Learning is compression.
Memorization is not.
In Zone 03, Q13, you measured the entropy of your own text. That number was Kolmogorov complexity in real time. You were using his definition without knowing his name.
Two sequences. Same length. Write the shortest rule you can that generates each. Watch which one resists.
Your rule is shorter than the sequence. The pattern compressed. This sequence is learnable.
No rule shorter than the sequence exists. It must be memorized.
Kolmogorov defined complexity in 1963 as the length of the shortest description of a thing.
A truly random sequence cannot be compressed: its shortest description is itself. A learnable pattern can always be compressed.
This is why neural networks work. They are compression engines, finding the shortest description of the structure in their training data. A model that has truly learned can describe a million examples in fewer parameters than one that memorized them.
Kolmogorov worked in Moscow under Stalin. He navigated ideological pressure on mathematics while simultaneously defining probability theory, turbulence, and computational complexity.
In 1963 he defined algorithmic complexity: the length of the shortest computer program that can generate a given string. Truly random data has maximum Kolmogorov complexity — it cannot be compressed. Learnable patterns have low complexity — they can be described in fewer rules than examples.
This distinction defines the difference between memorization and learning.
The entropy meter in Zone 03 of The Inquiry measures Kolmogorov complexity in real time. You were using his definition without knowing his name.