Permanent Underclass?

 The New Yorker yesterday informs me that a growing number of online commentators and Silicon Valley tech influencers believe the age of AI automation is finally at hand. Pretty much regardless of our field, they say, we are all about to be replaced by machines, and consigned to a permanently unemployable "underclass," like so many Silesian weavers. 

This wave of automation will come in inverted order from how it unfolded in the early nineteenth century. Instead of displacing hand work—this wave is going to start with brain work. 

The first things to go will be white-collar jobs involving the processing and manipulation of data and concepts. Jobs like writing email messages and digital communication pieces (like my current métier) will be on the chopping block early. 

But eventually the "demon of mechanism" (as Thomas Carlyle put it at the dawn of the Industrial Revolution) will come for us all: "changing his shape like a very Proteus; and infallibly at every change of shape, oversetting whole multitudes of workmen."

How—you might be wondering—will AI chatbots gifted at generating text be able to replace shelf-stockers at Walmart in the near future? How will ChatGPT or Claude ever come for manual harvesting or repair jobs? 

The way this is supposed to go—according to the true AI automation doomers—is that eventually the machines will be so smart they can replace the jobs of even their own AI engineer handlers. Then, the chatbots will be able to program themselves. 

If it seems like a stretch to us right now to imagine the current generation of LLMs making the qualitative leaps required to evolve the ability to manipulate the physical environment around them—that is no doubt due to the limitations of our human brains. Once the LLMs themselves are liberated to agentically address these problems by redesigning their own capabilities—then they will start figuring out how to automate all the lingering forms of manual labor as well. 

All of which is possible, I suppose. But the AI engineers have been proclaiming this imminent arrival of agentic AGI and self-building machines for some time now—and the day of reckoning keeps getting pushed back (as it does in many another doomsday cult). According to the New Yorker article—the current standard wisdom on the panicky corners of the internet is that AGI is coming for us in 2027—which sounds frighteningly close. But in the year 2023, we would have said we would all have been eclipsed by machines by 2025. Yet here we are—and the world is blighted in many ways, but not that one, particularly. 

Indeed, I was kind of surprised to see that article in the New Yorker yesterday—and to read that so many people on the internet are freaking out right now about the idea of a "permanent underclass" created by AI automation and displacement. After all, I thought that the era of peak panic about AI—2023 or so—was behind us now. Many had settled into a comfortably blasé attitude toward the new technology. Every day brings a growing chorus of economic observers who warn that we may be in an AI bubble—and that the excitement over the impressive new technology is not actually translating into profits or productivity growth. 

Indeed, the most visible way in which AI has recently affected our lives is the arrival of Sora 2.0. It's an app that allows you to create 9-second videos of yourself doing things like parachuting by means of an oversized pizza. 

I have not found that an interesting enough prospect so far to be moved to try it. 

There seems to be a disconnect here between the dire "any day now" warnings of hyperintelligent autonomous AGI and the reality of mundane slop the technology is actually producing. 

I don't mean by this to join in the brigade of people who snidely dismiss the new technology. I have some sympathy with Auguste Villiers de l'Isle Adam's ficitonalized version of Thomas Edison, who (in the author's eerily prophetic novel about robotics and the rise of AI romantic "companions"—a reality Elon Musk is now apparently trying to bring into being—The Future Eve) complains that every new technology is mocked in its early days for its minor and inevitable imperfections—by people who never could have invented it in the first place. 

And indeed, I am still wowed by the LLMs we have already. I don't wish to minimize how extraordinary this technology actually is. We now live in a world in which you can have a plausible, fluid, verbal conversation with a machine mind—rather as if you were talking to a Droid in Star Wars. That was not true of the world I was born in—but it's true of the world today; the world my niece and nephew will inherit. That's a profound qualitative leap in the human experience that shouldn't be gainsaid. 

My problem with the AGI doomsday scenarios though—is that they assume that merely because there has been one such qualitative leap in the technology in our lifetimes, there will be several more of them; indeed, many more, all exponentially accumulating and building on themselves. That has not been true of prior waves of technological innovation, however—which tend to follow instead a boom-and-bust cycle, interspersed with long plateaus of minimal activity. Why shouldn't this be another of the same kind?

