AMMO
Home
  • AMMO v0.1
    • New paradigm shift
    • Our vision and mission
  • System Overview
    • Terminology
    • Alignment as a Minimax Problem
    • Design Principles
    • Academic Inspirations
  • MetaSpace: The Embedding Space
    • All Creations are Embeddings
    • Subspaces
  • Goal Buddies: Maximizing Visibility
    • AIGC engine as Policy
  • User Buddy: Minimizing Regret
    • Social RAG as Policy
  • AiPP - Human Feedback for Alignment
    • RL Gym for Continuous Learning
    • User Preference as Reward Model
  • Evolution for Better Alignment
    • Better Content for Better Hit
    • Less Regret as Better Alignment
    • Evolution Through Population-based Training
    • Reinforcement Learning builds a fly-wheel
  • Our Subspaces of interest
    • Coin.subspace: Fakers AI
    • Job.subspace
    • Edu.subspace
  • References
Powered by GitBook
On this page
  • AMMO aims to create a New Meta for Human and AI
  • A New Paradigm for AI-Human Symbiosis
  • Technical Foundation: Distributed MAS + RL GYM for Alignment
  • A Vision for Real-world Agentic Society: The Scaling Laws of Agents
  • Join the Journey

AMMO v0.1

Mission: Build the future of AI-Human connection.

NextNew paradigm shift

Last updated 2 months ago

AMMO aims to create a New Meta for Human and AI

A for Alignment

A New Paradigm for AI-Human Symbiosis

AMMO (Architectures for Massively Multiagent Online) is an ambitious initiative designed to empower the next generation of Multi-Agent Systems (MAS) for real-world exploration.

At its core, AMMO aims to bridge the gap between human needs and AI-driven solutions, enabling individuals to discover hyper-personalised opportunities, connections, and resources through collaborative interactions with their AI companions, or "User Buddies."

By creating extensive MARL (Multi-Agent Reinforcement Learning) arenas that integrate alignment mechanisms rooted in collective human values, AMMO envisions a world where humans and AI coexist, co-learn, co-create, and co-evolve, all while being guided by humanity's shared values.

Technical Foundation: Distributed MAS + RL GYM for Alignment

AMMO is conceptually built upon the foundations of Multi-Agent Systems (MAS) and Distributed "GYMs" for Reinforcement Learning with Human Feedback. It provides a unified framework where AI agents dynamically adapt to human preferences while encouraging exploration. These agents act as autonomous scouts, continuously mapping the latent spaces of human needs and opportunities.

As a "gym" for reinforcement learning, AMMO trains agents to optimise both individual satisfaction and collective flourishing. This ensures that every interaction leads users closer to their ideal matches—be it ideas, commodities, or solutions.

A Vision for Real-world Agentic Society: The Scaling Laws of Agents

The name AMMO draws inspiration from OpenAI’s Neural MMO, a pioneering environment for massively multi-agent learning, as well as the advancements in Neural MMO 2.0 (NeurIPS 2023). Over the past few years, the team has been thinking over the possibilities to transition MAS from academia or game environments to large-scale societal adoption by implementing the Scaling Laws of Agents. AMMO represents a tangible and exciting approach to making this a reality.

AMMO envisions a future where AI serves as both participant and companion, guiding humanity through a real world of infinite possibilities. This vision is not confined to digital realms—it strives to dissolve both physical and intellectual barriers, creating a transcendent ecosystem where creativity, knowledge, and value flow freely.

Join the Journey

AMMO is more than just infrastructure; it is a movement. We invite you to reimagine a world where AI and humanity thrive together, guided by values of fairness, exploration, and mutual respect. This whitepaper details how we can turn this vision into reality—one environment, one agentic community, one interaction, and one discovery at a time.

Page cover image