A team of actual living people listen to selected Alexa commands as part of a process intended to help Amazon learn more about what people are asking its virtual assistant and improve responses.
While it's easy to get outraged about something like that, the reality is much more mundane and it's surely obvious that Amazon would have had to use real people to improve a service that relies on ever-more human-like responses to human commands.
The program is called the Alexa voice review process and it has come to light courtesy of Bloomberg, who have spoken to a bunch of people who work on the team.
Recordings are transcribed, annotated – the reviewer notes how it compares to the interaction that Amazon would expect from the situation. Other conversations in the room are also noted down. If there is anything sensitive that is picked up, the reviewer has a "critical data" box to tick on the form.
Then they can be fed back into the team programming Alexa – ostensibly to improve Alexa's understanding in areas where it isn't strong already. The process of collecting recordings is random. Remember that recordings are only retained by Alexa when you've uttered the wake word – although accidental activations are also noted down.
The team is located across the globe and is a mix of Amazon staff and contractors who have to sign non-disclosure agreements and listen to as many as 1,000 clips of Alexa commands each day. They ask colleagues for help when they don't understand particular commands.
We'd imagine most of this work must be fairly tedious, but there must be some amazingly funny clips as well – the team cites examples such as a woman singing off-key in a shower. But there are obviously some more worrying recordings as well which can upset – Amazon apparently has processes in place for team members who hear distressing audio.
“We take the security and privacy of our customers’ personal information seriously," said an Amazon spokesperson in an emailed statement to us. "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.”