In early September, Anthropic’s economists put out scenarios for how AI could change the US economy by 2030. These are no forecasts, we’re told, and no probabilities there. But the most radical one looks like this: GDP up by a third, work down by lots of percent. Only for those sitting at PCs so far.
\nPlumbers and electricians win: they’re scarce, they’re paid a third more. But the model “doesn’t allow rapid progress in robotics.” Slow progress goes unmentioned, and if AI is good at intellectual work, it can build robots too. I think plumbers will follow in 5–15 years. And even without robots, those white-collars we threw out in the cold in the previous paragraph have to go somewhere.
\nStill, these fears aren’t new. Let’s see whether this is just about jobs, and how much bullshit is in here.
\nIn October 2015, Sam Altman promised that self-driving cars would fill the streets in 2–4 years. Elon Musk agreed — two or three. Ray Kurzweil calculated 2019 — in The Age of Spiritual Machines, back in 1999. Taxi drivers, against their usual, were much more optimistic: that’s all talk, and AI can’t change a tire.
\nWell, the taxi drivers were right, and the three Cassandras above and I with them weren’t. It’s 2026, and robot cars are still confined to scattered pockets. By and large, they haven’t come that far since 2015: they drive only where they’re told to and only in good weather. Still don’t change tires either.
\nBut robots are hard. Remember “Humans Need Not Apply”? It also goes over those much-hyped robot cars and scares us with Baxter, a general-purpose robot, slow and clueless but cheap, meant to undercut low-skilled labor. In our reality, Baxter’s sales limped along, and they shut the project down in 2018, along with the company — the robot kept missing the machine with the part.
\nLet’s leave robots alone. Anthropic was threatening the office plankton anyway. A little look back: in 2016, Geoffrey Hinton suggested we stop training radiologists: in five years, AI would read scans better than humans. Ten years later, the US has more radiologists, and they’re paid more too.
\nTurns out reading scans is only part of a radiologist’s job. Once that got faster, we did more scans and ran short of radiologists for the other parts. Markets growing as prices fall, and here as things speed up too, is a paradox known since the 19th century. In 1865, Jevons noticed that although steam engines burned less coal, England as a whole burned more. Similar stories happened to:
\nEach time, Jevons paradox and market elasticity come to our rescue. Besides, Anthropic itself talks about automating tasks, not professions. So? Relax?
\nWaymo clocks up 500,000 paid rides a week and has 20–30% of the market in its San Francisco service area. Apollo Go (Baidu) does 300,000+ a week across 20+ Chinese and Arab cities. Both are below one percent nationwide, but two years ago Waymo did ten times fewer rides. Bringing the national share up is probably just a matter of time and scale.
\nBaxter failed, but cobots kept developing. Now humanoid robots are making a comeback in both China and the US, from factories to the sex industry.
\nBank tellers didn’t lose out to ATMs, but their numbers started falling once mobile banking arrived.
\nAnd the main reason this time might be different: before, we had narrow AI or plain automation. LLMs lay claim to generality — AGI — and even if that generality turns out jagged, the results could be different.
\nA common answer: people will do creative things and just bum around on basic income. Where it’ll come from is less clear — historically, benefits came from above out of necessity, not for our pretty faces.
\nBismarck introduced the world’s first health insurance and pensions in 1883–1889, not out of the goodness of his heart: he needed to outbid the socialists for the workers and tie them to the state. In 1914, Ford doubled wages and cut an hour off the working day only because the assembly line was so dull that turnover had shot up to 370% a year.
\nWhat about democracy? Won’t people just vote for the socialists?
\nClassical and the most well known democracy arose in the city-states of Ancient Greece. Everyone has heard of it, few know the details.
\nAthens got its most universal democracy in 462 BC. The poor finally had enough weight in the assembly to decide stuff. They had been sitting there before, but twenty years earlier Athens had built two hundred triremes. The fleet came to determine its military power, partly economic too, and the rowers gained more weight.
\nBefore that, it was the hoplites — heavily armored soldiers — whose voices counted. If you were a woman, disabled, or just couldn’t afford the expensive armor, you were out of luck. Such a property-qualified democracy. Before the hoplites came the cavalry, infantry didn’t count for much. Keeping a horse costs even more than armor. You need plenty of land, usually ancestral.
\nSo power spread to more of the population. Looks like social progress. Except the fleet was crushed in Sicily in 413 BC and finished off a little later, and power went back to the hoplites. Plenty more twists followed, but take that up with Thucydides and Xenophon.
\nSo much for progress and fighting for rights — those who are needed get a say. Aristotle writes quite plainly that after the kings, power in the city-states belonged to those who fought: first the cavalry, then, as heavy infantry grew stronger, “more persons came to have a part in the government.”
\nStill, as Ford and Bismarck already showed us, the military aspect isn’t the only one.
\nIn the mid-14th century, the plague wiped out a large part of Europe. There was more land than people, and peasants could bargain: wages rose, serfdom in the West weakened. A peasant simply left for the lord’s neighbor, who’d hide him because he needed the hands.
\nUniversal male suffrage came with industrialization, which needed both mass armies and masses of workers. In many countries this happened in 1917–1919, in the wake of World War I, the first truly industrial war. You can also connect the abolition of serfdom in Russia to defeat in the Crimean War, and find more parallels.
