This project builds an early prototype exploring artificial consciousness through a simplified brain-like agent architecture. The system models autonomous thought, motivation, prediction, and learning in a fully transparent grid-world/Minecraft-style environment using dopamine prediction error as the core learning mechanism.

The agent ("Player") navigates connected "Neurons" representing objects, states, or experiences — each with energy, dopamine reward, and hierarchy-of-needs values. The agent predicts paths, estimates future costs, chooses actions, experiences reward, and updates synapse strengths based on dopamine prediction error (better-than-expected strengthens, worse-than-expected weakens). Eligibility traces handle temporal credit assignment. Dopamine depletion and regeneration model habituation and novelty sensitivity.

The larger goal: test whether consciousness-like behavior can emerge from three ingredients: structured knowledge representation, an endless internal thought/impulse process, and a hierarchy of needs providing intrinsic motivation.

Project website: https://asi-one-pi.vercel.app/