← back to timeline

EXP-META-MODIFIABLE

pending

Meta-modifiability — Bedau OEE rule-space expansion test (paper6 step 4)

2026-05-04 L6 level paper6 6 runs $0.08
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

This experiment tested if artificial agents could expand their own rule sets to create new behaviors. Researchers found that agents did propose new rules, but it is still being determined if these new rules were accepted and used to create even more complex behaviors over time.

Technical details

Research hypothesis

Step 1-3 used a fixed RULE_SPACE (4 LLM axes + 3 CA axes, each with 3-5 values). The developmental program operated within that fixed space. Bedau OEE requires UNBOUNDED compositional growth — rule space must be able to expand. Step 4 (this experiment): agents may propose NEW values for existing axes (not in initial RULE_SPACE). Accepted novel values are appended to a per-run mutable rule space. Tests whether agents (a) propose novel values at all, (b) get them accepted, (c) build on them in subsequent generations. Falsifiable predictions: - OEE-positive: ≥1 novel value accepted in ≥3/6 runs, AND ≥1 of those novel values gets re-modified in a later generation (compositional depth on novel-value chain).

  • OEE-conservative: novel values proposed but rarely accepted (≤1/6 runs). Voters reject anything not in their training distribution.
  • OEE-null: agents never propose novel values. Conservative within fixed rule space.

This is CONSTRUCT — new architecture variant that adds genuine rule-space mutability (vs step 1-3 fixed-space modifications).

Experimental setup

Type: factorial

Condition Parameters
META_MODIFIABLE coupling: true, must_propose: true, axis_diversify: true, meta_modifiable: true, n_runs: 6

Factors: meta (true)

Parameters

n_agents
4
n_rounds
100
model
google/gemini-2.5-flash-lite
temperature
0.85
autopoiesis paper6-step4 meta-modifiability bedau-OEE construct