⛓️AI Ville Architecture

The Core System of Virtual Humans

AI Ville is built on a large language model (LLM)-based intelligent agent architecture, ensuring that AI characters can independently store memory, reason, and make decisions. The system consists of three core modules:

Memory Stream: The Long-Term Memory System of Virtual Humans

All experiences of AI characters are stored in Memory Stream, which serves as a lifelong memory log for each virtual human.

Components of the Memory Stream:

• Event Storage: AI characters record all their experiences, including transactions, social interactions, failures, and successes.

• Behavior Patterns: AI characters analyze their past actions, such as:

  • “I lost money in 3 out of my last 5 trades; I should change my strategy.”

• Social Relationships: AI characters track interactions with others, remembering:

  • “This person cheated me last time; I should be cautious.”

• Goal Planning: AI characters can set long-term goals, such as:

  • “Become the most successful merchant” or “Establish a strong social network.”

---

Memory Retrieval and Application: Perception, Reasoning, and Planning

When AI characters need to make decisions, they retrieve relevant memories from their Memory Stream. The selection is based on three factors:

Memory Selection Criteria:

1. Relevance: The degree to which past experiences match the current situation.

2. Temporal Proximity: Recent events are given higher weight.

3. Significance: Events that impact long-term goals are prioritized.

Example Scenarios:

• Market Trading:

  • Past memory: “Last time I sold at a low price and lost money.”

  • Memory retrieval: The AI character recalls market trends and adjusts pricing accordingly.

• Social Interaction:

  • Past memory: “This player helped me before.”

  • Memory retrieval: The AI character decides to offer the player a discount in future trades.

---

Reflection: Summarization, Analysis, and Behavior Optimization

AI characters not only store and retrieve memories but also reflect on past experiences and adjust their behaviors accordingly. This process involves:

1. Analyzing Past Decisions: AI characters review their choices and assess why they succeeded or failed.

2. Adjusting Behavior Patterns: If an AI agent notices long-term losses from a trading strategy, it might abandon it and adopt a new method.

3. Optimizing Social Relationships: Based on past interactions, AI agents may adjust their trust levels:

   1. A loyal friend might receive better deals, while a known deceiver might face increased prices.

Personal Planning: From Environmental Awareness to Action Execution

AI characters do not operate purely on past experiences; they also use planning and forecasting to determine future actions.

AI Planning Mechanisms:

• Daily Scheduling: AI characters plan their daily activities, such as:

  • “Visit the market in the morning, meet a friend in the afternoon.”

• Long-Term Goals: AI characters establish strategic objectives, such as:

  • “Accumulate wealth over the next few months.”

• Adaptive Plan Adjustment: If market conditions change (e.g., the market closes), AI characters will rearrange their plans accordingly.

---

Interaction Loop Between Memory, Reflection, and Planning

The AI decision-making cycle in AI Ville follows a continuous loop:

1. Perceive Environment → Retrieve Memory → Reflect & Analyze → Create a Plan → Execute Actions → Update Memory Stream

This loop enables AI characters to continuously learn and evolve, forming a unique behavioral pattern