Interpreting a spoken request isn’t magic, rather it has taken a team of underpaid, subcontracted linguists to make the technology possible.
“Do you believe in magic?” Google asked attendees of its annual developer conference this May, playing the seminal Lovin’ Spoonful tune as an introduction. Throughout the three-day event, company executives repeatedly answered yes while touting new features of the Google Assistant, the company’s version of Alexa or Siri, that can indeed feel magical. The tool can book you a rental car, tell you what the weather is like at your mother’s house, and even interpret live conversations across 26 languages.
But to some of the Google employees responsible for making the Assistant work, the tagline of the conference – “Keep making magic” – obscured a more mundane reality: the technical wizardry relies on massive data sets built by subcontracted human workers earning low wages.
“It’s smoke and mirrors if anything,” said a current Google employee who, as with the others quoted in this story, spoke on condition of anonymity because they were not authorized to speak to the press. “Artificial intelligence is not that artificial; it’s human beings that are doing the work.”
The Google employee works on Pygmalion, the team responsible for producing linguistic data sets that make the Assistant work. And although he is employed directly by Google, most of his Pygmalion co-workers are subcontracted temps who have for years been routinely pressured to work unpaid overtime, according to seven current and former members of the team.
These employees, some of whom spoke to the Guardian because they said efforts to raise concerns internally were ignored, alleged that the unpaid work was a symptom of the workplace culture put in place by the executive who founded Pygmalion. That executive was fired by Google in March following an internal investigation.
But current and former employees also identified Google’s broad reliance on approximately 100,000 temps, vendors and contractors (known at Google as TVCs) for large amounts of the company’s work as a culprit. Google does not directly employ the workers who collect or create the data required for much of its technology, be they the drivers who capture photos for Google Maps’ Street View, the content moderators training YouTube’s filters to catch prohibited material, or the scanners flipping pages to upload the contents of libraries into Google Books. Read more