8 January 2019

WE NEED TO NOT FREAK OUT ABOUT THE ROBOT REVOLUTION


"I want to caution the fetishization of manufacturing," says R. David Edelman, currently the director of MIT’s Project on Technology, Economy, and National Security. "Because building things is great and important, but let's not have nostalgia for the days where my great grandfather would be on an assembly line losing a finger every four years."

YOU, LIKE ME, may sometimes (or all the time!) feel that the world is spiraling out of control—trade wars and political strife. And, oh right, climate change, arguably the greatest threat our species has ever faced. Or maybe artificial intelligence and robots will put us all out of work before the world actually ends.


But know this: Smart people are trying to think us out of our predicament. One of them is R. David Edelman, formerly President Obama's special assistant on the digital economy, and now the director of MIT’s Project on Technology, Economy, and National Security. We sat down with Edelman to talk about the rise of robots, America’s labor woes, and the subtleties of rotten strawberries.

This conversation has been condensed and edited for clarity.

Matt Simon: We are at a fascinating moment in robotics, where the hardware side is increasingly imbued with AI. It used to be, “Here's AI doing cool stuff in a virtual world, and here are robots that are stupid.”

R. David Edelman: There is absolutely a convergence of the fields and the people who work in them. I think the cross-collaboration between the folks in those fields is vastly higher than it's ever been before. That's making these robots incredibly valuable, particularly in a consumer context, and also in an industrial context.

MS: So say one of these strong and newly smart machines threatens your job. What about retraining? Even if robots aren't necessarily coming in and wiping out all the jobs in the factory, they might augment the jobs there, which would require the humans to adapt.

“A dirty little secret about autonomous vehicles,” says Edelman, “is there won't be enough people to service them because these are trade skill programs. We don't pay these people enough.” 
MIT

RDE: It's a concern and an opportunity. The concern is that the United States has historically been poor at job retraining. A certain mythology—I think a wrong-headed mythology—is that you can train every coal miner to code. Which has two problems associated with it. One, the idea that someone in a coal mining job, or anything like it, that their closest adjacency, the job they would be best at doing next, is an entirely different industry with an entirely different skill set with a language they don't speak. And secondly it presumes that coding jobs are not the sweatshop jobs of the next decade. I'm not sure that's a safe bet.

The other thing we do terribly in the United States is the apprenticeship. I talk to these companies in autonomous driving or other robotics spaces, and ask where they see the big gaps. A lot are in maintenance—they can't hire trained hardware technicians fast enough. A dirty little secret about autonomous vehicles is there won't be enough people to service them because these are trade skill programs. They can't live near enough to the city where they would work because of the skyrocketing cost of housing, or because we have stigmatized career and vocational training in the United States in a way that is wildly self-defeating. We don't pay these people enough.

But AT&T for instance has a very interesting program where basically they take line men and women and train them for higher-paying jobs within the sort of core of network management. They pay for two years of their education and bring them back in, and this is classic upscaling.

MS: For the foreseeable future, we’ll need humans to help train the machines, right? So a worker with human eyeballs has to tag images to teach self-driving cars what a pedestrian or a tree looks like.

RDE: Let's think about winemaking or berry production. You've got all these conveyor belts, you've got fruit moving down the line and you've got individual people. They see a strawberry that's rotten and they pluck it off. There are hundreds of these people doing that right now. That has in the last five years become a totally solvable computer-vision-plus-robotics problem.

Lots of people could be put out of a job or will need to move to an adjacent role. But that adjacent role could absolutely be training the model—it's not as if we're going to train models once and for all and be done with it. These folks know how to intuitively pluck a rotten strawberry off a line. I could do it if it were a very rotten strawberry, but one that's going to be rotten in three weeks and not four weeks? There's real skill there. So I see that as a model training component. I see someone as being able to adjudicate hard calls.

MS: Here in San Francisco, we've had a supervisor float the contentious idea of a robot tax. Basically, replace a human with a robot and pay a fee. What are your thoughts there?

RDE: It's rightly contentious. Framed in that context, it is almost certain to be a bad idea by virtue of the obvious disincentive it creates for the increase in labor productivity. If you look at the economic statistics, you will see in the last decade we've had almost a lost decade of labor productivity. Even though the advent of the internet should have given us this dramatic leap forward, it's not totally captured in the economic metrics. So this notion of increasing labor productivity is crucial for GDP growth, it's crucial for being internationally competitive, and it's crucial for individual workers to be able to work in jobs that are less dirty, dangerous, or demeaning.

You can go back 100 years and find newspaper articles predicting the end of jobs, the end of invention, the end of creativity. The truth is, none of us can predict what job categories will exist 30 years from now. If you had asked me in grade school if a mobile app developer would be a job category that would employ a million Americans, you'd have to be crazy to speculate that.

MS: What about the opposite? Using robotics, especially collaborative robots that augment a human’s work, to bring manufacturing jobs back to the United States?

RDE: I think that's not only possible, I think it's happening. A lot of folks that I've talked to in industry, particularly the high-tech sector, are wanting to bring as much manufacturing back to the US as possible. There are a few reasons. One, IP protection is of grave and increasing concern as you get to more proprietary robotic systems. Two, a lot of American executives and others are worried about safety and persistence of the Chinese economy. People are pretty downbeat about how things look right now, they see a slowdown. They see this trade war as really unwinding a lot of the effective economic flows between the two countries.

I also want to caution the fetishization of manufacturing. Because building things is great and important, but let's not have nostalgia be a substitute for economic policy, or somehow longing for the days where my great grandfather would be on an assembly line losing a finger every four years.

MS: It seems to me that we’re living through this changing notion of work. Not in the Elon Musk view of things, where we're all suddenly replaced by a machine next week.

RDE: Yeah we'll destroy ourselves long before they manage to rise up and conquer us. If we get this AI thing wrong and dramatically accelerate class stratification, we'll do that to ourselves.

I think this isn't necessarily a robot panic that we're in, but we do have to recognize the potential for different kinds of displacement than we've seen. Everybody cites that Oxford-Yale report. They asked the question of machine learning researchers: By what year will there be no jobs? They all conjectured, Oh, 50 percent of the jobs will be gone in X number of years. But what really tickled me about that report is that also asked them what do you think will be the last job to be automated. Do you know what their answer was?

MS: Their job?

RDE: AI researcher, of course! Because everybody believes theirs will be the last job to be automated.

No comments: