Nemobot: A Strategic Game Agent Platform Enabling AI to "Self-Program"

Nemobot is an interactive agent engineering environment based on large language models, supporting users to create, customize, and deploy game agents. This platform extends Shannon's game machine taxonomy into four actionable AI game paradigms. By combining crowdsourced learning with human creativity, it enables AI to achieve self-programming capabilities and explores the direction of general artificial intelligence (AGI).

In 1950, Shannon first classified chess-playing machines into five categories (brute-force, practical, adaptive, luck-based, ideal). More than seventy years later, the emergence of large language models has made it possible to re-examine this framework. The Nemobot project is the culmination of this exploration, representing a brand-new paradigm for game AI.

Nemobot defines four game agent paradigms:

1. Dictionary-based Strategy Compression: For games with clear rules and limited states (e.g., Tic-Tac-Toe), it compresses strategies using state-action mappings and leverages LLM few-shot learning to quickly adapt to new games;
2. Mathematical Reasoning for Strictly Solvable Games: For perfect-information games (e.g., simplified Chess), AI explicitly calculates optimal strategies and generates natural language explanations, addressing the traditional black-box problem;
3. Hybrid Intelligence for Heuristic Games: For complex strategy games (e.g., Texas Hold'em), it combines minimax algorithms with crowdsourced data to evolve a strategy ecosystem through human-machine collaboration;
4. Self-Iteration for Learning Games: Based on reinforcement learning with human feedback and self-criticism mechanisms, AI actively reflects on decisions to improve learning efficiency.

The Nemobot platform is open: users can customize agents, experiment with tool combinations, or fine-tune models via programming interfaces; tool-enhanced generation capabilities allow agents to call external resources (databases, computing tools, etc.), dynamically acquire information to expand cognitive boundaries, making it both a consumer tool and a research platform.

Nemobot's ultimate goal is AI self-programming: autonomously discovering new tactics in strategic games and adaptively adjusting behavioral logic in role-playing games. Its essence is to integrate crowdsourced learning and human creativity, making AI a creativity amplifier that absorbs community wisdom, transforms it into executable strategies, and drives innovation.

Nemobot adopts a modular design: each game category has an independent strategy engine but shares a unified interface and knowledge representation, supporting cross-game migration. Future applications include: developing intelligent tutoring systems in education (utilizing interpretability); creating adaptive AI opponents in the game industry; and serving as a testbed for self-improving agents in AI research.

Nemobot demonstrates the potential of large language models as the core engine of complex decision-making systems. By combining Shannon's taxonomy with modern AI technologies, it provides new answers to AI playing games. Its self-programming paradigm is an important step toward general artificial intelligence. When AI can autonomously improve strategies, integrate knowledge, and collaborate with humans, the era of intelligent agents will be one step closer.