Tech giants change human review policies for recordings of AI assistant interactions
Both Google and Amazon have followed Apple in changing their policies for human review of voice assistant recordings to head off a growing controversy, MacRumors reports.
After reports last week that Apple contractors were hearing private conversations and other interactions from accidental activations of Siri, Apple suspended its quality control review program. The company also reportedly plans to release a software update allowing Siri users to opt out of recording reviews.
Amazon has added a similar opt-out option to the settings menu of the Alexa mobile app, and Google has said its reviews of Google Assistant audio have been suspended. Google took action in the EU earlier in July, after German privacy regulators began investigating a Belgian media report to the same effect.
Policies were already in place to protect user data, such as Google distorting the voice recordings and Apple assigning a random device identifier after removing identifiable information.
Each company has noted that only a small fraction of recordings are analyzed, but without annotating those samples, it seems likely that voice assistant performance would not only not improve, but could actually degrade over time. This is because unlike biometric systems, which are trained on new data to improve performance, speech recognition systems must take into account the natural evolution of language, including slang and accents.