After it faced public backlash against the practice of using people to review voice recordings from Alexa devices, Amazon gave users increased privacy controls and came clean about how those recordings were used.
But it also considered a more extreme measure that would have opted users out of the practice by default.
In the end, the company decided not to institute such changes, said Dave Limp, Amazon’s devices and services chief, during an interview at the 2019 GeekWire Summit. Limp defended the decision by saying that human review is “critically important to making Alexa better.”
“We don’t want data for data’s sake,” Limp said. “We want to use data where we can actually improve the experience on behalf of the customer.”
Apple, which faced scrutiny for a similar practice on its Siri personal assistant, recently changed its privacy setting so that users are, by default, opted out of human review of audio samples.
Limp argued that human review was critical to Alexa’s launch in the Hindi language. Through information gathered in the first 90 days, the company was able to improve the accuracy of the assistant by 30 to 35 percent, Limp said.
“I want that improvement to get customers. And you do need a corpus of data that’s broad enough from all sorts of different geographies and demographics to make that advancement,” he said.
Amazon’s practices came under the microscope earlier this year following reports that a team consisting of thousands of people listen to Alexa voice recordings as part of a program designed to improve the company’s voice assistant. In the world of artificial intelligence, the use of human reviewers was widely understood, but it was little known by the general public.
Amazon responded with a privacy setting that let users opt-out of human review. It also changed the language on its privacy page and, later on, introduced a feature that lets users automatically delete recordings.
Given the chance to go back in time, Limp said he “would have been more transparent about why and when we are using human annotating.” In the future, the company may be able to employ AI methods like federated learning so that humans are no longer needed, but such methods are not yet possible, Limp said.
The privacy issue has become a central question as Amazon looks to move Alexa beyond smart speakers. Limp is fresh off a torrent of hardware announcements, including Alexa-powered wireless earbuds, Alexa eyeglasses, a hand-worn “smart ring” called the Echo Loop, as well as smart dog collars, ovens and clocks.
“We’re starting to experiment with what it means to have Alexa on the go,” Limp said. In announcing the new devices last month, Limp spent several minutes at the outset of his presentation outlining the company’s efforts to provide customers with privacy controls.
European regulators have responded to data privacy concerns with a set of regulations known as GDPR. “[GDPR] on balance is very, very good,” said Limp, adding that some aspects of the law needed further clarification.
Amazon CEO Jeff Bezos last month said that “good regulation” would be welcome. The tech giant’s facial recognition software, which is in use at several law enforcement agencies, has drawn opposition from civil rights groups such as the ACLU.