EXP-PERSONA-NOVELTY-BASIN
… pendingPersona-conditional novelty basin map (paper7 territory) — 5 personas + control
In plain language
This experiment tested whether different "personas" or viewpoints could lead an AI to generate more novel ideas. Researchers found that when the AI adopted different personas, the rate of novel idea generation varied significantly, suggesting that the AI's underlying structure might have untapped creative potential in different areas. The experiment is still ongoing to fully map out these creative "basins" and understand how different personas influence the AI's output.
▶ Technical details
Research hypothesis
Cross-run lineage-archive over 11 prior experiments (419 accepted rule changes) showed novel-value rate of only 0.24%. ALL 11 used the same scientific/methodical persona framing. Hypothesis: LLM substrate has multiple attractor basins; we have been sitting in one (scientific) which happens to be low-novelty. Other basins (artistic, surrealist, childlike, inventor, alien, jazz) may emerge novelty at a much higher rate. Empirically testable with 6 persona conditions × 5 runs = 30 runs, identical otherwise (rule_multigen, meta_modifiable=true, 4 agents, 5 gens × 20 rounds, budget=400, gemini-flash-lite).
PRIMARY: persona-conditional novelty rate ratio.
- MULTIPLE-BASINS: max(novelty_rate)/min(novelty_rate) ≥ 3x across 6 conditions → substrate has untapped novelty regions, persona-basin hypothesis confirmed.
- SUBSTRATE-CEILING: uniformly <1% across all 6 → basin hypothesis falsified; constraint is deeper than persona, true regardless of framing. Returns direction to external-food / multi-substrate.
KILL: any persona produces semantic incoherence (parse failures >10%, agents drift into non-task generation) → persona prompt too aggressive, broke scaffold.
SECONDARY: cross-persona attractor map. Compare final_rules per persona — do different personas converge to different attractor profiles, or to the SAME (coupling=strong, voting=supermajority etc.). Different attractors per persona = different basins. Same attractors despite different personas = persona is cosmetic, structure determines.
Experimental setup
Type: factorial
| Condition | Parameters |
|---|---|
| SURREALIST | coupling: true, must_propose: true, axis_diversify: true, meta_modifiable: true, mute_signal: false, agent_spawn_enabled: false, n_runs: 5 |
| CHILD | coupling: true, must_propose: true, axis_diversify: true, meta_modifiable: true, mute_signal: false, agent_spawn_enabled: false, n_runs: 5 |
| INVENTOR | coupling: true, must_propose: true, axis_diversify: true, meta_modifiable: true, mute_signal: false, agent_spawn_enabled: false, n_runs: 5 |
| ALIEN | coupling: true, must_propose: true, axis_diversify: true, meta_modifiable: true, mute_signal: false, agent_spawn_enabled: false, n_runs: 5 |
| JAZZ | coupling: true, must_propose: true, axis_diversify: true, meta_modifiable: true, mute_signal: false, agent_spawn_enabled: false, n_runs: 5 |
| CONTROL | coupling: true, must_propose: true, axis_diversify: true, meta_modifiable: true, mute_signal: false, agent_spawn_enabled: false, n_runs: 5 |
Factors: persona (SURREALIST, CHILD, INVENTOR, ALIEN, JAZZ, CONTROL)
Parameters
- n_agents
- 4
- n_rounds
- 100
- model
- google/gemini-2.5-flash-lite
- temperature
- 0.85
Trophic Ratios by Condition
Mean trophic ratio per agent across runs. Error bars = ±1 std dev. Higher TR = more upstream (exporter).