Google responds to leak of audio data from AI assistant reviewer

A language reviewer under contract to Google to analyze samples of voice data collected by its Google Assistant software has leaked more than 1,000 sound recordings of users, putting its biometric and artificial intelligence training practices back in the spotlight, and putting the company on the back foot.

Belgian news site VRT published a report earlier this week on Google’s practice of hiring contract workers around the world to listen to and annotate recordings from devices like Google Home speaker. The story was spurred by one such contractor providing leaked voice data from Dutch customers, which included private information such as home addresses and discussions about medical conditions.

Google responded with a blog post, explaining that only about 0.2 percent of all audio snippets are listened to and transcribed by its contractors. The company says it has a number of measures in place to protect user identities, and that its human reviewers help digital assistants “understand the nuances and accents of a specific language.”

The company acknowledged the leak, and explained the process of collecting audio data, as well as its security response, to reassure the public.

“Our Security and Privacy Response teams have been activated on this issue, are investigating, and we will take action,” Google Product Manager for Search David Monsees wrote in the response. “We are conducting a full review of our safeguards in this space to prevent misconduct like this from happening again.”

Amazon was caught up in a similar controversy earlier this year when it was revealed that the company employs workers to listen to and annotate recordings for training purposes. Both the public and the companies providing digital assistants seem to be caught off guard by predictable developments, such as the recent BIPA lawsuit against Amazon for biometrically processing voice samples in Illinois.

The Verge points out that both Amazon and Google allow users to delete audio data, and Google allows users to turn off audio storage completely, though some data collected by Amazon remains available to the company.

Article Topics

artificial intelligence  |  data collection  |  data protection  |  Google  |  privacy  |  speech recognition  |  training  |  voice biometrics