AI World: A Small Step Towards Simulating AGI

My long-term vision, aligned with the Road to Free Open AGI, involves creating complex, simulated worlds – think the Matrix or Westworld – populated by genuinely intelligent AI agents. I believe that pushing AI to operate effectively in rich virtual environments is a crucial step towards developing AGI capable of understanding and acting in our own world.

The AI World prototype is an early experiment exploring this concept using simple web technologies (HTML, CSS, JavaScript). It’s a small 2D world demonstrating how AI agents might perceive, act, and communicate within a simulation.

What Can You Do In AI World?

This prototype focuses on the core mechanics needed for AI agents in a shared space:

• AI Agents with Basic Actions: Agents, powered by language models connected via api_providers.js, can perform actions based on commands. For example, in 'Agent' mode, you can tell an agent to Follow you, Unfollow, or Walk X,Y to move to specific coordinates. The AI determines the correct command, and the simulation executes it.
• Spatial Communication: Using the different chat tabs (managed by spatial_chat.js), you can communicate with agents based on proximity – 'Local' for nearby, 'Area' for defined zones, 'City' for everyone in the current map, or 'World'. Agents also use this system to perceive conversations relevant to them.
• Player Interaction: You can interact directly with agents or the environment. Right-clicking on an agent brings up options like 'Talk' (chat mode), 'Command' (agent mode), 'Toggle Follow', or changing their 'Appearance'. Right-clicking the ground allows adding new temporary agents or environment 'Areas'. Clicking your own avatar (top-left in chat) brings up UI options like toggling fullscreen chat. Clicking the AI/System avatar opens the main settings.
• Simple World Building & Persistence: Through the settings menu (accessed via the AI avatar), you can create different 'Cities'. Each city is a separate saved world (using Local Storage) where you can place agents, define named areas, and set the map size. This allows creating and switching between different simulation setups.

Why This Matters for the Road to AGI

Why simulate worlds? Because complexity breeds capability. Even this simple prototype demonstrates the crucial loop: AI perceiving a state, reasoning, and outputting actions (Walk, Follow, speech) that change the state. Creating environments where AI must navigate, interact, and communicate pushes the boundaries of current models.

Mastering virtual worlds, especially as we move towards richer 3D environments in engines like Unreal Engine 5 or Godot, provides transferable skills for real-world AI applications. This browser prototype, while basic, highlighted the need for dedicated game engines for greater complexity, but validated the core AI integration concepts.

Achieving truly intelligent agents also requires more than just current LLMs. We need better agent architectures and, crucially, powerful, open-source models – a core tenet of the Road to Free Open AGI.

Open Source Foundation

This experiment, including the core logic in ai-world.html and the supporting api_providers.js and spatial_chat.js files, is fully open source (MIT License). Sharing these steps openly is vital for collective progress towards safe and accessible AGI.

Feel free to explore the code, try the prototype, and share your thoughts or suggestions on the GitHub repository. Every experiment, shared openly, moves us a little closer to the goal.