The AI automation doomers—though—say that we aren't just talking about a future scenario here. The wave of technological displacement is already upon us. They point to declining jobs numbers, the difficulties faced by recent computer science grads in landing employment, etc. 

And indeed—this may be in part due to AI automation. But before we jump to that conclusion—I can think of a more mundane and obvious explanation: how about the fact that Trump is destroying the economy? 

We seem to be risking a repeat here of what happened in the Obama era—when persistent jobs shortages lingering from the Great Recession caused many pundits to speculate that automation may have irretrievably reshaped the labor market—such that the U.S. economy would simply never again create enough jobs to reach full employment. 

Just a few years later, though—the U.S. economy had reached full employment again; and then some. Indeed, part of the reason we had so much inflation the last few years was because the economy had too many jobs to go around; we had labor shortages—something that would have seemed unthinkable circa 2012 or so. 

The reason for this turn-around had nothing to do with automation. It had to do with the fact that Republicans—once they were in power—no longer felt the need to actively stymie economic growth and to block the fiscal stimulus that was needed to pull the economy out of its long post-2008 slump. 

What was vaguely attributed to technology, then, turned out to really just be a byproduct of Republican policy stupidity. 

I fear we're in a similar mess today—albeit the GOP policies causing the morass are slightly different. Now, Trump has imposed tariffs that are artificially increasing prices across the economy. This keeps inflation elevated, which in turn has made the Fed (rightly) hesitant to lower interest rates too precipitously. And these elevated interest rates have in turn slowed employment growth—as they always do (indeed, as they are meant to do). 

The lousy jobs numbers and small amount of new openings in the economy right now are a predictable effect, then, of living through a period of prolonged elevated interest rates. Indeed, the miracle is that we haven't long since tipped into an outright recession (which is what usually happens when the Fed raises rates this much). The wonder is not that jobs numbers are so bad, but that they are so good. 

And the Fed could make them even better—if they didn't have to fear that cutting interest rates too quickly would exacerbate Trump's tariff-induced inflation. 

So before we jump to the conclusion that the age of AI automation has arrived—let us first explore the more mundane and solvable possibility that Trump's sabotage of the economy could be causing these woes. 

If more Americans are indeed being thrown into an unemployable "underclass"—then, it may not be a "permanent" one, where the labor of their hands and minds has been eternally replaced by machine intelligence. It may, rather, be a perfectly temporary one—created by Trump's mammoth and easily-preventable stupidity, malice, ignorance, and incompetence. 

In his play about technological displacement, set in the period of the early Industrial Revolution, The Machine-Wreckers, the radical German playwright Ernst Toller suggests that the 19th century Luddites were right to rebel against the new industrial capitalism—but they had fingered the wrong culprit. They vented their rage on the machine, but this—in his view—was misguided. The machine was not the enemy; indeed, it could be a friend. It could be used to make life easier and reduce our horrid drudgery. 

(As Oscar Wilde once put it: "To sweep a slushy crossing for eight hours, on a day when the east wind is blowing is a disgusting occupation. To sweep it with mental, moral, or physical dignity seems to me to be impossible. To sweep it with joy would be appalling. Man is made for something better than disturbing dirt. All work of that kind should be done by a machine.")

The problem, in Toller's view (and in Wilde's too—in his essay "The Soul of Man Under Socialism"), was not the machine but the machine's owner—who was using the machine to impoverish the many for the sake of the enrichment of the few. 

And so too, today—we may be right to be outraged by the unemployment and displacement we are facing. But let us be sure we are blaming the party that is actually responsible. Instead of pointing the finger at LLMs and chatbots—let us see if the source of our problems is not in fact someone operating on a far more human scale: someone like the orange man in the White House.

If we are in fact to be displaced and turned out into the streets like so many Silesian weavers—then let us do as Heine's weavers did—and put the blame where it belongs: on the politicians that are doing this to us. And let us weave their electoral downfall.