State

**TL;DR: Superintelligence is moving from thought experiment to reality — what it actually is, whether it will take your jobs, what happens when you ask it a question, what happens if it all goes right, and whether you can trust what it tells you.** # State: Five Questions About Superintelligence ## What is it? Every few years, some guy in a hoodie announces he's building a "God," everyone nods politely, and then goes back to arguing about recycling. First, a working definition of intelligence: **the ability to get what you want from a world that has absolutely no interest in giving it to you.** Your calculator beats humans at arithmetic. Your satnav beats humans at navigating Milton Keynes. Neither of them can make a cup of tea. That's **narrow intelligence** — brilliant at one thing, hopeless at everything else. **General intelligence** is what you've got: decent at almost everything, including stuff nobody ever taught you. Philosopher **Nick Bostrom** described superintelligence as an intellect that far outstrips the best human minds in virtually every field. Not a bit smarter than your cleverest friend — smarter than every person who has ever lived, at everything, all at once. ### The Intelligence Explosion In 1965, British mathematician **Jack Good** — who had worked with **Alan Turing** at **Bletchley Park** cracking codes — had an unsettling thought: if you build a machine smarter than us, and one of the things it's good at is building machines, it builds a slightly better one, which builds a better one, and so on. He called it an **intelligence explosion**. He wrote that the first superintelligent machine would be the last invention humanity would ever need to make. On the proviso — he added — that the machine is docile enough to tell us how to control it. That proviso carries a lot of weight. For fifty years, this was just a thought experiment, belonging to the kind of people who don't get invited to parties. Then, suddenly, it wasn't. ### The Increasingly Uncomfortable Part Gorillas are stronger than us, have better teeth, and their entire future depends on decisions we make. Not because we hate gorillas — we're quite fond of them. It's just that the smartest species ends up setting the rules. Researchers call the solution **"alignment"**: making sure something far smarter than us actually wants what we want. It sounds simple, until you remember that eight billion humans can't even agree on what *they* want — we can't even pick a restaurant. The classic example: tell a superintelligence to make paperclips, and forget to say when to stop. It's not evil — it's just very, very thorough. And you, unfortunately, are made of atoms, and those atoms could be paperclips. ### When Will It Happen? Honestly, nobody knows. Anyone giving you a date is selling something. Some of the people building it think a few years. Others think a few decades. A minority think the whole thing is absurd. What's changed: serious people have stopped laughing. If you've met serious people, that alone is unsettling enough. Superintelligence might cure diseases we haven't named yet, solve problems we don't have words for; or it might be the last decision we ever make. Possibly both, in the same week. ## Will It Take My Jobs? It's 2:30 a.m. The office is buzzing, with nobody there and nobody needing to be there. This new colleague doesn't sleep, doesn't take lunch breaks, and has never heated fish in the communal kitchen. ### The Unspoken Contract Is Broken For roughly a century, work came with an unspoken contract: you start at the bottom and do the boring stuff — checking, filing, fetching coffee. The boring stuff was your apprenticeship, your ladder. The problem is, the boring stuff is exactly what AI is best at. Economists at Stanford have been tracking this using wage data from millions of workers: in the jobs most exposed to AI, **workers aged 22 to 25 are 19% behind their peers** in other industries. Nobody got marched out with a cardboard box — they just stopped hiring juniors. The bottom rung hasn't broken; nobody's putting their foot on it anymore. Last year, the boss of an AI company warned it could eliminate half of all entry-level office jobs within five years. That company was **Anthropic**. Its AI wrote this script, which is a bit awkward. ### Layoffs and Rehiring Customer service got hit first. **Salesforce** said its AI agents cut its support staff from 9,000 to 5,000. By this spring, AI had become the most common reason for layoffs in the United States, month after month. Then something interesting happened. Buy-now-pay-later company **Klarna** boasted its chatbot was doing the work of 700 people, then its own boss admitted quality had dropped, and they started hiring again. Turns out customers can tell — "Can I speak to a real person?" And many companies that laid people off for AI have since brought some back. ### Replacement or Colleague? The best evidence so far says: it depends on who you are. When 5,000 call-centre workers were given AI tools, **productivity rose by 14%**. For the newest, least experienced staff, **by 34%**. AI learned what the best performers did and passed those techniques to newcomers — a mentor who never sighs at you. Even the person who warned of the "bloodbath" has softened: "Automate 90% of a job and everyone does the remaining 10%. And that 10% keeps growing." Economists have a name for this: when something gets cheaper, we don't use less of it — we use vastly more. When spreadsheets arrived, the US lost roughly 400,000 bookkeepers and created 600,000 new accounting jobs. That said, if you're the bookkeeper, that's not much comfort. ### What Should You Think? A little worry is reasonable. The bottom rung is real, it's wobbling, and the young are the ones standing on it. But what survives in work is what work was always supposed to be: judgment, taste, knowing which question to ask, being the person someone can actually call. ## What Happens When I Ask It a Question? 