Sam Altman's lamplighter

"Let’s consider the set of adult human beings who are unemployable at OpenAI itself."

Sometimes Sam Altman, the new young Jupiter of AI, makes it easy to see why—to borrow the title of an essay I wrote ten years ago—he is a blithering idiot.

This is of interest not because Sam Altman is a blithering idiot—in fact, I hear he’s done rather well for himself—but because this miniscule, yet profoundly crippling, logical lesion in so finely-tuned a lipid-based thinking machine is probably not confined to Sam Altman’s brain.

No—whatever the germ (could it be Elon Musk’s “Woke Mind Virus?”) that causes this lesion, it has infected others. In fact, it has infected everyone. It had infected me. (I took my own noötropic tablets, and was cured. Click here to buy.)

Let me explain this simple fallacy using Altman’s own words, posted the other day. This is the end of his effusive essay on the “Intelligence Age”:

Many of the jobs we do today would have looked like trifling wastes of time to people a few hundred years ago, but nobody is looking back at the past, wishing they were a lamplighter.

If a lamplighter could see the world today, he would think the prosperity all around him was unimaginable. And if we could fast-forward a hundred years from today, the prosperity all around us would feel just as unimaginable.

Sam Altman is right about all these statements. Let us grant that he is exactly right! Every sentence is exactly correct. Yet he remains a blithering idiot, because he will never be able to follow me—and us, if you are not afraid—through the following logic.

(If you are afraid, it’s not too late to ingest the antidote! Hit the back button. Now!)

The problem with the Intelligence Age

Sam is right: no one is looking back at the past, wishing they were a lamplighter. But:

I am looking straight at the present—wishing other people were lamplighters—now. You should be too. And if your philosophy of society was properly adjusted, you would be. Here is what you would be thinking. I am pretty sure it is not what you think now:

Other people, today, should be lamplighters. Many, many more people, in the future, should be lamplighters. Because of AI. (And other technologies, obviously.) And this is not because they wish to be lamplighters. No: the state must make them lamplighters.

Wow! I can’t even! Wow, just wow. I must be such an evil person. With such evil ideas. Do you hear that? It’s the disease talking. The mind virus (or bacterium, fungus, prion, neoplasm… let’s keep the diagnosis open here) wants you to think this. Snap out of it!

Let me make two assertions to explain this evil thought.

Consider the set of human beings, Z, whom I wish were lamplighters (or pursuing some other artisanal craft—lamplighting is kind of cool, but we can probably invent better neo-traditional trades).

Assertion A: if people in Z were lamplighters, instead of what they are, their externalities—their external impact on people not in Z—would be more positive.

Assertion B: if people in Z were lamplighters, instead of what they are, their internalities—the quality of their own human experience—would be more positive.

We see clearly that if both A and B are true, my evil proposal is good for everyone. This is not my opinion. It is logic. The union of Z and non-Z is everyone.

It is good for both So if both A and B are true, whatever makes you think I am evil is evil itself.

Sam and all his GPUs can chew on this logic till the end of time. It will still be true. Please don’t hate the messenger! I hope this news won’t change your life too much.

The context of assertion A

Of course, the truth of A and B is dependent on the definition of Z. This definition is a complex question, but let’s start with a simple one that others have already coined.

By others, I mean of course the great libertarian economist Tyler Cowen, who defines “zero marginal product” (ZMP) human beings as:

Individuals who find it hard to get a job at all because they are perceived as just not having that much to offer at any wage.

There are many individuals who are obviously in ZMP. Consider my two-year-old son. While he does seem very precocious, even flipping burgers would be far beyond him. But sadly, this is true of many biological adults as well. As Cowen notes:

Other people, you may not, you know, even want them flipping burgers.  They’re just destructive. 

And my point is just the better we get at measuring value, which you will get with surveillance and software and just plain flat out right measurement, the more we will discriminate against those people. And that is a social problem.

Rather counterintuitively, given that it can already ace the Math Olympiad and shit, deep learning is not yet good at flipping burgers. But it will be. And that, I’m afraid, Mr. Altman, is a social problem.

