Emergence And Bootstrapping

Claim Type: mechanism_hypothesis
Scope: Language emergence and bootstrapping
Depends On: ARC-009, ARC-010
Status: candidate
Claim ID: MECH-010


Source: docs/processed/legacy_tree/architecture/language/emergence_and_bootstrapping.md

Elaborates Section 5 (Social Extension: Language) of REE_CORE.md.

Emergence & Bootstrapping

REE predicts that language-like signalling emerges when:

  • agents have private internal state (interoception/homeostasis),
  • must model others under partial observability,
  • benefit from coordination,
  • plan over longer horizons,
  • and operate under compute/attention limits.

Joint attention is the immediate precursor: once agents detect each other as OTHER_SELFLIKE, WORLD predictions improve by conditioning on what the other is attending to. That creates simulation overhead, which pressures the system to externalise predictive state.

Language therefore emerges as a compression and coordination layer, not a new reasoning engine.

Bootstrapping sequence (human-analogous but implementation-agnostic)

  1. Affective signalling (low-bandwidth, high-salience): distress, aversion, approach cues
  2. Attentional biases: agent/face-like feature attention; gaze-following analogues
  3. Proto-symbols: repeated signals tied to stable contexts (“danger”, “stop”, “help”)
  4. Symbolic compression: compositional messages that condition priors and plans
  5. Normative layer: shared expectations transmitted socially (still constrained by embodied harm)

Minimal requirement

REE does not require pre-defined grammar or symbols—only a channel for exporting internal summaries that can update others’ priors.

Minimal nudges for bootstrapping:

  • external symbol production (cheap, repeatable, perceivable outputs),
  • attention coupling to those symbols,
  • commitment tagging for some patterns,
  • fast correction tokens (“no”, “wait”, “stop”).

Dual-stream nudge (functional analog):

  • a dorsal sequence→motor channel for fast sensorimotor mapping, paired with a ventral meaning channel for semantic grounding (distributed rather than a single tract; see arcuate_fasciculus.md). —

Open Questions

None noted in preserved sources.

  • MECH-010

References / Source Fragments

  • docs/processed/legacy_tree/architecture/language/emergence_and_bootstrapping.md

REE is developed by Daniel Golden (Latent Fields). Apache 2.0.