The encyclopedic body of the Compendium. Each type of AI from the panel (Part II) gets a chapter that details it end to end — including the specialized life cycle (the 11 stages of Part I, now with the concrete details of the category).
The chapter template (8 sections)
Every chapter follows the same structure, so that the Compendium is consistent and comparable across types:
| § |
Section |
Content |
| 1 |
Definition and short history |
What it is, in one page; lineage and milestones (links to Part III) |
| 2 |
Foundations |
Sciences and mathematics specific to the type |
| 3 |
Algorithms and architectures |
The machine on the inside: components, variants, key formulas |
| 4 |
Inputs |
Hardware, data, data structures, systems/MLOps |
| 5 |
Specialized life cycle |
The 11 stages (Part I) applied to this type |
| 6 |
Capabilities, modes and modalities |
What it can do; types of intelligence covered |
| 7 |
Limits, risks and ethics |
Failures, biases, cost, alignment; [metaphor] vs. mechanism |
| 8 |
State of the art and examples |
Real systems; frontier trends |
Size rule: a chapter starts as a single file NN-slug.kmd. When it grows beyond what is comfortable to read/edit, it is promoted to a directory NN-slug/ with the 8 sections in numbered docs (same pattern as Part I).
Chapter index (card → chapter map)
Organized by paradigm (Level 1 of the taxonomy). Each chapter corresponds to a card from Part II.
Connectionist (Deep Learning)
Symbolic
Probabilistic / Bayesian
Evolutionary
Reinforcement
Hybrid / Neuro-symbolic
34 chapters, one per card, plus ch. 35 (Markov Chain / n-gram LM) — the memoryless ancestor of the language model, added under the Probabilistic paradigm. The LLM pilot chapter fixes the template; the others follow it. Gaps in the taxonomy (intersections not covered in the panel) may become extra chapters numbered in sequence.