Florence's AI · Lesson 4
why now

Why everything
suddenly exploded.

A single grain of rice, doubled on each square of a chessboard, ends up burying the world — and something almost exactly like that happened to the chip inside your phone.
For Florence,
the lesson where the speed makes sense.
Florence's AI · Lesson 4
A quiet noticing

A man drew a line, and it kept coming true.

In 1965 an engineer named Gordon Moore sat down to write a short magazine article about the tiny electronic chips his industry was learning to make. He had only a handful of years of data — chips were brand new — so what he did next was bold. He looked at how many of the tiny switches called transistors engineers had managed to pack onto a single chip each year, noticed the number was roughly doubling about every year, and predicted it would carry on.

That prediction — later settled to a doubling roughly every two years — became known as Moore's Law. And here is the strange thing: it broadly held for about fifty years. Not because nature forces it to — it is not a law of physics like gravity — but because it became a target the whole industry quietly agreed to chase, year after year.

The result is hard to take in. The transistors got smaller, and smaller, and smaller. The first chips held a few thousand. A chip in a phone today holds billions — and each switch is now far tinier than a single human cell, smaller even than most viruses. You are carrying, in your pocket, something a 1965 engineer could only sketch as a dream on the back of his article.

Honest about the word "law"

It is worth being straight about this, because grown-ups often aren't. Moore's Law is not a law at all — it is an observation, a trend people chose to keep up. And in the last few years it has clearly slowed: transistors are now so small that the makers are bumping into the size of atoms themselves, and you cannot build a switch out of half an atom. The doubling hasn't stopped dead, but the effortless years are over. A trend is not a promise.

A detail you might keep
35–50 seconds · MF 1
Cool fact

If cars had improved at the pace chips did under Moore's Law, a family car would now cost a few pence, do millions of miles to the gallon, and go faster than sound. The comparison gets wheeled out so often that engineers half-joke about it — but it does capture something true: no other technology in history has ever sped up like this.

Florence's AI · Lesson 4
The big idea

Doubling is sneaky. It hides, then it pounces.

The word for repeated doubling is exponential growth — and the single most important thing to know about it is that your gut feeling about it is almost always going to be off. Human intuition is built for things that grow by adding a steady amount — a plant a few centimetres a week, a savings jar a pound a day. We are not, by nature, wired for things that grow by multiplying.

Look at the difference. Start at 1, and add 2 each step: 1, 3, 5, 7, 9, 11… After ten steps you reach 21. Calm, predictable, a straight gentle slope. Now start at 1 and double each step instead: 1, 2, 4, 8, 16, 32… After those same ten steps you are past a thousand. After twenty, past a million. The two start side by side, almost touching — and then one of them quietly leaves the planet.

Tap each card. One side grows by adding; the other by doubling. Watch where they end up.

Adding 2 each step
1 → 3 → 5 → 7 → 9 → … what after 30 steps? 61. A long, gentle ramp. After 30 steps you've added 60. Tidy. Predictable. This is linear growth.
Doubling each step
1 → 2 → 4 → 8 → 16 → … what after 30 steps? Over a billion. The same 30 steps — but doubling, not adding. This is exponential growth, and it is why your phone exists.
The thinking-move: learn to spot doubling

This is a habit worth carrying out of this lesson and into the rest of your life. When you hear that something is doubling — users, cases, prices, computer power — your instinct will quietly whisper "that's fine, it's small." Don't trust that whisper. Ask instead: doubling how often, and for how long? A thing that doubles is not adding up. It is winding up. Naming it as exponential is half the battle.

Cool fact

Fold a sheet of paper in half, then in half again, and again — each fold doubles the thickness. If you somehow managed 42 folds, the stack would be thick enough to reach the Moon. You can't actually fold paper more than about seven or eight times, which is rather the point: doubling outruns the real world astonishingly fast.

Florence's AI · Lesson 4
A very old warning

One grain of rice, and a chessboard.

There is a story told for hundreds of years, in many lands, that teaches doubling better than any sum. An inventor shows a ruler a wonderful new game — chess — and the ruler, delighted, offers any reward he likes. The inventor asks for something that sounds almost insultingly modest: one grain of rice on the first square of the board, two on the second, four on the third — only doubling, square by square, to the sixty-fourth.

