What a self-learning agent writes down
Self-learning engineering is the practice of deciding what an agent may conclude from its own attempts, and what has to be true before it is allowed to act on that conclusion.
The definition
A self-learned rule is a policy the agent wrote from its own attempt, and reads on a later attempt. Two halves, and both are load-bearing. The agent wrote it — nobody reviewed it. And it is a policy rather than a record: it says what to do, not what happened.
Four things it is not
The word "learning" carries a great deal of freight, and three of the four things it usually means are not this subject.
| what it changes | is it this volume? | |
|---|---|---|
| training, fine-tuning | the weights | no — you are not running gradient descent in a loop |
| a transcript | nothing; it is a recording | no — nothing in it claims how to act |
| a memory | a record of what was true | no — a memory makes a claim about the world |
| a self-learned rule | what the agent will do next time | yes |
The distinction from memory is the one worth being careful about, because the machinery looks identical — something is written in one attempt and read in a later one. A memory says the cap is 180. A rule says fold unicode before comparing. The first can be checked against the world; the second can only be checked against outcomes, and the whole of Part 2 is about what happens when the outcome you check against is the wrong one.
The shape of the loop
Steps 2 and 3 are where the subject lives. Step 4 is where most of the engineering effort goes, and step 2 is where the result is decided.