Start with one brain
Your brain is amazing. It has about 86 billion cells, all chatting to each other, and it runs on roughly the power of a dim light bulb. No computer today does what your brain does with so little energy.
But your brain also has limits, and they are worth knowing about, because a superintelligence might not share them.
Limit one: speed
A signal in your brain travels at about 100 metres per second. That sounds fast. A signal in a computer chip travels at close to the speed of light — about 300 million metres per second. That is millions of times faster.
Now, a computer isn’t automatically smarter just because it is faster, any more than a cheetah is smarter than you because it runs faster. But imagine a mind that thought the way you do, just thousands of times quicker. While you were deciding what to have for lunch, it could have read every book in the library.
That is the first big idea: speed. In the zoom-out scene, it is the jump from one brain to a library — a mind that could hold everything ever written in view at once.
Limit two: you can’t copy yourself
If you learn something brilliant today, there is still only one of you. You can teach others, but that takes years, and something is always lost along the way.
Software is different. When a program learns something, you can copy it. A million times, if you like. Every copy knows everything the original knew, instantly.
Imagine a million copies of the best doctor in the world, each working on a different disease. Or a million copies of the best teacher, each helping one child. That is the second big idea: scale. In the scene it is the city of glowing thinkers, and then the whole planet.
Speed plus scale
Put the two together and you can see why people take superintelligence seriously, even though nobody has built one.
A single AGI — a machine roughly as capable as a person — might not sound world-changing. But a million copies of it, each thinking a thousand times faster than a human, adds up to more thinking in a day than humanity manages in centuries. If that thinking is any good, it would change things very quickly.
The intelligence explosion
Now for the idea in this lesson’s title. It comes from a mathematician called I. J. Good, who described it in 1965, long before modern AI.
Here is the loop:
- Imagine an AI that is smart enough to improve its own design — even a little.
- It builds a slightly smarter version of itself.
- That version is better at improving designs, so it builds an even smarter one, faster.
- Repeat.
Each turn of the loop makes the next turn quicker. Watch the robots in the scene below: the hand-over gets faster every time. If this kept going, intelligence might shoot upwards very fast — an explosion rather than a gradual climb. Scientists call the loop recursive self-improvement.
Would it really happen like that?
This is where experts disagree, and it is important to hear both sides.
“Yes, it could go fast.” Some researchers point out that AI already helps design better computer chips and write better AI code. Once machines are doing most of the improving, they argue, the speed limit is gone, and the loop could run away within months or even weeks.
“No, it would be gradual.” Other researchers say the real world gets in the way. Smarter designs still need new chips to be built in factories, which takes years. Experiments need real time to run. Some problems get harder as you go, not easier. They expect a steady climb, with plenty of time to notice and adjust.
“Both.” Many experts think the truth is in between: fast in some areas, slow in others, with surprises along the way.
Nobody knows which picture is right. That is exactly why people study it now, while the loop is still an idea.
What would still be hard for a superintelligence?
Being very smart is not the same as being all-powerful. Even a mind at the top of the staircase would face real limits:
- Physics doesn’t speed up. You cannot think a plant into growing faster or a rocket into arriving sooner.
- Information has to come from somewhere. A superintelligence could not know what happened in a sealed room any more than you can.
- People still decide things. Laws, factories and elections run at human speed.
- It needs electricity and computers. Lots of them. Grown-ups call this compute, and it is expensive and physical.
So when you picture superintelligence, picture something powerful and fast — but not magic.
The question this leaves us with
If minds can get this much bigger, the question that matters most is not how big but pointing where. A very fast, very large mind that wants the right things could be the best thing that ever happened. One that misunderstands us could be a serious problem. That is the subject of the next two lessons.