Voice-assistant recordings are reviewed by human contractors despite corporate anonymity claims

Evidence

Pieces of your talk are regularly farmed out in bulk to third-party firms that conduct “audio review processes” in which virtual scorers, tasked to evaluate the degree of match between the machine’s text and the original chunk of human speech, review audio recordings retained from smartphones, messaging apps, and digital assistants. Companies such as Amazon, Google, and Microsoft use these audio analyses to improve the algorithms of their voice systems. The tech companies insist that such recordings are anonymous, nothing more than voices without identities. “Partners do not have the ability to tie the voice samples back to specific people,” one Microsoft executive asserted. But one journalist who signed on to a virtual job as an audio recording analyst concluded just the opposite, as she listened to recordings full of pathos, intimacy, and easily identifiable personal information: Within the recordings themselves, users willingly surrender personal information-information that is especially valuable in these review processes because they are so specific. — Shoshana Zuboff, The Age of Surveillance Capitalism, 819

Categorization

• XREF: Connects to surveillance-capital claims about behavioral surplus — talk captured as surplus raw material. Also echoes the reader’s broader attention/platform/surveillance research purpose. — connects to: Voice recordings reveal identities and intimate details that anonymity claims cannot protect — Shoshana Zuboff, The Age of Surveillance Capitalism, 820
• SEED: Angle: the ‘anonymity’ fiction in AI training — how contractors become the invisible human layer behind ‘automated’ systems, and the consent gap for users whose intimate moments are listened to.

Broader Connections

• Voice recordings reveal identities and intimate details that anonymity claims cannot protect #expands (0.88) — Shoshana Zuboff, The Age of Surveillance Capitalism, 820