If you’re an Amazon Echo or Alexa (Amazon’s AI-based voice assistant) user, chances are that your voice queries are reportedly being heard, recorded and transcribed by thousands of Amazon workers around the world. Well, parts of your audio if not everything you tell Alexa. This exercise is carried out to enhance and train the voice-recognition capability that sometimes requires human intervention.
A report by Bloomberg states that Amazon has employed contractors around the world including Amazon’s own employees for this very purpose. Although Alexa has a self-learning system, its training needs a human role because understanding the human experience is something it can’t do on its own. The teams are based in Boston, Costa Rica, India and Romania. According to the report, they would listen up to 1,000 audio clips in every shift of a nine-hour workday.
Workers have to listen and interpret the voice commands and then annotate them so that the query can be learned by the AI. One of the anonymous workers shared his experience about searching for instances where ‘Taylor Swift’ was spoken and he had to annotate the query to indicate that it meant the musical artist. Some instances also included private moments while some sounded unsettling. There are internal chat rooms to share clips that require extra help in interpreting what the user meant. Workers would also share their experiences in the chat room to relieve stress whenever they came across upsetting clips.
Amazon doesn’t explicitly state that customer audio recordings are accessed by its employees. The only mention of this practice on the company website is mentioned as a means to “train our speech recognition and natural language understanding systems.” According to the report, the audio clips don’t carry the identity information of the customers. However, the identifiers do include the customer’s first name and product serial number.
In a statement to Bloomberg, Amazon said, “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.”
Users have grown to become sceptical about products that have a microphone embedded in them. Although it’s essential for a human to aid the AI in interpreting voice commands coming from customers, anonymising that data is essential to protect the user’s privacy. It’s equally important for companies to be transparent of how customer data is being used by them even while training its systems to serve the users better.
Apple employs humans to train Siri in voice recognition. The audio recordings don’t include any form of identifiable information and according to Apple, it’s stored for six months with a random identifier. The identifier is later removed and the data may be stored for a longer duration. Google can also listen to user voice commands from its Assistant but it doesn’t include any personal information.