The AGI Game - Grand Parameter Sweep Dataset

Overview. Complete output of a parameter sweep over the AGI Game — a game-theoretic model of AI dispensability dynamics. The sweep explores 31,944 distinct configurations of the model's five governance parameters — M (number of builder states ∈ {1,2,3,4}), C (frontier firms per state ∈ {1..6}), O (open-source level ∈ {0.0..1.0}), R (regulation ∈ {0.0..1.0}), E (enforcement deployment ∈ {0.0..1.0}) — at 100 simulation timesteps per configuration. Contents. grand_sweep_summary.csv — one row per configuration; 54 columns covering parameters, dominant trajectory archetype, final leverage / capability / commoditization, value-layer distribution, narrative-richness metrics, cluster-displacement event times. grand_sweep_timeseries/ — 31,944 per-configuration CSVs; full time series of κ (5-D capability), leverage, alignment, mobilization, coalition status, commoditization, value migration, and dispensability for each of 14 occupation clusters at every timestep × state archetype. cluster_index.csv — mapping of 14 occupation clusters to their dispensability vectors. pattern_analysis/ — derived analytical outputs: trajectory distribution, classification ambiguity, parameter-pair heatmaps (10 pairs × 3 metrics), adjacent-cell transitions, per-trajectory exemplars, and dramatic single-step trajectory flips. data_driven_findings.txt — human-readable findings report. Methodology. The simulation engine implements a 7-phase per-timestep loop: Nash equilibrium solve via Population Replicator Dynamics; supply-chain constraint update; open-source commoditization dynamics; company state updates; capability ODE evolution; dispensability and leverage computation with the governance-suppression formula; collective-action coalition dynamics. Each configuration runs for 100 timesteps from the canonical initial capability state. Trajectory classification uses a continuous scoring system over 8 named archetypes (Governed Multipolarity, Competitive Tension, Bipolar Standoff, Regulatory Preservation, Open-Source Paradox, Captured Hegemony, Gatekeeping Inversion, Algocratic Convergence). License. Released under CC-BY-4.0. Cite the dataset using the Zenodo DOI.### 概述 本数据集为**AGI博弈（AGI Game）**的参数扫掠完整输出——AGI博弈是一款刻画AI可替代性动态的博弈论模型。本次参数扫掠共探索了该模型5项治理参数的31944种不同配置：M（构建体数目，取值范围{1,2,3,4}）、C（每个构建体的前沿企业数量，取值范围1~6）、O（开源水平，取值范围0.0~1.0）、R（监管强度，取值范围0.0~1.0）、E（执法部署度，取值范围0.0~1.0），且每种配置均运行100个仿真时间步。 ### 数据集内容 grand_sweep_summary.csv — 每行对应一种配置，共包含54列，涵盖参数、主导轨迹原型、最终杠杆率/能力/商品化程度、价值层分布、叙事丰富度指标、集群位移事件时间。 grand_sweep_timeseries/ — 包含31944个单配置CSV文件，存储每种配置下，14个职业集群在每个时间步及每个状态原型下的κ（五维能力）、杠杆率、对齐度、动员度、联盟状态、商品化程度、价值迁移与可替代性完整时间序列。 cluster_index.csv — 14个职业集群与其可替代性向量的映射表。 pattern_analysis/ — 衍生分析输出，包括轨迹分布、分类歧义度、参数对热力图（10组参数对 × 3项指标）、相邻单元格转移、每条轨迹的示例样本，以及显著的单步轨迹翻转案例。 data_driven_findings.txt — 面向人类读者的实证研究发现报告。 ### 研究方法 本仿真引擎采用每时间步7阶段循环流程：基于种群复制动态（Population Replicator Dynamics）求解纳什均衡、更新供应链约束、模拟开源商品化动态、更新企业状态、开展能力常微分方程（ODE）演化、通过治理抑制公式计算可替代性与杠杆率、模拟集体行动联盟动态。所有配置均从标准初始能力状态启动，共运行100个时间步。轨迹分类采用针对8种命名原型的连续评分体系，分别为：受治多极格局（Governed Multipolarity）、竞争张力（Competitive Tension）、两极对峙（Bipolar Standoff）、监管留存（Regulatory Preservation）、开源悖论（Open-Source Paradox）、俘获型霸权（Captured Hegemony）、守门人反转（Gatekeeping Inversion）、算法统治趋同（Algocratic Convergence）。 ### 授权协议 本数据集采用CC-BY-4.0协议发布，引用时请使用其Zenodo数字对象标识符（DOI）。