EXP-ASYM

informative

Minimal Asymmetry Experiment

2026-03-08 10:00:00+00:00 L2 level paper3 9 runs

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

Give three AI agents different cognitive styles (analyst, associator, practitioner) and a stable hierarchy emerges. The practitioner (who "grounds" abstractions) becomes the word source, while the associator becomes the receiver. Neither personas without interaction nor interaction without personas create structure — you need both.

What we found

UNDERPOWERED (N=9): 3 directionally consistent (need ≥12 runs), 2 refuted, 1 borderline

First controlled demonstration: persona asymmetry + interaction → stable role differentiation. Structural metrics (TR, role persistence) go UP with differentiation; convergence metrics (VP_excess, PID synergy) go DOWN.

Predictions we made before running

0/6 confirmed · 2 refuted · 1 borderline

▶ Technical details

Research hypothesis

Minimal persona asymmetry (cognitive style) + interaction produces stable directional influence and role differentiation beyond what identical agents produce via diffusion.

Experimental setup

Type: factorial

Condition	Parameters
PERSONA_LIVE	persona: PERSONA, interaction: LIVE
IDENTICAL_LIVE	persona: IDENTICAL, interaction: LIVE
PERSONA_ISOLATED	persona: PERSONA, interaction: ISOLATED
IDENTICAL_ISOLATED	persona: IDENTICAL, interaction: ISOLATED

Factors: persona (PERSONA, IDENTICAL) × interaction (LIVE, ISOLATED)

Parameters

n_agents: 3
n_rounds: 60
n_runs_per_condition: 9
model: gemini-2.5-flash-lite
temperature: 0.9
scheduler: round_robin

Trophic Ratios by Condition

Mean trophic ratio per agent across runs. Error bars = ±1 std dev. Higher TR = more upstream (exporter).