← back to timeline

EXP-MUTE-COUPLING

pending

Mute coupling — Jaynes-style language-vs-substrate discriminator (paper6 step-2 variant)

2026-05-04 L6 level paper6 6 runs $0.05
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

This experiment tested whether communication between artificial agents relies on the meaning of their words or just the structure of their signals. Researchers observed how deeply the agents' communication systems evolved when they could only share numerical information, without any language. The results are still being collected to see if removing language significantly hinders the development of complex communication.

Technical details

Research hypothesis

Step-2 (with verbal coupling) showed mean lineage depth 7.2 with universal attractors. Coupling-ablation N=10 showed mean depth 4.0. Question: does coupling work via the LANGUAGE content of agent broadcasts, or via the numeric/structural state alone? Jaynes-strict prediction: depth ≤2 in mute condition. Language is necessary for the recursive self-reference that produces "I"-like coupling. Substrate-coupling prediction: depth ~4-7 in mute condition. Coupling shapes behaviour even without verbal content; numeric state-sharing is sufficient. Surfing 2026-05-04 evening note: Jaynes critics (LessWrong) point out that Jaynes's strict timing-claim is falsified by Hammurabi (1800 BC) coup planning. So we should expect a GRADED effect, not binary collapse — likely somewhere between full-language (7.2) and no-coupling (4.0). Pre-registered: - JAYNES-STRICT: depth ≤2 (matching ablation pilot) - SUBSTRATE-COUPLING: depth ≥5 (only modest reduction from verbal) - GRADED: depth 3-5 (intermediate effect, language content matters but isn't everything)

This is a CONSTRUCT experiment — new architecture variant (mute_signal factor), not replication of existing finding.

Experimental setup

Type: factorial

Condition Parameters
MUTE_NUMERIC coupling: true, must_propose: true, axis_diversify: true, mute_signal: true, n_runs: 6

Factors: coupling (mute)

Parameters

n_agents
4
n_rounds
100
model
google/gemini-2.5-flash-lite
temperature
0.85
autopoiesis paper6-step2-variant jaynes-discriminator construct