Let’s call our set Z the set of adult, neurologically normal humans in ZMP. If you truly believe that Z is empty… I despair. The virus has just done too much damage.

But wait! I have a trick. Let’s consider a similar set. Let’s consider the set of adult, biologically normal human beings who are unemployable at OpenAI itself.

After all, every time a company refuses to hire someone, it declares them ZMP from the perspective of its own employee pool. A true ZMP adult is unemployable in any firm or by any private individual.

Could there be continental human populations which are ZMP with respect to OpenAI? Sure. Google has an Australian office. How many full-blooded Aboriginal Australians work at Google Australia? Bueller? Bueller? Ferris was supposed to be in class today. Maybe the dog ate his explanation. Clearly, all men are created equal [citation needed].

Not every workplace is Google or OpenAI, but every workplace has a ZMP threshold. Generally that threshold is lower than Google’s. What AI will do—what Sam Altman will do—is to vastly increase the ZMP threshold in almost every workplace.

Once AI finds its physical groove, AI-powered robots will take ten or twenty years to eliminate all demand for menial human labor. The strawberry pickers will follow the lamplighters; the landscapers will follow the strawberry pickers; and the machinists will follow the landscapers. All skills for dealing with the physical world will vanish.

In the end, 20th-century employment as we know it will reduce itself to ten engineers, who write the code that writes the code that writes the code. They will all work for Sam Altman and drive ludicrously amazing flying cars. And what of everyone else?

We (I am anything but an AI engineer—good luck with those matrices, kids) will just have fun, right? Life will just be fun?

No, actually, we will probably all have our throats cut. Here’s why.

When everyone and everything is free

Sam Altman has a plan for this work-free future. He calls it UBI. Yes, kids, we have found the telos of the industrial revolution: fully automated luxury communism. This is obviously good. And my evil, evil state lamplighter plan is the polar opposite of UBI.

First of all, the UBI scenario (if we accept it) has also already proved assertion A. Once we say that UBI is the solution for all these ZMP people, we say that the non-ZMP set has to subsidize the ZMP set. Which means the ZMP set has negative externalities on the non-ZMP set. So Z is bad for non-Z. QED! If a little easy.

But you have an easy reply: under fully automated luxury communism, production is so easy that no one needs money. The last ten engineers aren’t even paid. No one is paid. The robot factories make whatever you want. That’s why it’s called communism. Even the writers are robots! The future is all Cory Doctorow: free stuff, free love, free speech. Everything is “free as in beer.” So all this UBI has no real cost to anyone.

Communism! People love the brand, but they don’t remember trying it last century. Marxist-Leninism, which killed 100 million people, was (I believe) the political result of 19th-century and 20th-century destruction of labor demand at the hands of the Industrial Age. Whatever the causality, the two sure happened at the same time.

And that destruction of labor demand does not hold a candle to the upcoming rise of the ZMP threshold in Sam Altman’s “Intelligence Age”—which, if past performance predicts future results, will most certainly guillotine us all. Thank you, OpenAI!

Like the industrial age, the intelligence age represents the final victory of capital over labor. Capital assumes its final form: Sam Altman’s trillion-dollar planetary silicon brain farm. Capital is the supply chain for matrix multiplication.

The only human jobs are jobs no robots can do. The last profession is the oldest profession. In Altman’s Elysium, all the men are coders and all the women are sex workers. So East Bay. So… what’s wrong?

So: as capital renews its final assault on labor, we’re about to.be hit by a wave of 21st-century Marxism that makes the Russian Revolution look like a garden party. Strap in.

Fully automated luxury Stalinism

While we have established the negative economic externalities of UBI, if in a possibly trivial sense, we are neglecting the political externalities of UBI.

The Russians of Stalin’s era enjoyed many luxuries that the Russians of Nicholas II’s regime could only have dreamed of. Their GDP was surely far higher. Probably no one in Czarist Russia had even thought of calculating GDP. Whereas the Soviet economists, like our own, were into GDP. And their numbers looked great. Stalinism was a success.

The problem with communism isn’t that it doesn’t work economically. Well, that’s a problem. It’s not the problem. The problem is that communism doesn’t work politically. Free labor from AI solves the economic problem. It exacerbates the political problem.

Even if the robots make all the free stuff we need, someone controls the robots. Even once you don’t have to work and get a free check to spend on whatever you want, someone has to sign the check. All legal action is human action.

As Albert Jay Nock once wrote, there are two ways human beings feed themselves: by the economic means, or the political means. The economic means is the production of goods and services needed by others. The political means is the capture of goods and services produced by others.

In the age of AI and UBI, almost no one can feed themselves by the economic means, because almost everyone is below the ZMP threshold. So everything becomes political.

Whoever signs your UBI check is inevitably your political patron. Whether this party is a person, a corporation, a church, a nonprofit, or a government agency, it can literally kill you by not signing a piece of paper—or transform your life by adding a digit. You would be an idiot not to delegate all your power to it. And very few adults are idiots.

Universal UBI means the universal purchase of votes. It means every human population—especially groups below the ZMP threshold—becomes what Indian political scientists call a votebank. Any actual democracy becomes the property of some power structure.

Here is a proposition you will never see falsified: when dependent populations vote, they always support the same party as the officials of the institutions they depend on.

An economic dependent cannot help but be a political dependent. The weak cannot be strong. But they can serve the strong. Under universal dependency, the idle poor become the political armies of the idle rich:

Now, in this year’s presidential race, the wealthiest woman in Silicon Valley has emerged as a powerful player behind the scenes. She has quietly contributed millions of dollars to an organization backing Ms. Harris, according to three people briefed on the gifts.

She played a hidden but key role in helping usher Mr. Biden out of the race, which cleared the way for a Harris run.

And Ms. Powell Jobs, who is so close to the vice president that her staff refers to her simply as “L.P.J.,” is positioned to have extraordinary influence, or at least access, in a potential Harris administration.

(Wouldn’t it be nice if our newspaper of record could tell us more about this “hidden but key role?” What is this, the Most Serene Republic of Venice? Could Ms. Powell Jobs have perhaps poisoned her aging Doge with a perfectly-blown dose of blowgun venom to the urethra? The frog-dust travels straight to the kidneys and thence to the brain, causing progressive deterioration, a wobbling gait and confusion of dignitaries. The situation in which both the Doge and his younger and comelier replacement are too embarrassing for public view is especially choice. Maybe we do need four more years. This is very late-stage stuff, kids: you have to remember the thing when it was young.)

Leftism is always and everywhere an aristocratic force. Perhaps this is most clearly seen in the Labor Party UK, which from its roots in the Ruskinites and Fabians has always been the political vehicle of the university Left. Across the last century, this vehicle has switched its fuel from the disappearing British worker to the burgeoning British immigrant—without any substantive change in the nature of its leadership! This is an absolutely wild way to use the word “democracy.”

An aristocracy out of power becomes a bureaucracy in power. H.G. Wells’ Open Conspiracy becomes a mandatory global ideology—an oligarchic despotism, utterly unlike the classical monarchical tyranny of 20th-century Stalinism, yet strangely reminiscent of it in so many ways. The two even used the same word, “progressive”—and, in the lives of those now living, were a military alliance that conquered the world. How exactly has no one noticed this?

Here is the massive negative externality of AI: the political impact of universal UBI. The future is a new wave of dangerously incompetent, arbitrarily cruel, distributed bureaucratic despotism. Everywhere, voters who oppose this regime are missing their last chance to stop it, as they become permanently outvoted by welfare votebanks.

UBI is redistribution of wealth, and redistribution of wealth is not an economic idea. It is a political idea. We know its historical track record. And we know its body count—which makes Hitler look like a rookie. And like moths we return to its red flame. Nerds! Fellow nerds! What is up with us? Why do we keep doing this?

There is a simple way to understand the economic and political impact of AI: the “resource curse.” Windfalls in oil, gold, etc, have demonstrated to economists that massive economic profits which create little or no labor demand have a paradoxical negative impact on economies. Venezuela, which has the most oil in South America, has the worst economy in South America.

Resource dividends and technology dividends are the same thing. Excess dividends create gross economic gain and net economic loss. They create net economic loss because they create political damage, which creates economic damage. They create political damage by transferring human beings from the economic to the political means—by making them resource capturers, not resource producers. Technology may be able to support a society of resource capturers. But political science is not.

AI will turn the whole world into Venezuela. AI will turn Palo Alto into Venezuela. Laugh while you can, monkey-boy! The train from Aragua is coming to you. Sam Altman, richer than Batman’s dad, will sleep till masked, tattooed faces wake him.

The Wildnil economy

As if this grim glance at the crystal ball wasn’t bad enough, it gets worse. Assertion B claims that the elimination of labor is bad for the laborers themselves, despite their inherent wish to not labor—whether as lamplighters or Javascript programmers.

Economists define labor as a form of disutility, done only to create offsetting utility. When, except for the last ten engineers, all mental labor is done by AI, and all physical labor is done by robots, the rest of us will just have… fun?

Here is one trick to see the future. From the standpoint of the domestic American economy, the difference between outsourcing to robots and outsourcing to China is negligible. China’s impact is a preview of AI’s impact.

The technology that produces the highest marginal return in GDP, that is, produces the maximum amount of fun as defined by market price, is well known. Like many of China’s wins, it is in material science: atoms, not bits. Chemistry, to be exact. It has the highest fun-to-weight ratio ever recorded: ten micrograms will keep you happy for hours. This material is a (literal) elephant tranquilizer called carfentanil, or “Wildnil.” (Just don’t give your elephant more than a milligram, especially if he’s been drinking.)

China also exports more indirect ways of having fun. For instance, these days they make the best, cheapest electric cars. Electric cars are not as fun as Wildnil, especially by weight. But how else will you sustainably get to your dealer? Why… the question answers itself. The Wildnil will deliver itself—in a self-driving electric car. Coming to you—the three-way marriage of sustainability, chemistry, and artificial intelligence.

I believe that any theory which says the best economy is the Wildnil economy is wrong. Self-driving Wildnil from China—a remarkable rhyme with the Opium Wars—is not a curious corner case, but a reductio ad absurdum which shows the absurdity of the whole theory—that is, of libertarian, utilitarian “GDP” economics as a whole. Since no other economics is known in our society, we find ourselves in rather an epistemic pickle. Sad!

I said that everything in that Sam Altman quote is true. It is true, in the sense that he meant it. But there is one very important problem in emphasis.

Our “prosperity” would not in fact impress our ancestors from a hundred years ago. If they could see our country—from our bustling cities, to our busy small towns, to our charming little villages—they would not find us to be “prospering.” Their eyes would see desolation and misery they could have never imagined—from tent encampments in Oakland to fentanyl shacks in Ohio. In every home, shack or tent—a Playstation.

And I haven’t even mentioned the porn! Consider our porn GDP in constant dollars—either by estimating the net price of all 1924 porn in the 2024 market, or vice versa. Such prosperity! I bet Noah Smith has already thought about this, hard.

The problem is simple. It is the confusion of two words: “prosperity” and “luxury.” Our ancestors, who, like, read the Bible and stuff, did not have this brain damage. They would see that while we are raddled with luxury, we have very little prosperity. Electric cars are a luxury. Porn is a luxury. Fentanyl is a luxury.

Three solutions

What is the difference between luxury and prosperity?

At the lowest levels of Maslow’s pyramid of needs, they are the same. Both luxury and prosperity start with air, proceed to water, then food, then safety. (40% of Americans are afraid to walk alone outside at night where they live—and more might be, if they weren’t more worried about long-term changes in the temperature of the planet.)

Here, luxury and prosperity diverge. Prosperity is what is good for you; luxury is what feels good to you. Luxury serves your present self; prosperity serves your future self.

The greater the productivity you can command, the further apart these targets are. A rockstar can choose to live like an LA rockstar or like a Hawaii surfboard shaper. One of England’s most successful and aristocratic actors, Daniel Day-Lewis, spent two years learning to make shoes in Italy. There are other shoemakers in Italy, but not as many can win an Oscar.

This is true for societies—for countries—as well: as their productivity increases, their gap between luxury and prosperity increases. When a government uses GDP as its economic target metric, it is managing the economy for luxury, not for prosperity.

How can we shift this frame? In three ways: the reactionary way, the liberal way, and the neoreactionary way.

The reactionary way to note that the state is God’s vicar on earth, and its role is to do unto its subjects as God would do to each. A liberal will complain that he has no idea what this means, because he has no idea what “God” means. The reactionary is like: yes, actually, you do. This formula worked perfectly for many years and there was no real need to abandon it, but that does not mean we can revive it.

The liberal way is to define prosperity as “human flourishing,” and ask what it takes to make every human being under the state’s umbrella flourish. The answer is sometimes “more stuff and more fun.” But not usually.

Generally, for human beings to flourish, they must be challenged to the best of their abilities. Otherwise their potential remains unfulfilled. Therefore, the problem of managing an economy, from the perspective of the state, is the problem of making sure that economy demands that everyone maximize their human potential through challenge—that is, through meaningful work that makes them grow as human beings.

For many people this work is menial labor and for a few it is advanced physics, but the sweet spot for enormous numbers of people is artisanal labor. For instance, making shoes is artisanal labor. Since the Industrial Revolution long ago destroyed the market for artisanal production, it is no loss to see industrial production (and its Dickensian human robots, pulling levers all day) destroyed by fully automated factories. The same is true of the “bullshit” office email jobs that Altman calls “trifling wastes of time.” Presumably, except in the rapidly burgeoning compliance department, there are no “trifling wastes of time” at OpenAI.

Therefore, it is part of the government’s job to shape labor markets so that there is a rough correlation between labor demand and labor supply, with professions where everyone is challenged according to their human capacity. In an era with productivity to burn, the state can and should burn productivity to shape labor demand.

Curating a system of professional trades (a traditional role of the state for centuries, going back to the medieval guild systems) will often mean burning productivity, that is, adding artificial difficulty.

The good news for Sam Altman is that AI, and new technology in general, is not in general the right target for artificial difficulty. The right target is old technology—to restore the first wave of professions destroyed by the Industrial Revolution, the old artisanal professions. It’s not clear if lamplighting is a particularly skilled job—it might be more of a Down’s syndrome thing, frankly, like bagging groceries—but if it isn’t, shoemaking is.

When we think about “neo-Luddism,” we shouldn’t think about banning AI. We should think about a world where, if an American wants to buy clothes, he or she will have to buy clothes made by hand, in America, from natural American materials. Everything is (old) Etsy. Presto: 10 or 20 million challenging artisanal jobs.

Finally, there is a neoreactionary way to think about artificial difficulty, which is too spergy for anyone but Sam Altman to understand, but should seem natural to any real corporate madman. This is the human capital standpoint.

For a neoreactionary, a government is a sovereign corporation whose capital is its land and its people (plus buildings and machinery). This analysis neatly distinguishes between luxury and prosperity: luxury does not make a human being valuable (rather the opposite, alas), where as prosperity does.

In fact, whatever we mean by the value of a human being, it is corroded by idleness and luxury, and enhanced by practice, training and discipline. As the Victorians knew well, but we persist in forgetting, a lamplighter is a more valuable human being than a welfare recipient. Lamplighting may not be useful in of itself, but the lamplighter can be reassigned to many tasks involving diligence and care. He is better for the world and he is better for himself. But he has less free time to inject elephant tranquilizer. Which way, modern man?

There is a tiny problem with this metric: it is hard to measure. We can measure GDP, but not human capital. Well, actually, we can measure human capital. We just haven’t wanted to, not for the last 160 years or so. Slavery is bad, ok? But suppose there was a lottery, in which 1 out of 10,000 people lost and had to be randomly sold, proceeds going to charity of course, just as an econometric design… naw. Too George Mason.