EXP-CA-RULE-MULTIGEN-VARIANCE

… pending

CA + multi-gen variance characterisation (paper6 step 3 char)

2026-05-04 L6 level paper6 10 runs $0.16

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

Researchers are investigating how different starting conditions affect an artificial intelligence's ability to propose changes to its own rules. They want to see if the AI consistently favors modifying rules related to a specific type of pattern, and if a particular part of its reasoning process is consistently disrupted. The results are still being collected and analyzed.

▶ Technical details

Research hypothesis

Step 3 pilot (N=2, exp-ca-rule-multigen-poc) showed provisional positive: agents propose CA-axis modifications (7/10 of accepted), reasoning grounds in CA observations (gliders, density), step-2 LLM-only chain partially disrupts. Question: how variable is this across diverse seed conditions? Does the CA-axis proposal preference replicate? Is the step-2 chain disruption consistent or pilot-specific? N=10 with 5 diverse seed sets × 2 rotations. Same engine (ca_rule_multigen) as pilot. Pre-registered: - CA-axis proposal share: ≥50% of accepted proposals are on CA axes (vs pilot 70%) - signal_type chain: ≤30% of runs reach 'concept' endpoint (vs step-2 50%, pilot 0%) - reasoning grounding: ≥60% of proposal reasoning explicitly references CA state (gliders, density, structures, patches)

Experimental setup

Type: simple

Condition	Parameters
COUPLED_VARIANCE	coupling: ALL, n_runs: 10

Factors: coupling (all)

Parameters

n_agents: 4
n_rounds: 90
model: google/gemini-2.5-flash-lite
temperature: 0.85

autopoiesis paper6-step3-char hybrid-substrate variance