These were my opinions, and they continue unaltered, as far as regards the landlord and the capitalist; but I am convinced, that the substitution of machinery for human labour, is often very injurious to the interests of the class of labourers.
David Ricardo, “On Machinery”, 1821
A famous take on capitalism is that it is an outlier: a world of exponential growth closer to Taleb’s Extremistan, very different to most of our history, which had, at least since the Neolithic revolution (itself a huge jump in productivity and a time of great upheaval), been linear. There was almost no per capita growth. A successful startup took the form of a raiding party, and capitalism’s obsession with productivity, with adding value, was nowhere to be found. If anything, there was an aversion to inventions that could put people out of work, since idle hands meant unrest, and unrest could topple the established order. It was a very different world to ours.
Merit was far from commonplace: our ancestors lived in more static societies (with the occasional upheaval) or wandered their outskirts like the Cossacks. In fact, of all places, it was the church or the state bureaucracy where merit was likely to count the most. Industrial capitalism changed the paradigm: it funnelled the competitive and adventurous spirit of humanity into ingenuity, and what the Roman emperors had feared became sacred. Enterprises were built on the promise of making humanity more productive, and commodities came to be produced by means of ever more intricate commodities. Then, in the second half of the twentieth century, came the promise of going one step further: of decoupling the torrent of human knowledge from the humans themselves. We made machines in our image, machines that learn about the world, because we thought this would be useful. What seemed a reductio ad absurdum a few years ago is our reality now, and it puts the outlier itself in question.
Some years ago I found a wonderful book that described industrial capitalism’s inevitable torrent of social learning and what it does to a society: Structural Economic Dynamics (1993), by Luigi Pasinetti, a Cantabrigian economist (Cambridge, England, not Massachusetts: a difference of method more than of geography). It was a beautifully crafted apparatus that left everything unnecessary out of the picture to focus on one simple question: what does growing productivity imply for human societies? One very Keynesian implication is that if productivity keeps growing, so too must consumption, or else people end up unemployed. So consumption becomes moral! But Pasinetti also saw that consumption can only go so far. There is, after all, only so much toothpaste a person needs. So he focused on structural change: on what societies produce and consume, not just on how much.
Productivity in n sectors seemed to lead, automagically, to the creation of n+1 sectors. That is one endogenous way out of the conundrum, and in political terms the libertarian one: in spite of technological change, all is well with the world, and the limits of the system are the limits of human ingenuity. Pasinetti also took the other route and showed that productivity gains can be “harvested” simply by working less, either by retiring earlier or by working a shorter week. To an American Republican this would look like a socialist European invention rather than an emergent property of the market. Yet one level up, both solutions are emergent: two answers to the same problem, growing productivity, given through different institutions.
Enter AI, and the promise of the singularity, AGI. In a working paper I have just put out, I use the framework Pasinetti laid out to push the parameters to their extremes and see what happens.
Its first result is a reminder of why the old solutions worked so well, and why they may not work again. For two centuries, the people that machines released from the fields and the factories went into services, where productivity grew slowly and demand kept growing with income: teaching, care, health, administration and the paperwork of every other sector. William Baumol called the rising cost of those services a disease. It was also what kept us employed, because productivity grew slowest exactly where demand grew fastest (Proposition P3). There were other valves, too. For over a century Europe shipped its surplus people to the “new world” by the tens of millions, as Prabhat Patnaik reminds us, and in a bad year you could still go back to the farm (a third-generation New Yorker might manage the farming part; I wouldn’t bet on the subsistence part). So anyone who says we have always found new jobs is looking at a plane in flight and concluding that gravity is a myth: the data, after all, clearly show a machine in the sky. For two centuries the fuel was services, helped early on by emigration and, to an ever lesser degree, the farm. AI seems poised to land hardest on the first of these, services and the white-collar work inside every other sector, and if productivity outruns demand everywhere, no reshuffling of what we buy brings the jobs back (Proposition P4).
Which brings me to the singularity in the title. The whole apparatus hangs on a single condition: add up, across everything people buy, the work it takes to make what one person consumes, and for everyone who wants a job to have one, that total has to equal the work one person offers (Proposition P1), which is also the condition for everything produced to be bought. In the model this is quite literally a singularity, a determinant that has to vanish. It is not a forecast; it is a yardstick. And AGI, the singularity everybody is talking about, would be the most thorough way yet devised of breaking it, should it ever arrive.
Now push Pasinetti’s two solutions to the extreme. The n+1 sector still arrives, but nothing says it needs anybody: a hundred new industries with zero labour coefficients absorb nobody (Proposition P11), and if AGI is what its prophets promise, the new industries may be the first to run without us. There goes the libertarian dream. The European one survives on paper: instead of a quarter of the population out of work, everybody works three quarters of the time (Proposition P10). But hours have to fall as fast as the need for work does, and one person’s hours have to stand in for another’s, which in a specialised economy they often can’t. A cardiac surgeon cannot hand Thursday afternoon over to a paralegal.
