In a promotional video, Amazon.com states that its cloud cam home security camera provides “everything you need to monitor your home, day or night.” In fact, the artificial intelligence requires the help of a squad of invisible personnel.
Dozens of Amazon employees based in India and Romania review selected clips captured by CloudCam, according to five people who have worked on the program or have direct knowledge of it. Those video snippets are used to train AI algorithms to do a better job of distinguishing between a real threat (a home attacker) and a false alarm (a cat jumping on a sofa).
An Amazon team copied and annotated commands recorded in customers’ homes by the company’s Alexa Digital assistant, Bloomberg reported in April.
AI makes it possible to talk on your phone. This is helping investors to predict a change in market attitude. But technology is far from irrational. Cloud Cam sends alerts when it is just a paper roar in the wind. Apple’s Siri and Amazon’s Alexa still make occasional errors. One day, engineers will be able to overcome these shortcomings, but for now AI needs human help. That’s a lot.
At one point, on a typical day, some Amazon auditors were annotating about 150 video recordings each, typically 20 to 30 seconds long, according to people who asked not to be named to talk about an internal program.
The clips sent for review came from staff testers, an Amazon spokesman said, as well as cloud cam owners who submitted clips to resolve issues such as incorrect notifications and video quality. “We take privacy seriously and keep CloudCam customers in control of their video clips,” he said, adding that until the clips are submitted for troubleshooting purposes, “only customers can view their clips.”
Nowhere in the cloud cam user terms does Amazon explicitly tell customers that people are training the algorithms behind their motion detection software.
And despite Amazon’s insistence that all the clips be provided voluntarily, according to the two, the parties have picked up activities that homeowners do not want to share, including rare examples of people having sex.
Clips with inappropriate content are flagged, then discarded so that they are not accidentally used for AI training, people say. A spokesman for Amazon said such clips had been discarded to improve the company’s human review experience, but did not say why the video clips submitted voluntarily showed inappropriate activity.
According to staff, Amazon Cloud Cam has imposed strict security on the vaccine operation. In India, dozens of reviewers work on a restricted floor, where employees are not allowed to use their mobile phones, according to the two. This, however, did not deter other employees from giving footage to non-team members, another said.
Cloud Cam debuted in 2017 and with the Alexa-powered line of Echo speakers, Amazon hopes to give it an edge in the emerging smart-home market.
The $ 120 (approximately Rs. 8,500) device detects and alerts people about ongoing activity in their home and offers free access to footage for 24 hours. Users willing to pay around $ 7 to $ 20 for a monthly subscription can extend that access to one month and receive appropriate alerts for a crying baby, say or smoke alarm. Amazon does not disclose how many cloud cams it sells, but the device is one of many home security cams on the market, from Google’s Nest to Amazon-owned Ring.
While AI algorithms are getting better at teaching themselves, many companies, such as Amazon, deploy human trainers throughout their business; They help Alexa understand voice commands, teach the company’s automated Amazon Go convenience stores to distinguish one buyer from another, and even work on experimental voice software designed to detect human emotions.
The use of humans to train artificial intelligence inside consumer products is controversial among privacy advocates because of concerns that its use could reveal personal information. An Amazon team listened to Alexa Voice Command and subsequent revelations about similar review programs from Google and Apple drew the attention of European and American regulators and lawmakers. The uproar even prompted some Echo owners to unplug their devices.
In response, both Apple and Google have stopped their own human review programs. For its part, Amazon has started allowing Alexa users to exclude their voice recordings from manual reviews and has changed its privacy policies to include an explanation so that people can listen to their recordings.
Reports from information and intercept technology websites examine the role of humans in training software behind the security cameras built by Ring last year. Sites have reported that staff used client-shared clips via the Ring app to train computer vision algorithms and, in some cases, shared unencrypted client videos with each other.
Amazon doesn’t tell customers much about the cloud cam troubleshooting process. Under these terms, the Company reserves the right to process images, audio and video captured by the devices in order to improve its products and services.
In response to a question about CloudCam on its website, Amazon said, “Only you or those with whom you have shared your account information can view your clips, unless you want to submit a clip directly to us to resolve the issue. Customers also choose to share clips through it.” You can email or social media. “
According to the trio, CloudCam teams in India and Romania do not know how the company selects the clips for comment, but they say there is no obvious technical flaw in the footage that needs to be submitted for resolution.
At an industry event this week, David Limp, who manages Amazon’s Alexa and hardware teams, acknowledged that the company could be more forthcoming about using people to audit AI. “If I could go back on time, that’s what I could do better,” he said. “I would be more transparent about why and when we use human vaccines.”
© 2019 Bloomberg LP