AGI experiments

A five-year personal research sandbox approaching AGI top-down: start with the simplest possible "minimal" agent, then iterate capability outward. Each experiment line is a hypothesis about what computational substrate might support general intelligence - predictive coding, free-energy minimization, quantized distribution learning, attractor dynamics, active dendrite models, sparse spiking ensembles, self-organizing maps. The structure is intentionally promiscuous: numbered notebooks (01.ipynb, 02.ipynb, …) track seed-and-fork iterations within each line, year-stamped folders capture later organized sprints, and notebooks run standalone rather than living inside a package.

Implemented in Python/PyTorch, with Weights & Biases instrumentation on the later runs. Specific lines stand on their own as cite-worthy investigations: Quantized Distribution Learning, Attractor Learning, domain quantization, EngramSOM, the Active Synapse work, and a standalone Pattern Machine implementation. Paused at the end of 2024 as the most promising threads migrated into focused projects - Pattern Machine, the carapace-intelligence series, and the fly connectome simulation - leaving this repo as the seedbed they grew from rather than an active project.