\nMachines made the most visible and least disputed difference between men and women — physical strength — a lot cheaper. And if women’s economic importance as workers evens out, their rights, after a bit of upheaval, catch up too.
\nBut it’s useful to look at the reverse too. That’s the fear, after all: AI will replace people.
\nIn rentier states, the money comes from oil, often extracted by foreign companies. For military muscle, there are mercenaries or some ally’s wing. People aren’t needed. How does that turn out?
\nDepends. In Angola, the rich simply live behind fences with guards. Until 2016, Saudi Arabia set regular gas at about $0.45 a gallon and kept its citizens on benefits and government jobs, with no VAT or income tax.
\nThose with money and power have a choice: bring society up to a comfortable level or shut themselves away behind three fences guarded by gunmen. That’s the third reason to keep people — humanitarian. Man needs man: to talk to, to out-flex, or just so it isn’t scary out there.
\nThen there are personal services. A robot can be a nanny, a masseur or a waiter, but many will still prefer a human. These services are only a mass market while there’s a middle class though, i.e. this is a multiplier, never the base of the pyramid.
\nThat’s what it comes down to. The answer will define the new balance of power and determine whether society degenerates or finds a new equilibrium. Let’s use the three aspects above: military, labor and humanitarian.
\nFirst, let’s throw out the useless definition of AGI as AI that does everything better than humans. If it’s literally everything, people aren’t needed by definition. Besides, AI is still jagged: it solves hard math problems but can’t turn a mug over on the way to the car wash. Let’s start there, maybe plus inevitable advances in cost, wider use and spread into areas barely touched so far.
\nMilitary. On one hand, in the most modern war — in Ukraine — drones already account for 70–90% of casualties. On the other, it’s people who still fly them, and you still need infantry. People win through resourcefulness; automation and AI through speed. Where speed matters more, as in air defense, the human just switches it on and keeps an eye.
\nLabor. OpenAI models are solving Millennium Prize Problems; robot cars still won’t drive in the rain. AI can generate a game that works, but it still can’t make it not boring. New ideas and taste remain with people.
\nEven in math, intuition can tell you where to go next. A big proof is a long chain of formally checkable pieces, but how promising each piece is often comes down to intuition. Without that taste, the chance of completing the chain falls with its length. At least for Navier–Stokes, human hints were needed.
\nIt’s very possible OpenAI will roll out another couple of proofs and show that math is no longer for people. Still, the example shows that human intuition can be of value in unexpected places.
\nHumanitarian. If we take a closer look at the oil states and sort them by barrels per person then the Gulf monarchies, Norway and Brunei come out on top, with Gabon and Venezuela further down. People turn out more generous and big hearted when there’s enough to share.
\n| Country | Year | Barrels* | Homicides | Fences |
|---|---|---|---|---|
| Qatar | 2012 | 2159 | 0.4 | ✗ |
| Kuwait | 2008 | 1002 | 1.4 | ✗ |
| UAE | 2022 | 540 | 0.7 | ✗ |
| Saudi Arabia | 2011 | 244 | ~1.0 | ✗ |
| Brunei | 2012 | 188 | 1.0 | ✗ |
| Norway | 2022 | 145 | 0.5 | ✗ |
| Oman | 2008 | 142 | 1.8 | ✗ |
| Bahrain | 2011 | 138 | 1.0 | ✗ |
| Libya | 2008 | 112 | ~2.9 | ✗ |
| Gabon | 2008 | 65 | ~13.8 | ✓ |
| Venezuela | 2008 | 45 | 52.0 | ✓ |
| Trinidad and Tobago | 2008 | 40 | 39.9 | ✓ |
| Kazakhstan | 2011 | 38 | 8.3 | ✗ |
| Angola | 2008 | 31 | ~19.0 | ✓ |
| Russia | 2012 | 29 | 11.3 | ✗ |
So if AI and robots really do produce lots of useful things, instead of or alongside people, perhaps we can count on a Saudi-style basic income after all.
\nAnother dimension: how concentrated ownership of AI and robots will be. Even a few tens of thousands of owners could be very different from the kingdoms of a handful of Samodei. But let’s not dig into that now.
\nAnyway, if any of the factors above play out — if some part of aesthetics, originality, resourcefulness or cunning stays with people — even better. Back in 2013, I wrote that people would always have something to offer each other. Looks like we can hope for now.
\nP.S. Maybe the guys will even slow down and give us more time to prepare. And to make a few more gloomy predictions.
","headings":[{"level":2,"text":"Robots and Cassandras","id":"robots-and-cassandras"},{"level":3,"text":"…","id":"section"},{"level":3,"text":"Or not…","id":"or-not"},{"level":2,"text":"Basic Income and the Balance of Power","id":"basic-income-and-the-balance-of-power"},{"level":3,"text":"Hoplites and Rowers","id":"hoplites-and-rowers"},{"level":3,"text":"The Economy Side","id":"the-economy-side"},{"level":3,"text":"Where People Aren’t Needed","id":"where-people-aren-t-needed"},{"level":2,"text":"So Will We Be Needed?","id":"so-will-we-be-needed"},{"level":2,"text":"Whose Robot? And What’s Next?","id":"whose-robot-and-what-s-next"}]}}