"How is the weather today?", "How do I get home?", But maybe "How do you remove blood from a carpet?": Such questions are asked by the Google Assistant, which the company staged as a digital everyday help. The Assistant is constantly evolving. Therefore, part of the assistant voice recordings was not only processed by Google's software until the summer, but later submitted to human resources. The should check how well the Assistant has edited the voice inputs and so contribute to its optimization.
This practice has caused a sensation in recent weeks - not only from the Google Assistant, but also from competing products like Amazon's Alexa and Apple's Siri. Many users of the services were apparently unclear that what the digital helpers intentionally or sometimes unintentionally recorded, sometimes get to hear other people. Against this background, Google had announced this summer that the controversial function was stopped on July 10 throughout the EU.
Today, Monday, the company has now outlined how it should continue. In a blog post, Google Manager Nino Tasca is initially self-critical. "Unfortunately, in this case, we did not live up to our high standards of making it easy for you to understand how your data is being used," Tasca writes. "We apologize for that, and when we heard about your concerns, we paused the process of human transcriptions around the world and extensively reviewed our systems and controls."
Only recordings of certain users are listened to
Google wants to reintroduce human audio controls soon - but limited to users who agree. All assistant users could decide for themselves whether or not their recordings could in principle end up in the listening rounds, suggests the blog post.
Therefore, only people who have activated the "Voice and Audio Activities" in Google's My Activities settings menu are checked by people. The setting, according to Google, among other things, that voice inputs are stored for a certain amount of time, but also that your own voice is better recognized over time. If you want more comfort, you have to live with the release of your photos.
Google says who activates the "voice and audio activities", it is pointed out "that people may listen to your voice input in order to improve the speech recognition technology." In a brief SPIEGEL test on Monday, such a hint in the selection menu but not yet appeared. Google did not announce a specific timeline for the adjustments.
Users who have activated the setting in the past want to ask Google for confirmation before their voice recordings are redirected to people. "Without a new confirmation from you, your voice input is excluded from the process," it says in the blog post.
Set how sensitive the Assistant reacts
In the post Google again emphasizes that only a relatively small part of the Assistant recordings is ever evaluated by humans: The speech is of "about 0.2 percent of all audio recordings of users." It is also emphasized that, in principle, voice inputs submitted to auditors can not be assigned to any specific user account. However, that was already the case in the past - and yet there were always sound snippets that landed personal data with examiners.
Such situations Google wants to counteract with increased security measures for the listening process, in the form of other filters to protect privacy. For example, the company wants to do more to ensure that unintentional activations by the assistant are excluded from the listening rounds.
Specifically, the Google Assistant user should "select shortly" how sensitive the software is to requests such as "Ok Google". "This gives you more control over Assistant to reduce false activation or, if desired, ease of use in noisy environments," says Google.