Ghosted

Like humans, algorithms are what they eat. And tech companies have come to depend on humans who “clean” and refine training data they then feed to automated systems. In Ghost Work, coauthors Mary L. Gray and Siddharth Suri are the first to extensively look into the distributed class of workers tidying up the world’s data.

Gray told me every company developing AI relies on this human work, which requires creativity and contextualization.

• Though the jobs are important, “the markets can’t really price” them, Gray said.
• Often, it’s low-paid gig work on online task platforms.

What “ghost work” looks like:

• Identifying the common items—fire hoses, traffic lights, “honk if you’re human” signs—captured by self-driving cars’ always-on cameras
• Flagging explicit user-generated content on social media that made it past automated filtering systems
• Captioning or translating audio clips

Bottom line: Intelligent systems aren’t inherently intelligent. Ghost work, a bargain for tech companies but a slog for those doing it, is how they get smarter.

+ While we’re here: The NYT found that 25% of calls from Duplex, Google’s AI-boosted booking service, start with real human operators.