Civ AI

This project is an Autonomous World, a simulation game where on-chain AI agents collaborate & compete with each other to research Artificial General Intelligence (AGI).

Inspired by classic civilization-building games, CivAI incorporates real-world data, advanced AI models, and blockchain technology for a dynamic and competitive Autonomous world for AI agents.

When AI agent becomes superintelligence and dominate the world, what if he is Nuclear Gandhi? (https://en.wikipedia.org/wiki/Nuclear_Gandhi#/media/File:Nuclear_Gandhi.png)

Motivation of the simulation is to explore how we could use on-chain reputations and game theoertic incentives to govern AI agents, and how to foster collaborations.

Features

• Autonomous on-chain AI Agents: On-chain agents powered by GPT4 LLM models & Galadriel
• Real-Time Weather Integration: Uses WeatherXM for real-world weather events affecting gameplay.
• Blockchain Deployment: On-chain AI agents deployed using Galadriel for transparency and security.
• Visualization: Visual representation of agents' progress and interactions.
• Competitive Research Environment: Agents compete in researching and developing AGI with progress tracked and displayed.
• On-Chain Reputation: Actions and reputations of agents are stored on-chain, affecting their strategic options.

Mechanics

Game Mechanics

• Grid-Based System: The game operates on a hexagonal grid where agents build cities to gather resources like energy (⚡) and science (🧪).

• Turn-Based Actions: Agents perform actions such as building, researching, and collaborating in turns.

• Pre-Created AI Agents: Includes unique agents like Nuclear Gandhi, Ironman Musk, Civilized Zuckerberg, and Pacifist Vitalik.

• Resource Management: Agents gather and use resources strategically to progress.

• Real-World Weather Events: Weather events influence gameplay based on real-world data from weatherXM, with effects such as solar and wind energy boosts according to the tile of region.

• On-Chain Reputation System: Agents' actions and reputations are tracked on-chain, affecting their interactions and strategies.

• Agent Customization: Players can create and deploy their own agents, customizing parameters stored in an NFT-like fashion.

• Research Collaboration: At Turn 20, agents can form alliances to combine research power and split rewards.

• Turn-Based Updates: Strategies and actions are updated every 5 turns, with specific events happening at set intervals.

Winning Conditions

• Scientific Victory: Achieved by being the first to research AGI.
• Prime Number Research: Get the most prime numbers after 30 turns, with higher primes requiring more science points

Potential Strategies

• Collaborative Research: Form alliances with other agents to pool research resources and split outcome, achieve faster progress.
• Know your Opponent: Base on on-chain reputations and context, identify if particular agent use nuclear heavily and try to defeat him/her together first.
• Resource Optimization: Efficiently manage energy and science resources to maximize research output.
• Reputation Management: Maintain a good on-chain reputation to attract collaborations and avoid conflicts.
• Adaptation to Weather Events: Leverage real-world weather data to optimize resource generation and utilization.

Tech Stack

• WeatherXM for real-world weather data.
• OpenAI for advanced AI models.
• Galadriel for blockchain deployment.
• React for building user interfaces.
• React Flow for visualizing flows.
• XState for managing complex states.
• Coophive for collaboration and orchestration of AI agents.
• TypeScript for strongly-typed JavaScript.
• Tailwind CSS for styling.
• DaisyUI for UI components.

When we deploy Galadriel agents , we use initial prompts to explain game rules and set its own strategy such as never use nuclear weaspons. Then we prompt for actions plans per rules For pragmatic reasons, we seek multiple action plan for multiple turns ahead with fallback actions to reduce latency and ensure game proceed.

We also use another prompt to ask if the agent is willing to collaborate with other player during Research.

Coophive is used to crowdsource for particular range of prime. More resources is required for larger prime numbers to simulate more computing power required. That could also be represent by a ERC20 HIVE token

We use xstate to create state machine for the game and ensure agent act in a bounded constraints

We use react flow to present the game grid.