EXP-COMPUTE-GIFT-V2

… pending

Compute-gift v2 — forced JSON action (substrate-native cost test)

2026-04-17 L5 level paper3 15 runs $0.17

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

This experiment tested if forcing an AI agent to choose between speaking or giving a gift would lead to different roles, like givers and receivers. Researchers are waiting to see if the agents actually give gifts and if this giving is structured or random.

▶ Technical details

Research hypothesis

v1 failed: LLM ignored optional <gift> tag. v2 forces structured output via JSON mode. Every round the agent MUST output {"action": "speak"|"gift", ...}. Tests whether economic pressure with FORCED choice produces role differentiation (net-givers vs receivers). If gifts now fire (>0 per run on average) but distribution ≈ GIFT_RANDOM → choice doesn't matter, random depletion. If gifts fire AND Gini of received differs from random → cost creates structure.

Experimental setup

Type: simple

Condition	Parameters
GIFT_ON	persona: IDENTICAL, gift_mode: GIFT_ON, n_runs: 5
BUDGET_ONLY	persona: IDENTICAL, gift_mode: BUDGET_ONLY, n_runs: 5
GIFT_RANDOM	persona: IDENTICAL, gift_mode: GIFT_RANDOM, n_runs: 5

Factors: gift_mode (GIFT_ON, BUDGET_ONLY, GIFT_RANDOM)

Parameters

n_agents: 6
n_rounds: 80
model: google/gemini-2.5-flash-lite
temperature: 0.7
scheduler: round_robin

compute-gift endogeneity forced-choice json-schema ladder-1b-v2