Bringing Up AI: How People Are Teaching Their Jobs to Machines


The NewCo Daily: Today’s Top Stories

Audrey Watters | Flickr

The economy stands at a threshold moment in the era of machine learning. The artificial intelligences that companies are increasingly deploying are just beginning to take on roles and jobs that used to demand a human being at the controls. But in most cases they’re nowhere near ready to take over entirely. They still need people at their sides — in some cases to generate the data that will train them, in others to provide judgment that’s beyond them.

Welcome to the world of the hybrid human-machine workplace. A couple of recent articles have begun to give us a portrait of this emerging work environment, with its awkward encounters, unemployment fears, and potential for both efficiency and exploitation.

In Wired, Davey Alba talked with a bunch of people who screen YouTube videos for content that might offend advertisers. In the long run, Google (which owns YouTube) aims to hand this task over to an AI. But the judgments involved are complex, opaque, and subjective, and distressed advertisers aren’t going to wait for the technology to mature. So low-paid, part-time contractors hired through an agency called ZeroChaos do the work. Their video ratings serve two purposes — protecting YouTube’s revenue right now, and building up a trove of data to help the AI learn what humans (and advertisers) find objectionable.

A job’s a job, and a lot of the people doing this one are glad to have it. But it’s high-pressure, high-volume piecework, and working for Google sometimes feels like working for an inhuman AI; the company barely communicates with workers, dismisses them precipitously and without explanation, and provides no benefits, job security, or guarantee of steady work.

The problem with treating your AI tutors this way isn’t just a matter of ethics — it could also warp the outcome of the whole project. As Alba puts it: “if it turns out you’re training your AI mainly on the perceptions of anxious temp workers, they could wind up embedding their own distinct biases in those systems.”

Portraits of AI Tutors In All Walks of Life

Machine-learning tools are extending their reach far beyond the giant tech platforms that pioneered them. In The New York Times, Daisuke Wakabayashi offers a compendium of case studies of the propagation of AI techniques into other industries.

At Lola, a travel-booking app, human travel agents have been guiding the education of an AI named “Harrison” that has become proficient at recommending hotels. (The human agents are still better at offering users travel tips, or helping with upgrades.)

Legal Robot is developing another AI that can parse complex contracts and other legal documents, identifying problematic passages and suggesting improvements. Its CEO points out that legal agreements — with their repetition, formality, and structured nature — make good fodder for machine learning.

At Magoosh, the test-prep company, customer service reps are speeding up their answers to incoming student questions now that they have an AI at their disposal that’s gotten steadily better at suggesting email replies. But employees don’t think they’re going to get edged out any time soon: Too many questions still require human intuition, and people are still better than AIs at knowing when it makes sense to break a rule.

In all these cases, the relationship between human worker and AI is neither coexistence nor warfare but rather a continuous process of reaction, adjustment, and evolutionary change. The technology’s advances have been prodigious, yet it still can’t do most of the things we expect it to eventually master. We’re still waiting to find out just where people will fit in when these software machines have caught up with our imaginations.

Let a Hundred Hardware Businesses Bloom

As AI transforms information work and leaves us wondering where future jobseekers will turn, one answer may lie in an unexpected place: the world of hardware manufacturing. If the dozens of startups clustered in Brooklyn at a former shipbuilding facility are any indication, there’s a boom looming in the making of physical goods supercharged by new digitally driven techniques like 3-D printing and laser-cutting (Steve Lohr in The New York Times).

These companies seem to be leading a modest renaissance in urban manufacturing in the U.S., though data is scarce, since firms entering markets like medical devices, consumer goods, and electronics are separately tracked in each of their fields.

This kind of physical-goods business will rely on in its own species of AI, naturally, to speed productivity, detect patterns, and solve problems. All that efficiency means such companies won’t ever end up hiring armies of human workers like their industrial-age predecessors. But the flourishing of boutique urban manufacturing still offers some hope that not everything solid will melt into the ether.

He’s Signing Books in the Lobby

Missed the Pope at TED this year? For a Friday afternoon pick-me-up, try this TED talk by one Jesus of Nazareth (from The Late Show with Stephen Colbert). Hard to say how the moderators — human or AI — will rate it.

Leave a Reply