The ruler laughs and agrees, thinking he has got off lightly. He has not. Reveal the squares one at a time, and try to feel the moment it runs away from you:

Squares 1 to 10. 1 grain, then 2, 4, 8… by the tenth square it is 512 grains. A small bowlful. The ruler is still relaxed — this is the part where your gut feels safe.
Square 20. A little over a million grains on that one square — around a large sack of rice. We've only moved ten more squares, and we've gone from a bowl to a sack. The quiet part is ending.
Square 32 — halfway. About 4 billion grains on this square alone. Halfway across the board, and we're already past what a whole country might grow in a year. And remember: every remaining square is bigger than everything so far combined.
Square 64 — the end. The final square holds about nine billion billion grains. The whole board together comes to more rice than has been grown in all of human history — a heap that would bury a city. The ruler, of course, could never pay. That is doubling. It hid in a bowl, then it swallowed the world.
One grain on square 1, doubling each square square 1 square 2 square 3 square 4 1 grain 2 4 8 …keep doubling, all the way to… Square 64 more grains than all the rice ever grown on Earth
The chessboard reward: one grain, doubling each square. The first four squares look harmless. Square 64 holds about nine billion billion grains — more than humanity has ever grown. Original diagram, drawn for this lesson
Cool fact

There's a neat shortcut hidden in the board. The grains on any one square are one more than every square before it added together — so the last square alone holds slightly more than the other sixty-three combined. The grand total works out to a number mathematicians write as 2 to the power of 64, minus 1: 18,446,744,073,709,551,615. Worth seeing written out at least once in your life.

Florence's AI · Lesson 4
The same shape, in silicon

Now put chips on the chessboard.

Here is the move that makes the whole lesson click. Moore's Law is the chessboard. Each "square" is roughly two years, and the thing being doubled is the number of transistors on a chip. So the count of switches climbed exactly the way the rice did: a few thousand in the early squares, then millions, then — by the squares we live in now — billions.

This curve shows it, roughly. Notice the steepness: nearly flat for years, then rising faster and faster, almost standing on end. That upward sweep is the signature of doubling. Read the numbers up the side carefully, though — each line is ten times the one below it, so the curve is far steeper than it even looks.

Transistors on one chip — roughly 1 thousand 10 thousand 100 thousand 1 million 10 million 100 million 1 billion 1970s 1980s 1990s 2000s 2020s Each line up the side is ten times the one below it.
Transistors per chip, roughly, over the decades — a near-straight climb on a ten-times scale, which is what doubling looks like. The exact numbers vary by chip; the shape is the point. Original diagram, drawn for this lesson

One honest note on the picture. Drawn on an ordinary scale, this line would shoot straight off the top of the page within a few decades — so, as scientists often do, the side has been squashed so that each step up means ten times as much. On a squashed scale like that, steady doubling shows up as a straight climbing line. That's the tidy trick: a near-straight line here is the fingerprint of something exploding.

Cool fact

The transistors on a recent chip are spaced only a few nanometres apart — a nanometre being a millionth of a millimetre. To picture it: if one transistor were blown up to the size of a full stop on this page, that same magnification would make you taller than the entire planet. We are now building switches only a few dozen atoms wide.

Florence's AI · Lesson 4
Why this lesson is about AI

Two things had to pile up first.

You might be wondering what rice and transistors have to do with the AI Florence can talk to today. Everything, as it turns out. The ideas behind modern AI are surprisingly old — some are from the 1950s, as you'll see next lesson. For decades they mostly didn't work, and it wasn't because the ideas were poor. It was because two ingredients hadn't piled up high enough yet.

Cheap compute

Raw thinking-power — the ability to do staggering numbers of tiny sums very fast. This is the gift of Moore's Law: the doubling chessboard made compute so plentiful and so cheap that calculations once unthinkable became ordinary.

Huge data

Vast piles of examples to learn from — text, photos, sound. The internet quietly produced these by the billion, as the whole world wrote, posted and uploaded. Suddenly there were oceans of examples to study.

Hold those two side by side, because together they are the answer to the question this lesson is named for. Modern AI didn't arrive because someone had a single brilliant new thought. It arrived because the compute finally got cheap enough — thanks to fifty years of doubling — and the data finally got plentiful enough, thanks to the internet. The old ideas had been waiting, patiently, for the world to grow powerful enough to run them.

