When I was teaching artificial intelligence or machine learning, occasionally a journalist or journalism student would drop by my office to ask about the threat of AI. This is always a strange question to an AI researcher — the practice of AI is so different than the public perception that we tend to forget that we’re not even really talking about the same thing at all. We’re thinking about particle filters and POMDPs and worrying about whether we can get the robot to consistently fold towels, while, thanks in large part to Hollywood, the general public are largely worried about Skynet’s robots climbing piles of skulls or droid armies invading.
I usually tried to gently explain that AI was so far from such scenarios that it was not worth panicking about just yet. Our worrying on the coming robot revolution would be a bit like Leonardo da Vinci spending stress on the threat that passenger jets pose to sky scrapers and the socio-political consequences of such a use. He could envision flight, but he was so far from achieving it that he could not possibly guess what the true implications would be.
Yet… After a few such questions, I began to realize that there is a very real and very immediate threat, even from the level of AI we’ve already achived. It’s just that it looks nothing at all like Skynet or Cybermen.
The true threat, today, of AI is the natural extension of the same threat that we’ve been facing for two centuries, with slowly growing urgency. It’s not a physical threat, but an economic one: automation.
Since the days of the Jacquard loom and the steam drill, society has been shifting away from human craftsmanship and manual labor to increasing levels of automation and machine power. In industry after industry, the invention of cheap power and automation techniques have shifted labor from human workers to machine. First raw, physical labor; later, skilled craftsmanship.
This struggle has been the centerpiece of labor/capital disputes since at least 1800 and, in spite of sabotage, labor movements, government protectionism, and protest songs, the story has played out essentially the same way every time: labor loses. The economic forces at play are just too strong — the economic advantage of automation is simply overwhelming.
The classic story bandied about by politicians and the pop economic press is, “No worries! Sure, the fact that automation is annhihilating the labor segment of industry X causes some short-term pain to the workers in that industry, but, thanks to ingenuity and drive, a new industry is always around the corner! Those folks will simply re-train, re-tool, and move on to a better position in a different industry.”
It’s a pat story, albeit not exactly comforting to those who are facing “re-training and re-tooling” as their jobs evaporate in the face of the onslaught of automation. Only… What happens when there are no more new industries to absorb workers?
It turns out that the role of computers, robots, and automation is at the center of current economic arguments about the sources of income inequality in the US and even the causes of the Great Recession. The question is whether the same old story about the effects of automation continues to hold, or whether something new is going on.
The truth is that it’s something completely new in human history, and it’s very, very different than what came before.
The thing about the computer — the thing that I don’t think economists and pundits have really grasped in their bones yet — is: the computer is the first truly general purpose machine that the human race has ever invented.
Before the computer, individual machines essentially did one specialized thing apiece: cars drove, backhoes dug, combines harvested, lathes turned. Had a new job to do? You needed to find or build a new machine to do it. Had a new industry to automate? You needed to invent a new automation process and batch of machines to do it.
Computers aren’t like that. They are, in a very literal and mathematically deep sense, infinitely flexible machines.
As computers began to enter the industrial scene, all the old automation scenarios began to change. Now one machine could be quickly and easily programmed to do a wide variety of different tasks. You no longer needed typesetters (or type manufacturers or linotype smelters or…) when one smart laser printer could print all possible fonts, texts, and images. You don’t need a pile of master machinists to run a manufacturing floor when you have CNCs — you only need one person to program and maintain the CNCs. And 3D printers are still in their infancy, but it’s clear that they’re the new generation of automated stuff creation, displacing who knows how many sculptors, mold-makers, and so on. Even fiddly assembly work is getting increasingly dexterous robots to do the job, again displacing skilled and semi-skilled workers.
So far, the story fits right in with the two-century-old tale of “blue collar workers get shafted by automation while white collars reap the profits”. But there’s more to the story this time…
The thing that’s sneaking up on us now is that it’s no longer all about manual labor and manufacturing. More and more, white collar jobs are also under the axe of automation. It just hasn’t yet really sunk into the public consciousness.
ATMs — how many non-automated bank tellers did those replace? Every time you hit a voice mail tree and swear because you can’t get a human? That missing human is a displaced desk job. Robots are making travel reservations for us, trading stocks for us, routing shipments for us, and doing all sorts of accounting for us. Soon, they’ll be driving cars for us.
And it’s not just “rote” information jobs either. While AI is still immensely far from realizing Turing’s dream, our still-primitive creations are already invading things that are traditionally the realm of the creative classes. Machines already grade essays. I recently ran into a company that has a system that writes news articles. A group in the UK built a robot scientist that can formulate and test hypotheses (and, incidentally, do the work of half a dozen lab techs and grad students). The big internet companies are making scads and oodles of cash by understanding your shopping psychology better than you do. Pandora understands music very primitively, but well enough to make millions of listeners happy with its recommendations. And the tech behind “MOOCs” (perhaps debatably) enables one teacher to do the job of hundreds.
So… Where do we go from here, and what do we do about it?
I think the history and the trend is clear: so long as there’s economic incentive, the tech will continue to advance to capture that profit. And there’s enormous economic incentive to make machines even just a tad bit smarter (often by displacing expensive human workers). I don’t see that incentive changing, regardless of how smart the machines get. So the only limit is… How smart can the machines get?
Personally, I see no particular technological limit. We are far from understanding how brains really tick, but we’re making surprising inroads there, and all signs are that the human mind really is just a big, sophisticated machine under the hood. We still have a lot of puzzles to crack, but I think it’s possible to crack them and really get to full AI.
But the robot overlord threat hits far before full AI. Already, automation is driving the 1%/99% divide in the first world. As the tech gets smarter, it will continue to ruthlessly eliminate not just jobs but entire industries and types of jobs. The human workers will end up in a growing pool of third-class citizens, behind the machines who took their jobs and the people who own those machines. And, eventually, I suspect, even the humans at the top of the heap will end up second fiddle to the machines, as the computers become better at manipulating capital and making investments than humans are.
We’re really still in early days of this path, but we’re already feeling the first throes of what I think is going to be an enormous social and economic upheaval. I think the Great Recession is one of the first really big upheavals of this process, and our fights about capitalism, socialism, and other economic -isms are a symptom of having to confront it.
We no longer have the luxury of our centuries-old pat story about workers re-tooling and retraining. It’s not “them” any more. It’s “us” now. All of us. We have to actually confront the issue now, and think about whether we’re really prepared for a post-labor society, what such a society would look like if we are, or how to avoid it if we’re not.