Where technology and economics collide
Over at Mother Jones, Kevin Drum has a quick post rebutting the two worst objections he hears to the fear that AI-empowered robots will take all our jobs. The arguments are:
Drum dismisses these as follows. The first, he says, is meaningless, “because AI doesn’t exist yet.” He predicts that “we’ll see the first glimmers of true AI in about ten years, with full AI coming 20 or 30 years later,” but says that no one believes we have AI today, “so of course it’s not reflected in the current productivity statistics.”
The second argument, he says, is just a poor analogy. The AI revolution has nothing to do with the industrial revolution. “AI, by definition, implies human-level intelligence,” he writes. “Thus, by definition, if the AI Revolution creates new jobs, those new jobs will also be done by AI-equipped robots.”
Unlike Drum, I’m skeptical that AI will lead to mass unemployment in the next 50 years (which is about as long a time frame I’m comfortable making any kind of prediction about), and perhaps this is a good moment to lay out why.
First, I take slow productivity growth more seriously than Drum does. It’s true we don’t yet have AI. But we have seen an explosion of information technology that should’ve put many jobs at risk, and in some cases clearly did. The stories we can tell about why ATMs would replace all bank tellers, or online learning would displace most teachers, or digital diagnostic tools would make many doctors superfluous, are at least as convincing as the stories we can tell about AI-driven job displacement, and the technology is already here.
And yet there are more bank tellers today than there were in the 1970s. And online learning hasn’t dented the demand for teachers. And WebMD has mostly sent people scurrying to see their doctors. What happened?
Toward the end of his post, Drum suggests that a better argument for why AI might fail to deliver mass unemployment is that “robots will be smart but never very sociable, so humans will all move into jobs that require social skills.” This, however, is part of the explanation for why IT hasn’t done more to move the productivity needle either. People want to interact with other people, even when they’d be almost as well off interacting with a computer interface. And so even when IT makes it cost-effective to replace people with computers, that often leads companies to plow the savings into more people to work alongside the computers (which is largely what happened in banking).
The productivity story isn’t surefire proof that AI won’t upend employment. But it should make us skeptical of confident predictions that it will.
Which brings us to the industrial revolution. Drum is right that we shouldn’t assume a past period of technological change is directly analogous to a future period of technological change. But one lesson of the past few hundred years of technological change is that human beings are pretty good at inventing things to do after automation pushes out the things we used to do.
As I understand the argument for AI-driven mass unemployment, the basic theory is that human beings won’t be needed to do most jobs, and so they will be replaced, and left in a state of useless pleasure seeking and mischief making. When I had him on my podcast, Yuval Harari, the author of Sapiens and Homo Deus, offered a particularly grim vision of this future:
Drum agrees, and puts some stark numbers to the point:
This is where the Industrial Revolution, and subsequent technological revolutions, really do feel relevant. A hundred years ago, or 400 years ago, people did much more useful jobs — huge swaths of the human race, for instance, were directly involved in the production of food and the collection of water.
Compared with those ancestors, humans today are a massive useless class. What sort of job is “editor of an explanatory journalism web site” next to “farmer”? Would our ancestors value the work of psychologists or customer service representatives or wedding planners or computer coders?
But this, to me, is the story of labor markets in the past few hundred years: As technology drives people out of the most necessary jobs, we invent less necessary jobs that we nevertheless imbue with profound meaning and even economic value.
The AI revolution, if it comes, does not seem likely to follow a wildly different path. The technology’s diffusion is likely to be slower than people think — I suspect we will have trucks capable of making driverless deliveries long before regulators permit them to operate without a human in the cab, and computers capable of making diagnoses for decades before human doctors aren’t required to look over and explain the readout — but as it comes, humans will find things other humans want them to do, and they’ll decide those things have value. We’re good at that.
A possible response to this is that if AI achieves human-level intelligence, there will be no jobs AI can’t do, and so no jobs we can invent that AI won’t outcompete us for. But that’s a narrow view of what makes human beings human: Human-level AI is not human-level sociability, or appearance, or personality, or a million other things. There is a reason that yoga tapes haven’t replaced yoga instructors, and in fact, simply seem to increase demand for them.
The AI conversation often seems to me to be dominated by very analytical people who overestimate the value of analytical intelligence and underestimate the value of other skills and aptitudes that make people good at their jobs and valuable to each other. But those other skills are real, and numerous, and I’d want to see more evidence of job displacement coming from the information technology we already have — much of which can simulate our analytical capabilities just fine — before I believe that AI will replace them and the jobs that rely on them.