The latest data leak showed that Microsoft employees earn between $ 12 and $ 14 per hour and must process 200
Apple, Google, Amazon, and more recently, Facebookwere seen using human labor to decrypt audio recordings of their own products. According to Motherboard, Microsoft is doing the same for Skype, despite the fact that many companies are abandoning such a workforce.
A recent data leak seen by Motherboard,shows the backstage of artificial intelligence development by IT giants: hired workers repetitively study application data to improve the quality of speech recognition. This means that promising applications for working with human speech are unthinkable without monotonous human labor.
The volume of work increases many times due to the wide integration of Cortana into completely different devices: these are Windows 10, Xbox One, and mobile OS.
“I did a hell of a lot of transcripts and annotationsCortana teams as part of their work for Microsoft, ”says one of the corporation’s employees. Motherboard guaranteed the respondent anonymity for a more frank dialogue about the internal affairs of the company; moreover, the source signed a non-disclosure agreement.
The instruction for internal use containshundreds of pages, including special requirements for classification of information and punctuation. The source says that Microsoft employees need to process 200 voice fragments per hour. Moreover, in such fragments, personal and intimate information is relatively often found.
Payment for such work varies in smalllimits: one contract speaks of $ 12 per hour with the possibility of receiving an additional dollar as a bonus, a contract for a similar type of work implies about $ 14 with the same negligible bonus.
A whole section of materials is devoted to the phrase “Hello,Cortana ”, especially its pronunciation with different accents and in different languages, including German, Chinese, Japanese, as well as various dialects of English.
Motherboard received a letterthe following content: “We always try to make our business transparent and help clients make their choice. We always inform that we use user data to improve our products, and also involve third parties to help with this, while at the same time anonymizing the data. ”
As for the work of third parties, theirThe main task is data classification. Workers were required to distribute transcripts by topic. There are more than twenty of them, including: “Calendar”, “Alarm clock”, “Camera”, each of which is responsible for its functions of the same name. Other topics include games, email, communications, feedback, smart home, and food ordering, found in Microsoft docs. The topic "general" is intended for decryption, suitable for several categories.
Each theme or "domain" has its own functions: turning on / off and setting the alarm, setting and turning on the timer - this is what applies to the "alarm" domain.
The main task of employees is precisely the distribution of information in such categories. But another document shows that categories change very often, again and again adding jobs to people.
A source: Motherboard