A report by the Belgian public broadcaster VRT NWS revealed how subcontractors paid to transcribe audio clips collected by Google's AI Wizard can end up listening to sensitive user information, including their names, addresses, and details about their personal lives.
This is the latest story that shows that our interactions with AI assistants are not as private as we might like them to be. Earlier this year, a report from Bloomberg revealed similar details about Amazon's Alexa, explaining how audio clips recorded on Echo devices are sent without users being privy to human subcontractors, who have transcribed what is being said so to improve the company's AI systems.
Even worse, these audio clips are often fully recorded by accident. Usually, artificial intelligence assistants such as Alexa and Google Assistant only start recording audio when they hear their waking word (eg, "Ok Google"), but these reports show that devices often start recording by mistake.
In the history of VRT NWS, The broadcaster reviewed about 1,000 recordings, 153 of which were accidentally captured. One contractor told the publication that he was transcribing about 1,000 audio clips from Google Assistant every week. In one of the clips that he watched, he heard a distressed female voice and said that he felt that "physical violence" had been involved. "And then it becomes real people that you listen to, not just voices," said the entrepreneur.
You can see more in the video report below:
Technology companies say that sending audio clips to human beings to transcribe is an essential process to improve their voice recognition technology. They also point out that only a small percentage of the records are shared in this way. A Google spokesman said wired that only 0.2% of all recordings are transcribed by humans and that these audio clips are never presented with user identification information.
According to Michael Veale, a privacy researcher at the Alan Turing Institute in London, these obfuscations could pose legal problems for society. He said wired this level of disclosure may not meet the standards set by the EU's GAR regulations. "You have to be very specific about what you are implementing and how," Veale said. "I think Google did not do it because it would look scary."
We solicited comments from Google and will update this story if we hear more.