Thousands of Amazon employees are listening to your Alexa conversations

Alexa is installed on over 100 million devices around the globe, acquiring billions of user queries in the process. Like all digital assistants, Alexa relies on data models to offer better responses, but it looks like Amazon also gave the digital assistant a helping hand in the form of human reviews. An investigation by Bloomberg has revealed that Amazon has a global team that transcribes and annonates Alexa recordings from around to world to "eliminate gaps in Alexa's understanding of human speech" and allow the assistant to serve up better responses to your queries.

The publication found that Amazon employs thousands of people around the world — both contractors and full-time employees — to review Alexa recordings, with teams spread out across Boston, Costa Rica, India, and Romania. According to an unnamed source, each reviewer parses over 1,000 audio clips over a nine-hour shift, annotating them and feeding them back to the system to improve Alexa's responses. From Bloomberg:

One worker in Boston said he mined accumulated voice data for specific utterances such as "Taylor Swift" and annotated them to indicate the searcher meant the musical artist.

Occasionally the listeners pick up things Echo owners likely would rather stay private: a woman singing badly off key in the shower, say, or a child screaming for help. The teams use internal chat rooms to share files when they need help parsing a muddled word—or come across an amusing recording.

The publication also found that employees come across recordings that are of a disturbing nature:

Sometimes they hear recordings they find upsetting, or possibly criminal. Two of the workers said they picked up what they believe was a sexual assault. When something like that happens, they may share the experience in the internal chat room as a way of relieving stress.

Amazon says it has procedures in place for workers to follow when they hear something distressing, but two Romania-based employees said that, after requesting guidance for such cases, they were told it wasn't Amazon's job to interfere.

The recordings don't have a user's full name or address, but they do include the user's first name, an account number, and the device's serial number. Amazon has previously mentioned that it relies on natural language processing to train Alexa, but it has admitted to Bloomberg that it uses a human element to annotate a "small sample of Alexa voice recordings:"

We take the security and privacy of our customers' personal information seriously. 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.

It's not just Amazon that's turning to humans helpers to develop its digital assistant. Bloomberg found that Apple also has a human team that checks whether Siri's interpretation of requests matches what was asked by users. Google has reviewers that train Assistant, but the clips don't have any personally identifiable information and the audio itself is distorted to prevent any identification.