Human Listeners and Virtual Assistants: Privacy and Labor Arbitrage in the Production of Smart Technologies

Abstract

In spring 2019, public outcry followed media revelations that major producers of allegedly automated voice-activated devices recruit human operators to listen to, transcribe, and label users' conversations. These high-profile news stories originated in journalistic and scholarly investigations, and in the voluntary disclosures of whistleblowers. The scandal started with a Bloomberg report on April 10, 2019, disclosing that thousands of Amazon workers listen to the private conversations of users of the smart speaker Echo. Workers include both employees and subcontractors of Amazon, located in the US and in India. A company spokesperson claimed that these privacy violations concerned "an extremely small number" of the 100 million-strong Echo user base (Day, Turner, and Drozdiak 2019). A few months later, the Flemish Radio and Television Broadcasting Organization (VRT) revealed that third-party outside contractors were eavesdropping on audio recordings of users of Google Home smart speakers and smartphone apps. The practice was not characterized as an exception; Google subcontractors were said to "systematically" listen after logging into a secure platform. Some of the conversations they accessed contained sensitive information, such as private arguments, confidential business conversations, and the whereabouts of children (Verheyden et al. 2019). Even more intimate details were said to reach Apple. According to a whistleblower who used to work for a subcontracting company, these included "confidential medical information, drug deals, and recordings of couples having sex." This data was collected via the company's voice assistant, Siri, which is included in most of its devices from iPhones to HomePod smart speakers and Apple Watches. As Amazon had done earlier, an Apple spokesperson claimed that this concerned "a very small random subset, less than one percent of daily Siri activations"