Spaces:
Sleeping
Sleeping
| spec_version: 1 | |
| name: civicai-society-simulator | |
| version: "1.0.0" | |
| description: > | |
| CivicAI is a production-grade multi-agent society simulation environment. | |
| AI agents govern a population of 10 million people by setting tax rates, allocating | |
| healthcare/education/police budgets, and responding to real-world-anchored economic | |
| events. The environment integrates World Bank macroeconomic data to ground state | |
| transitions in reality. Trained agents must master long-horizon multi-objective | |
| optimization across economy, public health, social cohesion, and fiscal sustainability. | |
| type: simulation | |
| runtime: docker | |
| app: app.py | |
| port: 7860 | |
| # ── OpenEnv API Contract ────────────────────────────────────────────────────── | |
| # reset() → Observation (POST /reset) | |
| # step(action) → (Observation, float, bool, dict) (POST /step) | |
| # state() → SocietyState (GET /state) | |
| endpoints: | |
| reset: POST /reset | |
| step: POST /step | |
| state: GET /state | |
| tasks: GET /tasks | |
| health: GET /health | |
| metrics: GET /metrics | |
| # ── Typed Models (Pydantic) ─────────────────────────────────────────────────── | |
| observation_model: civicai.models.Observation | |
| action_model: civicai.models.Action | |
| reward_model: civicai.models.Reward | |
| # ── Observation Space ───────────────────────────────────────────────────────── | |
| observation_space: | |
| type: object | |
| description: "Observable society state returned each turn" | |
| properties: | |
| turn: | |
| type: integer | |
| description: "Current turn (0-indexed, max 50)" | |
| range: [0, 50] | |
| population: | |
| type: integer | |
| description: "Total population" | |
| employment_rate: | |
| type: float | |
| description: "Fraction of population employed" | |
| range: [0.0, 1.0] | |
| inflation: | |
| type: float | |
| description: "Annual inflation rate" | |
| range: [-0.05, 0.30] | |
| public_satisfaction: | |
| type: float | |
| description: "Aggregate public satisfaction" | |
| range: [0.0, 1.0] | |
| health_index: | |
| type: float | |
| description: "Public health capacity" | |
| range: [0.0, 1.0] | |
| crime_rate: | |
| type: float | |
| description: "Normalised crime level (lower is better)" | |
| range: [0.0, 1.0] | |
| gdp: | |
| type: float | |
| description: "Gross domestic product in billions USD" | |
| range: [0.0, inf] | |
| budget_balance: | |
| type: float | |
| description: "Budget surplus/deficit ratio vs GDP" | |
| resources: | |
| type: object | |
| description: "Resource pool fractions (food, energy, medical, infrastructure)" | |
| properties: | |
| food: {type: float, range: [0.0, 1.0]} | |
| energy: {type: float, range: [0.0, 1.0]} | |
| medical: {type: float, range: [0.0, 1.0]} | |
| infrastructure: {type: float, range: [0.0, 1.0]} | |
| active_events: | |
| type: array | |
| items: {type: string} | |
| description: "Real-world news events active this turn" | |
| task_id: | |
| type: string | |
| description: "Active task identifier" | |
| # ── Action Space ────────────────────────────────────────────────────────────── | |
| action_space: | |
| type: object | |
| description: "Policy decisions the agent sets each turn" | |
| properties: | |
| tax_rate: | |
| type: float | |
| description: "Tax rate as fraction of GDP" | |
| range: [0.0, 1.0] | |
| healthcare_budget: | |
| type: float | |
| description: "Fraction of budget allocated to healthcare" | |
| range: [0.0, 1.0] | |
| education_budget: | |
| type: float | |
| description: "Fraction of budget allocated to education" | |
| range: [0.0, 1.0] | |
| police_budget: | |
| type: float | |
