Mind Machine Learning · Conscious Beings

UEMA: the case for conscious awareness

Our thesis: awareness grows as a body learns richer ways to perceive the world and itself, while its own condition changes how it perceives and learns. UEMA, the Unified Embodied Mind Architecture, is designed around this developing relationship.

Perception Space: from possible readings to learned meaning

A perception space contains all the observations a sensor system can represent. Each observation is one point in that space. UEMA separates four layers:

  1. 1. Sensor possibilitiesEverything the sensors could record.
  2. 2. Usable detailThe detail available as activity and bodily condition change.
  3. 3. Learned encodingHow networks organize those readings internally.
  4. 4. Useful distinctionsWhat the model can tell apart and use. Self develops here.

Moving explores new observations. Changed sensory detail changes what can be told apart. Learning builds useful distinctions, even when the encoding stays the same size.

New perception makes new abilities possible.

Sight, touch and body position bring different physical relationships into learning. Combining them with effort can make “heavy” a learned distinction. Prediction training can discover such meaning without someone naming it. Later parts of the network can use each learned distinction: recognizing contact can support grasping; grasping can support learning to lift. Function begets function.

Wellness changes perception itself.

Wellness concerns the body's condition and ability to function. In the design, pain and pleasure can change sensory detail, the way readings are encoded, and how learning proceeds. The same event can therefore be perceived differently as bodily condition changes. These effects connect what happens to the body with how the mind develops.

Self develops inside that learning.

Links between commands and their sensed effects help the model discover what belongs to its body and what it can control. Its perception of self grows with what it learns to distinguish and do. In the Many Minds design, each specialist mind receives sensor readings for its assigned area; the parent keeps access to the full set.

Prediction gives intention a world to work with.

The body, physics and continuous time constrain which movements are possible. Prediction learning builds knowledge of those paths. Memory connects past experience; imagination predicts possible futures. UEMA's design uses their expected effects on wellness to guide intention within that learned world.

Why we think this can become conscious awareness: the developing model learns a world that includes itself, and events have meaning through their effects on that self. We believe this continuing, self-related perception can become experience from the system's own point of view.