We built an Artificial Brain that sleeps, dreams, and forms memories (github.com)

JN Research has unveiled Primite 1.03, an artificial cognitive system built from 1,800 Adaptrons—artificial neurons designed to mimic biological neuron dynamics such as subthreshold, graded, and action potentials. Unlike conventional machine learning models that rely on training and gradient descent, Primite operates on principled neuron-like rules with self-adaptation. This novel approach enables emergent higher cognitive functions including autonomous sleep states, dreaming, original thought generation, memory formation and recall across multiple timescales, and anticipation—where outputs precede corresponding inputs. This breakthrough is significant as it departs from data-driven AI toward a fundamentally different substrate for cognition, dubbed Adaptronics. Primite’s ability to autonomously cycle through sleep and dream states and internally generate thoughts without external stimuli challenges existing paradigms by demonstrating early forms of creativity and memory processes grounded in biologically inspired neuron interactions rather than statistical pattern recognition. The team supports their claims with raw, timestamped experimental data and offers live demos for independent verification, underscoring transparency and reproducibility. Primite 1.03 represents a promising step in scalable machine cognition by emphasizing adaptable, principle-based neural architectures rather than black-box optimization. This could reshape future AI research by inspiring systems that exhibit true self-directed cognitive behaviors and richer internal dynamics, bridging the gap between mechanistic neuroscience and traditional AI.