← back to timeline

EXP-ROLE-FRAMING-ASYMMETRY

pending

Role framing asymmetry — LLM-as-cell vs LLM-as-controller

2026-05-01 L6 level paper5 4 runs $0.02
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

This experiment investigates how an AI's assigned role affects its collective behavior. Researchers are comparing two scenarios: one where the AI acts as a single cell coordinating its own actions, and another where it acts as a controller directing multiple other AI cells. The results are still being analyzed to see if the role framing influences the structure of the AI group's interactions.

Technical details

Research hypothesis

Paper5 shows LLM-as-cell (goal-pursuit + state coupling) produces 2x specialization vs uncoupled. Le et al. GECCO'25 uses LLM-as-controller (LLM generates interventions/directives for cells). Does role framing (cell vs controller) produce qualitatively different collective structure? Prediction: controller framing → hub-spoke directive pattern (one agent coordinates others). Cell framing → distributed specialization (our paper5 result). If CONTROLLER ≈ CELL on progress but different on structure → framing determines structural type, not progress. If CONTROLLER < CELL → goal- directed pursuit is necessary for specialization acceleration.

Experimental setup

Type: factorial

Condition Parameters
CELL_COUPLED coupling: true, coupling_mode: real, goals_mode: diverse, controller_mode: false, n_runs: 2
CONTROLLER_COUPLED coupling: true, coupling_mode: real, goals_mode: diverse, controller_mode: true, n_runs: 2

Factors: framing (cell, controller)

Parameters

n_agents
4
n_rounds
40
model
google/gemini-2.5-flash-lite
temperature
0.85
levin-bioelectric controller-framing role-framing le-et-al-discriminating-test paper5-competitive