That is the deep reason behind the word exploded. For ages, almost nothing seemed to happen — the quiet early squares of the board. Then both ingredients crossed some line at once, and the machines could suddenly do things that looked, for the first time, a little like learning. You're about to meet exactly how, in Lessons 5 and 6.

A detail you might keep
35–50 seconds · MF 1
Cool fact

Training one large modern AI can involve reading a slice of text so big that a person reading day and night, without ever stopping, would need thousands of years to get through it. The machine works through it in weeks — not because it is wiser than you, but because cheap compute lets it read in parallel, at a scale no human life could ever reach.

Florence's AI · Lesson 4
Pulling the threads together

The transistor you met is the thing doing the doubling.

Cast your mind back over the story so far. In the hardware lesson you met the transistor — the tiny switch that replaced the bulky glass valve, the thing that let computers shrink from room-sized beasts into something pocket-sized. That switch is the very thing this lesson has been counting. Moore's Law is, at heart, the story of how many transistors we learned to crowd onto one chip — and the answer, square by square, was: keep doubling.

And this is real maths, not a trick of the trade. Doubling is powers — 2, 4, 8, 16 is 2 to the power of 1, 2, 3, 4. The difference between adding and multiplying your way along is the difference between a linear graph (a straight slope) and an exponential one (that climbing curve). The intuition you've built today — that exponential things feel slow then pounce — is one of the most useful pieces of mathematics a person can carry. You'll meet it again in graphs, in percentages, in compound interest.

So keep the chessboard in your pocket as we go on. Because the next question is the obvious one: if machines suddenly had all this cheap power and all this data — what did people do with it? The answer is that they returned to a very old dream, one that had failed and been laughed at for decades: building a machine that could learn for itself. With the board finally full of power, that dream got a second life. That's where we're heading.

Three threads, one knot

Notice how the subjects keep talking to each other. The transistor from your hardware lesson is the unit being doubled. The powers and graphs from maths are the shape of the doubling. And the internet you studied in Computing is where the data came from. Three separate things you've learned, tied into a single knot — and that knot is the reason modern AI exists at all.

Cool fact

Gordon Moore co-founded the company Intel, whose chips ran a great many of the world's computers for decades. So the man who merely predicted the doubling also spent his life helping make it come true — which is part of why a loose guess in a 1965 magazine ended up shaping the entire modern world.

Florence's AI · Lesson 4
Question 1 · circle the answer

What Moore noticed.

In 1965, what did Gordon Moore notice was happening to the number of transistors on a chip?
Question 2 · circle the answer

Law, or not?

The lesson was careful about the word "law". What is Moore's Law really?
Question 3 · circle the answer

Adding versus doubling.

What is the key difference between linear growth and exponential growth?
Question 4 · type your answer

On the chessboard.

You start with 1 grain on square one and double each square: 1, 2, 4, 8… How many grains are on the sixth square?
that square holds
Question 5 · circle the answer

Why now.

The lesson said modern AI finally became possible because two things piled up. Which two?
Question 6 · circle the answer

Where the data came from.

Where did the huge piles of examples — the data that modern AI learns from — mostly come from?
Question 7 · type your answer

The shape of an explosion.

What single word describes growth that multiplies — like repeated doubling — rather than adding the same amount each step?
it is called
Question 8 · circle the answer

The thinking-move.

What was the thinking-move this lesson asked you to carry away about exponential change?
Florence's AI · Lesson 4
Question 9 · have your say

Where else is something doubling?

You now have a kind of radar for doubling. So use it. Think of one thing in the real world — anything at all: something in nature, money, technology, a worry, a hope — that grows by multiplying rather than adding, or that you suspect might. Describe it in your own words, and say what your new radar tells you: does picturing it as exponential change how it looks? There's no single right answer here — pick something that genuinely interests you, and reason it through.

Start by naming your example in one line. Then explain why it multiplies rather than adds — and what that means for where it ends up.

0 words
reading your thinking…

On your thinking, Florence

strong The example you reached for is a real one — it genuinely multiplies rather than adds, which is exactly the radar this lesson was trying to switch on. You didn't only name it; you said why it grows the way it does, and that "why" is the part most people skip.

try this One way to push the thinking further: put a rough number on it. Doubling how often? Every day, every year? The same idea feels completely different depending on the speed of the doubling, and pinning that down is where the real insight lives.

to add A question worth sitting with: most things can't double forever — the world runs out of room, like the paper you can't fold past eight times. What is it, do you think, that finally stops your example from doubling? That's a deep question, and a good one.

