The Ladder, End to End

Fifteen stops

The interactive figure below gathers the whole story of the course onto a single continuous canvas. Every stop is a technological step we built together: the passage from one to the next is a change of perspective, a step back through which what before seemed the whole universe becomes merely a component of the next storey.

Where the measuring ruler changes

The first five stops can be measured in metres and nanometres, in a classical spatial hierarchy: electric charges move inside a crystal of silicon; the crystal forms the transistor; transistors make up the logic gates; the gates form memory and flip-flops; and these join to make up the processor. This stretch of the journey is an ordinary geometric magnification, from the microscopic scale of atoms to the visible scale of chips.

But from the sixth stop onwards the spatial ruler loses its relevance. A number is no larger in centimetres than a processor, and a neural network takes up no more physical space than a line of code. They sit one on top of another in a different way: as levels of abstraction. What does stay intact is the fundamental principle of the whole course: every storey uses the components beneath it as ready-made bricks, without needing to know how they are built in detail.

This, at bottom, is the great architecture of computers and of artificial intelligence: a transistor does not know what a logic gate is; a logic gate does not know what a number is; an artificial neuron does not know what a sentence is; and a language model does not know what the software tool it calls actually is. Every level receives only what the level below offers it and produces a stable result, ready for the next storey.

What a machine really knows

Looking at this ladder from one end to the other, we discover something liberating: none of the individual components is conscious or “intelligent” in itself. The electron obeys the physics of semiconductors, the transistor switches on differences of voltage, the weights are adjusted by a mathematical derivative, and the next word is selected by a statistical distribution. What we perceive as intelligence, understanding or creativity comes from no magical spark hidden somewhere along the way, but from the enormous number of simple steps laid one on another at a monumental scale.

From this basic simplicity, however, follow the natural limits we met along the way: a model that merely predicts the most plausible continuation of a text has no consciousness of error and no direct anchor in physical reality. The same statistical principles that give it its power also draw its boundaries.

The title of our course was a simple question: what does a machine know? After forty-eight lessons taken step by step, the most honest and profound answer is that the machine knows what pattern naturally follows at each level, and nothing more. All the rest — what seems to us thinking, intuition or living dialogue — is the spectacular way in which these modest steps, combined billions of times, come to reflect the complexity of the human mind. Whether this reflection is only a sophisticated mirror or the beginning of something entirely new remains an open question. What we can say with certainty is that we now know, step by step, what it is built of.