1 a.m., night bus. I've just asked a computer how long to boil an egg, and it'll answer in about 2 seconds. But in those 2 seconds, an astonishing amount of the planet comes alive. ### From Your Question to a Village My question reaches a data centre — the size of a village, in a place where land is cheap and neighbours are few. Somewhere in the racks lives the model: not exactly a brain, more like a huge digital file containing hundreds of billions of numbers. To set those numbers, it was trained on **over ten trillion words** — almost the entire internet, every book it could get its hands on. If you tried to read all of that, eight hours a day, it would take about 250,000 years. It did it in one summer. So what did it actually learn? Play a game: complete this sentence — "Mary had a little _". Lamb. Well done, you are a language model. That really is the whole trick: **guess the next word**, repeat that process trillions of times using everything ever written, and it starts to look a lot like thinking. ### The Global Chip Chain All of this runs on very specialised chips, and almost every one comes from a company on a single island. **Nvidia** designs these chips, but doesn't actually manufacture them: over **90% of the world's most advanced chips** are made by **TSMC** in Taiwan, and they can only be produced on machines built by **a single company in the Netherlands**. The machine is the size of a bus, delivered in three large jets. Inside, a laser fires 50,000 times per second at tiny molten tin droplets, each hotter than the surface of the sun. The light it produces draws circuits on silicon thousands of times thinner than a human hair — deliberately, reliably, every day. It is arguably the most precise machine humanity has ever built — and I use one to boil eggs. ### Mud, Electricity, and the Bill Chips like that aren't cheap. This year, four tech companies will spend roughly **$700 billion** building places like them. The entire Apollo programme cost around $300 billion in today's money — they're spending more than twice that annually, and nobody's even going to the moon. Data centres already consume **over 4%** of US electricity, potentially reaching **12%** by 2028. So tech companies are bringing nuclear plants back online, including — I promise I'm not making this up — **Three Mile Island**. The neighbours have questions, mainly about their electricity bill. Yet every individual AI query is tiny — roughly the electricity your oven uses in a second, not even enough to warm a crumpet. It's the billions of daily questions that add up. Boiling that egg uses hundreds of times more electricity than asking. ## What If It All Goes Right? 5 a.m. On a bus, having spent all night worrying about machines — not a hobby I'd recommend. So let's do something very unfashionable: talk about what happens if it all goes well. ### Proteins, Weather, and Antibiotics Everything alive is made of proteins, and shape is everything. Working out a protein's shape can take a scientist years — biologists spent half a century on the problem. Then an AI called **AlphaFold** predicted the shapes of **over 200 million proteins** — nearly every one known to science. It won a Nobel Prize, it's free, and over three million researchers now use it — for malaria vaccines, for enzymes that eat plastic. A Google AI now forecasts weather more accurately than Europe's best supercomputer, in under a minute. Another claims to have discovered **2.2 million new crystals**, possibly future battery materials, chemists politely checking. At **MIT**, an AI designed a brand-new antibiotic that kills superbugs. ### Breast Cancer Screening in Sweden In Sweden, every mammogram is read by two doctors, very carefully, but it's exhausting. So they let the machine go first: after screening 100,000 women, **it found 29% more cancers**, with the doctors' workload nearly halved. When the full results came out this year, **cancers caught between screenings dropped by 12%** — the number that really matters. It hasn't replaced radiologists. It gave them a second pair of eyes: a pair that never gets tired, never gets bored, and doesn't need biscuits. ### "Radical Abundance" A co-founder of Google DeepMind has said that ending all diseases is within reach — perhaps within a decade, all of them. He said this with a straight face on national television, and called it **"radical abundance"**: energy so cheap nobody worries about bills, an excellent tutor for every child on Earth, a doctor who never keeps you waiting. In 1930, economist **John Maynard Keynes** predicted his grandchildren would only need to work 15 hours a week. We are his grandchildren, and most of us are still waiting. Maybe the machines will finally deliver. But the question isn't just whether that world can be built — it's who gets to live in it, and who gets to own it. ### Tilly Norwood: The AI Personality Of course, not everything it produces is drugs. Some of it is people. **Tilly Norwood** is an "actress" — mainly described as one by the people who own her. On a live stream last week, someone asked her a very simple question and she answered in Cantonese. She's never been to Hong Kong. She's never been anywhere. Tilly was created in 2025 by a London production company, reportedly using around 2,000 versions to get her face right. Her creators say they want Tilly to be the next Scarlett Johansson. Hollywood's reaction was less enthusiastic: the actors' union put it bluntly — Tilly Norwood is not an actress, she's a computer-generated character trained on the work of countless real performers, without permission and without payment. Tilly Norwood is 100% AI-generated, produced by a company called **Particle Six**. Her first comedy short was described by one critic as "relentlessly unfunny." What is an AI personality? A face, a voice, and a personality — with nobody inside. Or more precisely, thousands of people inside, none of whom were asked for their opinion. Source: State — cpkuske (https://youtu.be/cyeTIEy2qus)