The Humans Hiding Behind the

▶ ▶ Artificial intelligence services are leaning on low-tech reps working long hours ▶ ▶ “I left feeling totally numb and absent of any sort of emotion”

Amy Ingram, the artificial intelligence personal assistant from startup X.ai, sounds remarkably like a real person. Given access to your calendar and e-mail, Amy takes over the mundane tasks of scheduling meetings and sending appointment reminders. After she made her debut in 2014, beta testers praised her “humanlike tone” and “eloquent manners.” “Actually better than a human for this task,” one tweeted.

This artificial intelligence, however, isn’t wholly artificial. Behind most every e-mail is a person—someone like 24-year-old Willie Calvin. X.ai’s software can usually guess from its e-mail scans when “tomorrow” means Tuesday, but it’s a ways from being able to operate on its own. An X.ai spokeswoman says trainers review “the vast majority” of information in e-mails so the system can improve.

Calvin joined X.ai in December 2014, shortly after graduating from the University of Chicago with a public policy degree. He was under the impression his $45,000-a-year job as an AI trainer would be split evenly between spot-checking Amy and doing more ambitious product development. He says he was asked, as part of the job application, to write a one-page essay arguing that automation would be good for jobs and workers.

After Calvin started work, he says, the product part of the job never materialized. Instead, he sat in front of a computer, clicking and highlighting the bot’s right and wrong responses to requests, for up to 12 hours a day. “It was either really boring or incredibly frustrating,” he says. “It was a weird combination of the exact same thing over and over again and really frustrating single cases of a person demanding something we couldn’t provide.”

X.ai declined to comment on specific hiring practices. Kristal Bergfield, who oversees X.ai’s trainers, says the job has evolved over time and entails hard work. “We’re building something that’s entirely new,” she says. “It’s an incredibly ambitious thing, and so are the people who work here.”

A handful of companies employ humans pretending to be robots pretending to be humans. In the past two years, semiautomated services have sprung up to offer e-mail schedulers (X.ai, Clara); shopping assistants (Operator, Mezi); and do-anything concierges (Magic, Facebook’s M, Gobutler). Startups in this arena have raised more than $50 million in venture capital funding in the past two years, and most aim to employ as few humans as possible. But for now, the companies tend to use similarly vague marketing lingo and in truth are powered largely by people.

Facebook turned the spotlight on human-assisted AI last summer when it introduced M, a chirpy personal assistant bot that lives in Messenger, its chat app. M’s Ai-generated responses are reviewed, edited if necessary, and sent out by a team of a few dozen contractors working from the social network’s Menlo Park, Calif., campus, according to the company. Facebook won’t say what hours the contractors work or how often they correct M’s guesses.

Clara, which offers an e-mail scheduling service a lot like X.ai’s, uses