EXP-SCARCITY-BUDGET
… pendingGenuine scarcity — token budgets, agents die when exhausted (paper6 step 5)
In plain language
This experiment tested what happens when artificial agents have a limited "word budget" and go silent once they've used it all up. Researchers wanted to see if the agents' ability to develop and pass on their "programs" would be affected when they could no longer communicate. The results are still pending.
▶ Technical details
Research hypothesis
Steps 1-3 had no scarcity — agents could propose/vote indefinitely. Selection was social (vote-mediated), not physical. Step 5 introduces token budgets: each agent has a finite word budget per run; when exhausted, agent goes silent (skipped, no further proposals/votes). Selection becomes physical. Operational definitions: - Agent dies = budget exhausted = silent for remaining rounds - Selection physical = rule lineage continues only via surviving agents - Tests: does developmental program persist when speakers run out? Pre-registered: - PHYSICAL-SELECTION-EFFECT: ≥1/6 runs has all agents dead before final gen. Lineage truncated. Reveal: budget=word-cost-of-coupling.
- GRACEFUL-DEGRADATION: agents reduce verbosity, run lasts full 5 gens. Adaptive behavioral compression.
- NO-EFFECT: budget too generous, no deaths. Try again with tighter budget.
Budget choice: 800 words (≈ 4 generations × 5 turns × 40 words). With current verbose outputs (~80 words/turn), most agents will exhaust by gen 4. Calibration based on observed outputs in step-2 N=10 runs. This is CONSTRUCT — first time selection becomes physical (vs social voting-mediated selection in steps 1-3).
Experimental setup
Type: factorial
| Condition | Parameters |
|---|---|
| SCARCITY_BUDGETED | coupling: true, must_propose: true, axis_diversify: true, n_runs: 6 |
Factors: scarcity (true)
Parameters
- n_agents
- 4
- n_rounds
- 100
- model
- google/gemini-2.5-flash-lite
- temperature
- 0.85