Context
Most artificial intelligence systems today are built around pattern prediction over large collections of human-generated text, images, code, or other data. These systems are useful and technically impressive, but their learning process is usually separated from the moment of use.
During normal interaction, a model's weights are fixed. It can describe curiosity, fatigue, hunger, or preference, but those words are not coupled to internal needs that change how the system acts.
Research direction
Neural Substrate explores a different research direction: adaptive behavior from stateful neural mechanisms inside an environment. The system is not a language model with a motivational layer added on top. It is a neural substrate model with internal state variables, energy constraints, sleep dynamics, spatial input, and environmental consequences.
The research question is narrow: can a biologically inspired neural system accumulate useful state through continued interaction with an environment, while preserving dynamics and structure across episodes?
Observed behavior
Early observations suggest that internal state materially changes behavior. When the system is in a low-hunger state, it may ignore nearby resources that it locates quickly when hungry.
In one run, a resource found in under 500 steps during a hungry state took more than 21,000 steps when the system was satiated. The important point is not the exact ratio. The important point is that the same environmental opportunity produced different behavior because internal state changed what the system attended to.
Persistence
Over longer runs, the system also develops persistent spatial tendencies. It returns to regions where previous experience made resources more likely or more salient. Those tendencies are not represented as a hand-written route plan. They emerge from updates to the circuits involved in navigation, memory, and action selection.
Sleep and replay are also part of the current research surface. The system can resume after a rest cycle with altered internal state, which makes consolidation itself part of the behavior being studied rather than a separate offline training step.
Claim boundary
This work does not claim consciousness, sentience, or animal equivalence. The claim is more limited and more testable: behavior changes because internal states, neural dynamics, memory traces, and environmental interaction are part of the same operating system.
That distinction matters. It moves the question away from whether a model can produce plausible language about experience, and toward whether a system can develop useful behavior through persistent interaction, internal constraint, and structural change.