Learning AI #32

When AI is your boss

Karim Sawadogo drove for Uber in Denver for seven years and gave more than 20,000 rides. Then a rider accused him of driving drunk. Sawadogo doesn't drink, for religious reasons, and went straight to the police to get his sobriety on record. Uber locked his account anyway.

He stayed locked out for nine months. He messaged support again and again. Uber eventually blocked his phone number too. He fell behind on his bills. "I didn't know what to do with my life," he said.

"They treat you like a machine."

Karim Sawadogo, Uber driver

What got him back on the road was a plane ticket. Sawadogo flew to San Francisco and walked into Uber's headquarters. The first employee he found couldn't help him. He came back three days later, found someone else, and that person looked at his file and reinstated him in ten minutes.

Nine months versus ten minutes tells you what algorithmic management actually looks like. The algorithm is slow to change because there's no one on the other end who can look at a file and use judgment. A human did that in ten minutes, once Sawadogo could physically stand in front of one.

Uber is trying let the AI run autonomously for management decisions but is finding it still needs a human-on-the-loop

Uber’s default was a human-out-of-the-loop system, where the AI was fully autonomous. Works greats, most of the time, but when it doesn’t, it makes quite a mess.

Separately, European regulators are said to be preparing to fine Uber close to a billion dollars, for deactivating drivers without meaningful human review, a practice the regulator says violates European rules on automated decisions with serious consequences. Regulators are saying that Uber needs a human-on or human-in-the-loop.

"A computer should not make decisions on its own that have such major consequences."

Monique Verdier, deputy chair, Dutch Data Protection Authority

Uber says it will appeal, and that most of its driver suspensions are temporary and reviewed by people. Sawadogo waited nine months to find out his wasn't.

New York tried to legislate a fix. A city law passed in January required rideshare companies to give drivers 14 days' notice before deactivation, put the burden of proof on the company, and let drivers sue over it.

Uber and Lyft sued the city instead, arguing the law stripped their ability to remove dangerous drivers. A federal judge blocked the law days before it was set to take effect.

The record so far: one regulator says firing drivers by algorithm without a human violates the law. One city tried to require a human anyway. The company that would have had to provide the human-on-the-loop got a federal judge to say it doesn't need to, at least for now.

These are becoming everyday terms in our increasingly AI-driven world. We all should understand them.

Every industry that's handling scheduling, discipline, or performance reviews is building the same setup: a decision that used to come from a manager you could talk to now comes from a system you can't. That’s a move from human-in-the-loop to human-out-of the-loop, skipping over the all-important step on having a human-on-the-loop to intervene when the AI makes a bad call.

The gig economy got there first because it never had managers to begin with. Whatever gets decided about Uber's drivers over the next few years is going to end up as the template for jobs across the AI-managed economy.

Things I think about

Honey never spoils. Archaeologists have found 3,000-year-old jars of honey in Egyptian tombs that were still edible.