Google Changes Terms For Virtual Assistant Transcripts

CloudInnovationSecurityWorkspace

Company to require users to reconfirm if they want to allow humans to listen to Assistant recordings, following privacy row

Google has said it will only allow humans to listen to snippets of the audio recorded by its virtual assistant if users opt into the feature, following concerns about such processes at Google and its competitors.

Google said customers had always opted into the feature, but that previoulsy it wasn’t clear enough that humans would be listening to the recordings.  That is to be made more obvious under the new terms, the company said.

Customers who had previously opted in will be required to do so again before the process is restarted.

Google said about one in 500 recordings would be checked for accuracy by human listeners.

security and privacyEavesdropping

In August Apple switched to an opt-in model for human transcriptions of its Siri recordings, while Amazon and Facebook require customers to opt out of the process.

In April Bloomberg reported that Google, along with Amazon and Apple, all employed human staff to listen to recordings made by their virtual assistants to improve accuracy.

The practice is controversial due to the intimate nature of virtual assistants, which are typically used in homes, and due to the fact that the devices often begin recording without being asked to do so.

Many customers were unaware that strangers were listening to the audio recordings made by the assistants.

In July Belgium’s VRT NWS news service said it had obtained thousands of leaked Google Assistant recordings, which included mistaken recordings with sensitive business information and “bedroom conversations”.

In July Amazon acknowledged that it stored recordings made by its Alexa assistant indefinitely.

Stored recordings

Hamburg’s data protection commissioner opened an investigation into the issue in August, after which Google said it had suspended human reviews for European customers.

In a blog post, Nino Tasca, one of Google Assistant’s product managers, said Google had “fallen short of our high standards”.

He said Google plans to roll out an audio sensititivity feature later this year that will give customers the option of making Assistant less likely to switch on in error, although this means the device may also fail to detect activation words in noisy conditions.

Tasca said Google will now “automatically delete the vast majority of audio data associated with your account that’s older than a few months” and plans to “vastly reduce the amount of audio data we store”.

Google has not committed to using only its own staff to review audio recordings, unlike Apple, which said it would stop using third-party contractors.

Facebook recently gave users the ability to choose not to have commands stored and later reviewed by humans.

Read also :