Roger Fingas
Some voice snippets uploaded by Google Assistant are being reviewed by human contractors, as with Amazon Alexa, and can potentially include sensitive conversations, reports say.
Belgium's public broadcaster, VRT, recently gained access to over 1,000 audio files from a Google contractor tasked with reviewing Assistant audio, according to Wired. The audio was captured from devices such as phones, smartspeakers, and security cameras.
Google has acknowledged anonymously transcribing 0.2% of uploads to improve its technology, but instances in which Assistant is triggered accidentally appear to be capturing private information. In the case of a couple in Waasmunster, Belgium, audio included their address, their status as grandparents, and the voices of their son and grandchild.
Google's practices may violate the European Union's General Data Protection Regulations, scholars suggested. One worker screening audio said he encountered a recording in which it sounded like a woman was being physically attacked, but that Google didn't have clear guidelines on what to do in such cases.
A Google spokesperson said that the company has launched an investigation into the contractor mentioned by VRT, on the basis that it violated data security policies.
While Amazon and Google have been subject to the most scrutiny for their voice assistant policies, a technology policy researcher at the Alan Turing Institute in London — Michael Veale — has filed an Irish complaint regarding Apple's Siri, saying it violates the GDPR because users can't access uploaded audio. Both Amazon and Google let users review and delete voice samples.
Apple has responded, according to Veale, by saying that its systems handle data well enough that his recordings don't qualify as personal data.
William Gallagher
Christine McKee
AppleInsider Staff
Chip Loder
Malcolm Owen
33 Comments
While the recordings supplied for transcription are anonymized and not connected to any accounts it plainly would not mean that someone with time, money, sources and a reason to do so would not be able to connect dots within specific snippets. In this case some contractor illegally supplied anonymous recordings to a journalist (?) for the purpose of attempting to do just that. They were seemingly successful in at least one case where the owners address was mentioned.
Anonymous may not mean you cannot be identified given the time and resources. I'd also be shocked if humans were not involved in understanding problematic Siri recordings as well, and yes those are also anonymized and not connected to specific accounts. If "machines" misunderstood what was said or misinterpreted a wake word then using another machine to understand why it happened hardly seems appropriate.
“A Google spokesperson said that the company has launched an investigation into the contractor mentioned by VRT, on the basis that it violated data security policies. “
This statement can be interpreted in many ways:
But one interpretation feels like Google murdered someone with a knife and then handed the subcontractor the bloody knife and then accusing them of murder.
Another interpretation is that Google is investigating a whistle blower, as violating some term in an agreement to not share what Google provided to them.
It's a feature for that more personal touch. Any wonder why I don't have these things in my house.
It’s really surprising that people are shocked by this. People who are involved in the design and development of the product need to know it’s working and also need to find ways to improve it. If companies only relied on customer feedback on how to improve it then improvements wouldn’t be as quick. They need to be proactive so by listening to recordings enables them to do this.