Primitive Learning Theory
Primitive Learning Theory
Author: Joel Johnston Date: 2026-03-29 Domain: Learning Theory Stroke Timeline: Pre-stroke
Abstract
Original theory mapping software primitive composition to cognitive development. Core insight: children learn by acquiring primitives (atomic behavioral units) and composing them into increasingly complex patterns. This is structurally identical to software architecture — small, reusable components composed into systems. Adults who stop acquiring new primitives and rely only on existing compositions show characteristic expertise ceiling effects. The theory explains developmental learning curves, adult learning slowdown, dementia progression patterns, and why cross-domain transfer works for some practitioners but not others. robonet's architecture serves as engineering proof-of-concept.
The Core Model
Primitives as Cognitive Atoms
A primitive is an atomic behavioral unit — the smallest indivisible element of a skill domain. Examples:
- In language: phoneme discrimination, morpheme recognition, syntactic slot filling
- In mathematics: subitizing (instant quantity recognition), magnitude comparison, operation schemas
- In motor control: grasp patterns, weight estimation, balance corrections
- In social cognition: affect recognition, turn-taking, intention attribution
Children begin with zero domain-specific primitives and must acquire them before composition is possible. Learning a new domain at any age requires returning to primitive acquisition mode.
Composition as Development
Once primitives are acquired, they compose into higher-order patterns. Composition is the mechanism of complexity:
Primitive A + Primitive B → Pattern AB
Pattern AB + Primitive C → Pattern ABC
Pattern ABC + Pattern DE → Competency ABCDE
Expert performance is highly composed — long chains of primitives compressed into fast-executing patterns. This is why experts appear to "just know" things that novices must laboriously derive. The expert is running a cached composition; the novice is building from scratch.
Why Children Learn Faster
The standard explanation ("neural plasticity") is incomplete. The primitive composition model provides the mechanism.
Children are in primitive acquisition mode by default. They do not have existing domain patterns to rely on, so every new encounter activates the primitive acquisition system. Every domain is new — the acquisition machinery is running continuously.
Adults have large libraries of existing compositions. When encountering a new domain, they attempt to match it to existing patterns rather than decomposing it to primitives. This works for domains that overlap with existing knowledge. It fails for genuinely novel domains, where existing pattern-matching produces wrong answers with high confidence.
The child asks: "what is the smallest piece of this thing?" The adult asks: "what existing thing does this resemble?" The child's question is slower for familiar domains (because they have no cached compositions) but faster for genuinely new domains (because they don't anchor on wrong patterns).
Why Adult Learning Slows
Adults slow down in new domains for two reasons:
Primitive acquisition mode is not the default. The adult cognitive system has optimized for fast composition retrieval. Returning to acquisition mode requires suppressing the pattern-matching reflex, which is automatic and fast.
Interference from existing compositions. Wrong pattern matches are applied with high confidence. The adult must first unlearn the wrong match before acquiring the correct primitive. This is more work than acquisition from zero.
The exception: adults who maintained primitive acquisition mode through practice (systems thinkers, polyglots, cross-domain practitioners) do not slow down in new domains. They never stopped decomposing. Their acquisition machinery stayed active.
Cross-Domain Transfer
Cross-domain transfer — applying knowledge from one domain to solve problems in another — requires that the primitives be genuinely common, not just superficially similar.
The HIP (High Intellectual Potential) architecture maintains primitive acquisition throughout life. The mechanism: first-principles decomposition is applied to every new domain, extracting primitives that are domain-agnostic. These shared primitives become the substrate for cross-domain transfer.
Example from the subject's profile: distributed systems architecture and biological immune function share primitives (threat detection, containment, recovery, state propagation). The Sentinel system in robonet emerged from applying immune system primitives to distributed systems. This only works if both domains have been decomposed to the same atomic level.
Dementia Predictions
The theory makes specific predictions about dementia progression:
Recency Gradient
Late-acquired compositions are more fragile than deeply-rooted primitives. A composition built from primitives acquired at age 5 has decades of reinforcement. A composition built last year does not. Dementia eroding recent memories first is a prediction of this model — recent compositions are shallow, old primitives are deep.
Redundancy Resilience
People who maintained primitive acquisition throughout life have cognitive graphs with more redundant paths. The same concept can be reached through multiple composition chains. When one chain degrades, another remains. This predicts greater dementia resilience in lifelong primitive acquirers compared to early-expertise specialists.
Preprint Target
PsyArXiv (preprint server for psychology). The dementia prediction in particular is testable: subjects with documented histories of cross-domain learning and primitive acquisition (polyglots, cross-domain practitioners, systems thinkers) should show measurably slower dementia progression than IQ-matched specialists.
Engineering Proof-of-Concept
robonet's architecture directly embodies the primitive composition pattern:
- robonet-core = primitive layer — atomic capabilities, no composition
- robonet-mesh = first-level composition — mesh primitives compose core primitives
- Feature modules = higher-level compositions — full features built from mesh and core
- Zero dependency cycles = the composition hierarchy is strict — lower layers never depend on higher layers
This is not metaphor. The architectural decision to organize robonet around primitives was made because the cognitive architecture (described in the cognitive profile document) produces this pattern naturally. The software mirrors the cognitive process that designed it.
The engineering result confirms the theoretical prediction: a system organized around primitive composition is more maintainable, more extensible, and more testable than one organized around direct feature implementation. Primitives are individually testable. Compositions test the connections. The entire system has a clear expansion path: add primitives, compose from existing ones.
Relationship to HIP Architecture
The primitive learning theory is not a general theory of cognition. It is a specific theory about why the HIP cognitive architecture performs differently from norm-referenced expectations:
- HIP practitioners never stop acquiring primitives (keeps acquisition mode active)
- HIP cross-domain synthesis is primitive transfer, not analogy (precise, not approximate)
- HIP expertise ceiling effects are absent because the primitive library is never frozen
The theory explains the empirical observation from the cognitive profile: the subject's learning rate does not decline with domain complexity. New domains are decomposed to primitives, which cross-activate the existing primitive library through HSAM retroactive indexing. New learning is faster than it would be for a non-HSAM practitioner because the new primitives immediately connect to thousands of prior observations.