There’s No You in AI

Andrew Stuttaford:

This thoughtful piece on what ‘robots’ are going to do to employment by Kevin Drum might be published in Mother Jones (and it comes with quite a few Mother Jones flourishes), but take the time to read it, (very) stiff drink in hand.

Drum’s focus is less on robots (as conventionally understood) than on Artificial Intelligence (AI):

AI is improving exponentially, a product of both better computer hardware and software. Hardware has historically followed a growth curve called Moore’s law, in which power and efficiency double every couple of years, and recent improvements in software algorithms have been even more explosive. For a long time, these advances didn’t seem very impressive: Going from the brainpower of a bacterium to the brainpower of a nematode might technically represent an enormous leap, but on a practical level it doesn’t get us that much closer to true artificial intelligence. However, if you keep up the doubling for a while, eventually one of those doubling cycles takes you from the brainpower of a lizard (who cares?) to the brainpower of a mouse and then a monkey (wow!). Once that happens, human-level AI is just a short step away.

This can be hard to imagine, so here’s a chart that shows what an exponential doubling curve looks like, measured in petaflops (quadrillions of calculations per second). During the first 70 years of the digital era, computing power doubled every couple of years—and that produced steadily improving accounting software, airplane reservation systems, weather forecasts, Spotify, and the like. But on the scale of the human brain—usually estimated at 10 to 50 petaflops—it produced computing power so minuscule that you can’t see any change at all. Around 2025 we’ll finally start to see visible progress toward artificial intelligence. A decade later we’ll be up to about one-tenth the power of a human brain, and a decade after that we’ll have full human-level AI. It will seem like it happened overnight, but it’s really the result of a century of steady—but mostly imperceptible—progress.

Far from slowing down, progress in artificial intelligence is now outstripping even the wildest hopes of the most dedicated AI cheerleaders. Unfortunately, for those of us worried about robots taking away our jobs, these advances mean that mass unemployment is a lot closer than we feared—so close, in fact, that it may be starting already. But you’d never know that from the virtual silence about solutions in policy and political circles.

That, I suspect,  is because no one has any ideas that are, for now, politically palatable (Drum lists some policy options, all of which are on—to use dully conventional labels—leftish, but they merit much more than a look, even if only to think through why they might be wrong–and what the alternatives might be).

Drum also knocks down the argument that this automation wave will work out fine, just like all the others.

The Industrial Revolution was all about mechanical power: Trains were more powerful than horses, and mechanical looms were more efficient than human muscle. At first, this did put people out of work: Those loom-smashing weavers in Yorkshire—the original Luddites—really did lose their livelihoods. This caused massive social upheaval for decades until the entire economy adapted to the machine age.

Well, yes.

As I’ve mentioned a few times before, it’s worth reading about the ‘Engels Pause’. British working class wages stagnated for half a century or so after the first industrial revolution, and that did indeed lead to major social upheaval.

Now imagine what will happen when those at the losing end of this latest automation wave include just about everyone, including the best and the brightest, people who always assumed that the only way ahead for them was upwards (I wrote about that on NRODT here). They will not go quietly into the dole queue.

It’s also worth remembering Moravec’s paradox. I quoted the Guardian’s Larry Elliott on that topic in a post here:

Robots are likely to result in a further hollowing out of middle-class jobs, and the reason is something known as Moravec’s paradox. This was a discovery by AI experts in the 1980s that robots find the difficult things easy and the easy things difficult. Hans Moravec, one of the researchers, said: “It is comparatively easy to make computers exhibit adult-level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility.” Put another way, if you wanted to beat Magnus Carlsen, the world chess champion, you would choose a computer. If you wanted to clean the chess pieces after the game, you would choose a human being.

In the modern economy, the jobs that are prized tend to be the ones that involve skills such as logic. Those that are less well-rewarded tend to involve mobility and perception. Robots find logic easy but mobility and perception difficult.

Back to Drum:

The AI Revolution will be nothing like [earlier industrial revolutions]. When robots become as smart and capable as human beings, there will be nothing left for people to do because machines will be both stronger and smarter than humans. Even if AI creates lots of new jobs, it’s of no consequence. No matter what job you name, robots will be able to do it. They will manufacture themselves, program themselves, repair themselves, and manage themselves. If you don’t appreciate this, then you don’t appreciate what’s barreling toward us.

Read more