Learning = changing in a way that allows you to solve (a certain class of) problems more efficiently (on average).
Not learning = either not changing, or changing in a way that does not make you more efficient at solving problems.
(Note: I am saying "on average", because... suppose your original algorithm for solving math problems is simply yelling "five!" regardless of the problem. Now you learn math, and it makes you better at solving math problems in general... but it makes you slower at solving those problems where "five" actually happens to be the correct answer.)
It means different things in different contexts. Sometimes, especially in political/status discussions, it can mean multiple things at the same time (see https://rationalwiki.org/wiki/Motte_and_bailey).
Generally, it's about an agent's change in model, information, or information-organization that allows it to effect more reliable actions (including actions that are simply decision outputs, and actions which have a physical component, like swinging a cricket bat).
I want to know what learning actually is. Ideally you can provide a model that is compact and useful as possible, such that when I encounter the word 'learn' in daily life I could replace it with this model while seeing a complete set of unique gears within it such that I can look at a real-world learning system and see how its traits correspond to each of those parts.
Generative subquestions:
What results from Tabooing it? What are its components? Is there a precise General Theory of Learning that underpins humans, animals, ML agents, or any other things that learn? If there are multiple constructs being pointed to with 'learn': what are the differences between systems that 'learn'? How many different kinds of 'learning' are there and what do they look like?
What isn't learning?