it believed parts of background conversation were mistakenly heard by the device as "Alexa", a wake word. The company also promised to fix a bug in its virtual assistant after reports the device was randomly laughing at people.
