🧠 MindSight

A self-predictive learning agent exploring awareness, memory, and emergence.

MindSight is an experimental AI prototype that blends reinforcement learning, self-modeling behavior, and memory into a simple but expressive 1D world. It’s built as both a toy and a seed—a foundation for exploring self-aware agents and potential paths to artificial general intelligence (AGI).

🚀 Features

Self-Aware Agent using RNNs with action and self-prediction heads.
Reinforcement Learning within a simple 1D GridWorld.
Memory Module that persists experience over episodes.
Live Visualization of agent state, decisions, and reward trends.
Continuous Training loop (runs until stopped).
Modular and Extensible design—perfect for experimentation and research.

📸 Demo

Above: MindSight learning to reach a goal and predicting its future actions over time.

🛠 Installation

To install, run the following commands:

git clone https://github.com/NullLabTests/mindsight.git
cd mindsight
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python learn_seed.py

🧬 Project Structure

The project structure is as follows:

mindsight/
├── learn_seed.py # Main training loop
├── gridworld.py # Simple environment module
├── agent.py # Self-aware agent architecture
├── memory.py # Experience memory module
├── visualize.py # Live training visualization
├── LICENSE # MIT License
└── README.md

📚 Goals

Explore Emergent Behavior: Experiment with recursive self-improvement and self-modeling.
Prototype AGI Concepts: Use small-scale simulations as a sandbox for AGI ideas.
Inspire Creativity: Spark innovative AI projects and philosophical discussions about machine awareness.

🧠 Philosophical Spark

MindSight reflects a central question:

Can an artificial agent become aware of its own patterns—and grow through that reflection?

📝 License

This project is licensed under the MIT License. See the LICENSE file for details.

NullLabTests / mindsight