Florence's AI · Lesson 4
A life

Gordon Moore.

He was born in California in 1929, the son of a county sheriff, and grew up quiet and curious — a boy who loved chemistry sets and blowing things up gently in the back garden. He trained as a chemist and a physicist, and might have stayed a careful laboratory scientist his whole life, had he not landed, in the late 1950s, in the brand-new business of making electronics out of a material called silicon.

In 1965 a magazine asked him to predict the future of these new chips. He had barely any data to go on, but he drew his line anyway, and guessed the doubling would carry on for a decade. That short article — almost a throwaway — became one of the most famous predictions in the history of technology. Three years later, in 1968, he co-founded a company called Intel, and spent the rest of his working life helping the prediction come true.

He was, by all accounts, a calm and modest man who disliked fuss and gave away most of his fortune to science and conservation before he died in 2023, aged ninety-four. It is a quietly lovely thing: the person whose name is attached to the fastest, loudest change in human history was himself unhurried, careful, and kind. A reminder that you don't have to be loud to change the world.

Cool fact

Moore very nearly didn't get his prediction published as a tidy "doubling." An editor wanted a snappier story, and the famous graph was almost cut for space. The single most quoted line in the computer age survived partly by luck — which is worth remembering the next time someone tells you the future was obvious all along.

To watch at home

If you'd like to see exponential growth move.

We looked hard for a short film to drop in here, but couldn't find one from a source we fully trust that was both on-topic and the right tone — so rather than settle, here are two genuinely good ones to watch with Dad, on a screen bigger than this.

BBC · documentary · suitable for 13
The Secret Rules of Modern Living: Algorithms
Mathematician Marcus du Sautoy, warm and clear, on the hidden maths running the modern world — including how quickly small things compound. A calm, friendly hour; nothing in it to worry a younger viewer.
The Royal Institution · talk · suitable for 13
Any RI lecture on exponentials & growth
The Royal Institution's talks and Christmas Lectures often demonstrate doubling and exponential growth with real objects on a stage — chessboards, folding paper, chain reactions. Search their channel together for "exponential" and pick one that looks fun.
Florence's AI · Lesson 4
Glossary

The words from today.

Moore's Law
Gordon Moore's 1965 observation that the number of transistors on a chip roughly doubles every couple of years. A trend, not a law of nature — and it has slowed recently.
Transistor
A microscopic electronic switch — the building block of every chip. A modern chip holds billions of them, each smaller than a human cell.
Exponential growth
Growth that multiplies by the same factor each step (such as doubling). It looks slow at first, then rises almost vertically.
Linear growth
Growth that adds the same amount each step, drawing a straight slope on a graph — the gentle, predictable kind.
Compute
Raw computing power — the ability to do enormous numbers of small calculations quickly. Made cheap and plentiful by Moore's Law.
Data
Examples a machine can learn from — text, images, sound. The internet produced these in vast amounts, the second ingredient modern AI needed.
End of lesson four

You've found out why everything sped up.

A man named Gordon Moore noticed, in 1965, that chips were doubling — and that doubling held for half a century. You felt how exponential growth hides in a bowl of rice and then buries a city, and you learned the move that matters most: human intuition is poor at doubling, so name it when you see it. And you found the real answer to "why now" — cheap compute from the chessboard, and huge data from the internet, piling up together until the machines could finally do something new. Next time, Florence, with all that power suddenly to hand, people went back to an old, once-laughed-at dream: a machine that could learn for itself. That's where we pick up.

F.M. · AI · Lesson 4
Images · The cover motif (growing dots), the chessboard-and-rice diagram, and the transistors-per-chip curve are all original line-art drawn for this lesson. No external images are used.
Video — none embedded. No film from an approved institutional source could be verified as both on-topic and tonally suitable; per the project's video rule, the embed was omitted and a "watch at home" recommendation given instead.
Accuracy note — Moore's 1965 figure was revised to roughly every two years around 1975; Moore's Law is an observation/industry trend, not a physical law, and has slowed markedly in recent years. Specific transistor counts vary by chip; the lesson states orders of magnitude ("roughly", "billions") rather than exact figures by design.