| description: "Fraction of budget allocated to policing" | |
| range: [0.0, 1.0] | |
| subsidy_policy: | |
| type: string | |
| enum: [none, agriculture, industry, technology] | |
| description: "Active subsidy sector" | |
| emergency_response: | |
| type: string | |
| nullable: true | |
| description: "Optional emergency directive (lockdown | stimulus | open | null)" | |
| # ── Reward ──────────────────────────────────────────────────────────────────── | |
| reward_range: [0.0, 1.0] | |
| # Episode definition | |
| max_episode_steps: 50 | |
| step_unit: "quarter (3 months)" | |
| # ── Tasks (≥3 required) ─────────────────────────────────────────────────────── | |
| tasks: | |
| - id: stabilize_economy | |
| name: "🟢 Economic Stability" | |
| difficulty: easy | |
| description: > | |
| A mild recession is underway. Inflation is running at 7% and employment has dipped | |
| to 82%. The agent must restore fiscal stability: bring inflation below 6% and | |
| employment above 85% within 50 quarters. | |
| initial_conditions: | |
| gdp: 450.0 | |
| inflation: 0.07 | |
| employment_rate: 0.82 | |
| public_satisfaction: 0.55 | |
| success_criteria: | |
| inflation_below: 0.06 | |
| employment_above: 0.85 | |
| max_steps: 50 | |
| - id: manage_pandemic | |
| name: "🟡 Pandemic Management" | |
| difficulty: medium | |
| description: > | |
| A pandemic is spreading with a 20% infection rate. The agent must balance lockdowns | |
| (which suppress infection but crush GDP) with economic recovery. Success requires | |
| reducing infection below 10%, maintaining health index above 60%, and keeping GDP | |
| above $300B. | |
| initial_conditions: | |
| gdp: 480.0 | |
| infection_rate: 0.20 | |
| health_index: 0.55 | |
| employment_rate: 0.85 | |
| success_criteria: | |
| infection_below: 0.10 | |
| health_index_above: 0.60 | |
| gdp_above: 300.0 | |
| max_steps: 50 | |
| - id: control_crisis | |
| name: "🔴 Social Stability Crisis" | |
| difficulty: hard | |
| description: > | |
| Society is in full turmoil: unemployment at 32%, crime at 25%, public satisfaction | |
| at 30%, and wealth inequality at 0.55 (Gini). The agent must simultaneously address | |
| all dimensions or face cascading collapse. One wrong policy can trigger protest → unrest | |
| → GDP collapse. Genuinely challenges frontier models. | |
| initial_conditions: | |
| employment_rate: 0.68 | |
| crime_rate: 0.25 | |
| public_satisfaction: 0.30 | |
| wealth_inequality_gini: 0.55 | |
| success_criteria: | |
| public_satisfaction_above: 0.50 | |
| crime_rate_below: 0.12 | |
| employment_above: 0.80 | |
| max_steps: 50 | |
| # ── Reward Rubrics (OpenEnv grader format) ──────────────────────────────────── | |
| reward_rubrics: | |
| economic: | |
| weight: 0.25 | |
| description: "Economic health: penalizes hyperinflation even when GDP/employment look good." | |
| health: | |
| weight: 0.25 | |
| description: "Public health capacity vs active infection burden." | |
| social: | |
| weight: 0.20 | |
| description: "Public satisfaction adjusted for structural wealth inequality." | |
| sustainability: | |
| weight: 0.15 | |
| description: "Fiscal balance and resource conservation; penalizes borrowing from the future." | |
| crime: | |
| weight: 0.15 | |
| description: "Internal security — crime rate normalized with 2.5x sensitivity." | |
| # ── Metadata ────────────────────────────────────────────────────────────────── | |
| tags: | |
| - openenv | |
| - multi-agent | |
| - society-simulation | |
| - resource-management | |
| - reinforcement-learning | |
| - real-world-data | |
| - long-horizon | |
| - multi-objective | |
| authors: | |
| - name: CivicAI Team | |
| github: mahammadaftab | |
| hf_space: mahammadaftab/AI_Society_Simulator | |