Facebook has become the latest company to admit that human contractors listened to recordings of users without their knowledge, a practice the company now says has been “paused”.
Citing contractors who worked on the project, Bloomberg News reported on Tuesday that the company hired people to listen to audio conversations carried out on Facebook Messenger.
The practice involved users who had opted in Messenger to have their voice chats transcribed, the company said. The contractors were tasked with re-transcribing the conversations in order to gauge the accuracy of the automatic transcription tool.
“Much like Apple and Google, we paused human review of audio more than a week ago,” a Facebook spokesperson told the Guardian.
Facebook is the fourth major company to have been exposed using humans to listen in on audio recordings that users expected would only ever be heard by an artificial intelligence.
In April, Bloomberg revealed that Amazon was using humans as part of the quality assurance for Alexa, the company’s personal assistant. Contractors across multiple countries reported hearing “upsetting, or possibly criminal” recordings, they told the organisation.
In July, the Belgian public news channel VRT revealed that Google carried out the same quality control, after a contractor leaked more than 1,000 audio clips to the channel. Almost 15% of the recordings had been made accidentally, and some contained sensitive personal information. Google paused the practice after the voice clips were leaked, and a German data protection commission forced the company’s hiatus to be extended to a minimum of three months.
Just a week later, the Guardian revealed that Apple carried out the same practice, “grading” Siri by having humans contractors listen to recordings. At the time, Apple said: “A small portion of Siri requests are analysed to improve Siri and Dictation. User requests are not associated with the user’s Apple ID. Siri responses are analysed in secure facilities and all reviewers are under the obligation to adhere to Apple’s strict confidentiality requirements.”
Although the company insisted its voice assistant was more privacy-conscious than Amazon’s or Google’s, due to its practice of uploading the recordings without linking them to an Apple ID, rather than keeping them all in the cloud as part of the users’ account, it announced it was pausing the human grading a week later. The company told the Guardian that it was “suspending Siri grading globally”, while it conducted a “thorough review. Additionally, as part of a future software update, users will have the ability to choose to participate in grading.”
In August, Amazon, the first to be reported using human oversight, eventually followed Google and Apple in pushing changes to its program, offering users the ability to explicitly opt out by disabling a setting labelled “help improve Amazon services and develop new features” on the Alexa privacy page.
Finally, last week, Vice News caught Microsoft in the same trap, revealing that human contractors were being used to carry out quality assurance on recordings for the company’s Cortana voice assistant, as well as for voice conversations carried out over Skype if the app’s translation feature are turned on.
Human oversight of supposedly automated systems is common within the tech industry. But voice speakers and chat programs carried a significantly higher level of concern over privacy, which left all the companies struggling to match user expectations with their actual practices.
Voice assistants can be accidentally triggered, mishearing ambient noise, homophones or, in the case of Apple’s Siri, “the sound of a zip” as their “wake words”, resulting in them recording and transmitting audio without a direct command. Those accidental triggers led to some of the most egregious violations of privacy, as contractors reported hearing personally identifiable information, medical consultations and couples having sex, in the process of carrying out their work.
Similarly, the transcription and translation services offered by Skype and Facebook Messenger capture far more personal information than users could have expected would be played back to a human stranger, including, again, full addresses, job interviews, and “stuff [that] could clearly be described as phone sex”, according to one contractor.
The Information Commissioner’s Office in the UK has said it is investigating whether the undisclosed use of human oversight has breached the General Data Protection Regulation (GDPR). “We are aware of privacy concerns relating to voice assistant programs and will be assessing the available information,” a spokesperson said, adding that the regulate would meet with its European colleague “in order to ascertain the full facts and any possible risks to the rights of UK residents”.
Ireland’s Data Protection Commissioner, which oversees Apple and Google, said “we will be making our assessments and conclusions” about the practice. “We note that both companies have ceased this processing.”
In late July, the US congressman Seth Moulton introduced a bill called the Automatic Listening Exploitation Act that would empower the Federal Trade Commission to issue fines of up to $40,000 each time digital personal assistants and smart doorbells record private conversations of users who haven’t said the device’s wake word or phrase or activated the doorbell.