At the very extreme, the results get stranger. Elon Musk has said that money won’t matter by 2036, and inside the model he is right: when goods need no labour, their prices go to zero and they become free gifts of nature (Proposition P7). But only because nobody in the model owns the machines. Once someone does, a fully automated good keeps a price, depreciation plus profit, and ownership decides who gets it (the paper, §8), which is the main limitation of this simple model and a job for a sequel. The opposite fear, that everything collapses, is wrong too. The economy does not collapse; it settles into what a Keynesian would call an underemployment equilibrium, in which some work and the rest are supported out of what they produce (Proposition P12), and nothing in the model pushes it back out. Dario Amodei described that state better than the model does: “Cancer is cured, the economy grows at 10% a year, the budget is balanced — and 20% of people don’t have jobs” (Axios). A Star Trek economy is a grand vision that makes strange bedfellows of Musk and Varoufakis; achieving it is an altogether different proposition. The transition rests on the people still at work supporting everyone else, with the share of each worker’s time that goes to others rising towards all of it before abundance arrives (Proposition P13). Europe, where work is already heavily taxed, would start down that road with less slack than it thinks. Hence the political in political economy.
So what does a society do when the market no longer needs everybody? Much what it did before industrial capitalism. One option is a return to the mandarin: bureaucracy takes centre stage, not that it ever really left. Wherever a person must answer for a decision (regulation, liability, audit, compliance), the work a good requires cannot fall below a floor, however good the machine. Europe has already written one into law: Article 14 of the AI Act requires high-risk AI systems to be designed so that they “can be effectively overseen by natural persons”. Take that far enough and the whole world turns into Brussels, with a senior compliance post the best job going. The price is a culture of credentials, the examination halls of imperial China rather than a garage in Palo Alto, and an economy that comes to rest: with a floor in every sector, measured growth goes to zero while output per head sits at its peak (Proposition P9), a new road to the classical stationary state that John Stuart Mill rather looked forward to. If you believe in utopia and a smooth road to it, floors only delay a free gift. If you don’t, they are a safeguard, because they keep work, and with it a claim on output, in human hands.
The other option is the dole. A few years ago Varoufakis argued that today’s economy is closer to feudalism than to capitalism. I would reach for a different historical analogy with problems rather closer to ours: Rome. In the classic account of the late Republic, conquest brought slaves in great numbers (the ancient world’s “labour-saving technology”), the slaves displaced the small farmers, and the farmers drifted to the city. Rome fed them bread and entertained them with games, Juvenal’s panem et circenses, and, under the emperors, when that did not do the trick, it sent in the praetorians: bread, games and the occasional gladius. These were the plebs frumentaria, the citizens on the grain dole, whose claim on the city’s output was a ration rather than a wage. The emperors understood the arithmetic of power. Suetonius tells of Vespasian rewarding an engineer handsomely for a device to haul heavy columns up to the Capitol at little cost, and then declining to use it because, as he put it, he had to feed the common people. Emperors feared idle hands.
The dole was never an indulgence; it was demand. Somebody has to buy what the workers make, and once they make more than they use themselves (remember the toothpaste), the rest has to be bought by people who do not work: the plebs on their dole, or the elite. Take both away and nobody works, not because the machines do it all but because nobody is buying. This is where the model lands, and it is an old argument: Malthus’s case for unproductive consumers, “a considerable class of persons who have both the will and power to consume more material wealth than they produce”, without whom nobody could go on producing more than they consume at a profit. Keynes took his side, and noted that he had opposed Ricardo on it vehemently but vainly. A basic income, in the model, is exactly that buyer, and what it buys decides who works (Proposition P14). But in Rome, bread, games and the sword all took people. In an AGI economy, all three can increasingly be made by machines, which makes them cheap and employs nobody, and a technology this good at entertaining, watching and sorting people is an obvious tool for keeping order should the road get rough. The plebs have far less agency than workers did, since nobody depends on their work, and Amodei, again, put it as well as anyone: “The balance of power of democracy is premised on the average person having leverage through creating economic value. If that’s not present, I think things become kind of scary.” Nor was the dole ever free. Rome’s grain came from the provinces, as tribute. Who is the province this time? The people still at work, other countries buying our exports, or the owners of the machines (Amodei has floated a “token tax”).
So perhaps the capitalist outlier, an age of idolatry of productivity, ends and we go back to something older: bureaucratic compliance structures for the ambitious and connected, bread and games for the have-nots, and for the restless, like the Cossacks, the edges of the map, which Elon seems hellbent on extending to the rest of the solar system. If production no longer confers standing, the contest for rank leaves the market, and where it goes is anyone’s guess. Somewhere banal: a status feed in which the extremes, by dint of their looks, their behaviour or just their unconventionality in general, crowd out the average. Somewhere Keynes would have liked (and the socialists too, who at heart want us all to become a race of noble aristocrats, each of us a Leo Tolstoy, who decided he must be in love because he had bought a horse the day before for no reason at all), with people learning “how to occupy the leisure, which science and compound interest will have won for him, to live wisely and agreeably and well”. Somewhere more daring, with the startup becoming a raiding party once again. Or trade stays at the centre, in whatever human goods are left to trade, and Smith turns out to have been right all along about our “propensity to truck, barter, and exchange”. These are pictures, not forecasts. If AGI does arrive, the economic changes may not even be the largest ones.
Pasinetti saw in 1993 that the Promethean gift is not free. It reshapes the world around it, enchanting some and disenchanting others, and it demands action from both alike:
It imposes, at every moment in time, not just the possibility, but the necessity of making choices. There does not exist the alternative of not choosing! Technical progress does not present itself in the form of a free gift […]. Rather it presents itself in the form of an impetuous flow which cannot be stopped, but which must always be channelled into new directions, which must themselves, bit by bit, be discovered and invented, as the old ones continuously become saturated. And if the new directions do not emerge, or are not found, or are not found soon enough, that impetuous flow […] may well […] cause considerable damage (in the form of unemployment) to existing positions.
He was writing about the torrent in general. It has since become rather more impetuous.