The TEKnology

We can say that a high-level phenomenon is strongly emergent with respect to a low-level domain when truths concerning that phenomenon are not deducible even in principle from truths in the low-level domain. - David J. Chalmers

Fractal Neural Network TEKnology a completely different approach for the development of truly intelligent systems. Our Fractal Neural Network is an emergent computational architecture, not a deterministic one. Almost all other AI efforts, even those who attempt to model neurophysiology, have depended upon the human researchers to understand how each function of the brain works. The FnnTEK approach is to understand a much more basic building block and allow the higher functions to emerge from the complex non-linear interactions of those basic blocks.

Early Results - Our Motivation

Our initial work was in visual systems - we were attempting to create a visual system that utilized low-res inexpensive cameras to perform complex visual tasks. The hardware was a pair of 640x480 B&W cameras on independent pan / tilt mounts and a neural network with 105 neurons reading images from the cameras and connected to the pan tilt control mechanism. When the system was initially activated, the neural simulation moved the cameras randomly, which was expected. What was not expected, however, was that the system learned within a few hours to move the cameras in concert. A few hours after that, the cameras would track an object that moved through the visual field.

This last effect was an emergent property of the neural network simulation that was running - there was no programming that in any way that involved tracking objects in three dimensions, the system learned how to track those objects in three dimensions. Several years of investigation followed to determine why the tracking occurred, how to harness the effect, and how to generalize the effects into other problem domains.

Resources of Interest

Varieties of Emergence
David A. Chalmers
Mark A. Bedau

The Theory

Strong Emergence unwelcome guest in the discussion of emergent phenomenon (see the Bedau quote at the bottom of this page). Yet the one phenomena that is most often stated as a candidate (sometimes, the only candidate) for strong emergence is consciousness. FnnTEK's basic theory is that cognition (a strong subset of consciousness) is also a strongly emergent property of a sufficiently complex and non-linear interaction of neurons. Further, it is our contention that cognitive behavior necessarily emerges from such a complex system when certain minimal conditions are met with regards to the 'sufficient complexity' of the environment. Lastly, there is a real difference between 'cognition' and emulation of cognitive functions via constrained algorithmic approaches.

Sufficiently Complex Environment

For a system to develop emergent cognition, the environment of that system must be sufficiently complex. Obtaining some metric for exactly what 'sufficiently complex' means is one of the research goals of this project. From our observations thus far, it seems that the following are necessary:

  1. Senses - Input to the network
    1. Exteroception - external inputs related to the problem domain
    2. Interoception - Fast / Slow Pain and Pleasure
    3. Proprioception - Kinesthetic Sense of Body Position
  2. Musculature - Direct Control of the Environment
    1. External - Real World Manipulation Ability
    2. Internal - Perception Manipulation
  3. Neural Network - Memory, Learning, Starting Point for Feedback Loops
    1. Feedback - for a system to be 'cognitive' of the real world, there must be feedback from the world into the system so that the system has a chance to perceive the world and its actions therein.
  4. Cognition requires a "Mind Body Mapping"
    1. This Mapping is created during the growth of software neurons
    2. Such learning is necessary so that once the digital mind is interfaced to the real world it can learn:
      1. What is under its direct control
      2. What is available for indirect control
      3. What is not controllable
    3. This provides a means to learn 'internal' vs 'external' phenomena

The Software Neurons

In our early writings we referred to the software itself as a kind of neural simulation we stated that, "The neurons are simulated the cognitive response is real" - after examining that phrase again and again, a realization struck; our neurons were not simulations of biological neurons, they are neurons rendered in software. Since we used nature as a starting point for our work, it was easy to think of the software as a simulation of nature. But simulating nature was never the goal, the goal was to create an AI that could build a truly cognitive model of the world in which it operates. From day one, we have moved away from deterministic methods of creating that system to one of causing the system to emerge from much simpler elements. At this point in our development, the spatial organization of the neurons into a fractal structure, the creation of separate organelles that have defined functions within that structure, existence of pain / pleasure, motor control, and visual information processing centers, have all been reduced to simple interactions between the neurons and their environment.

Digital Mind Overview

  • 42.9 Million (3503) Neurons in the Cerebellar structure
  • 1.7 Million (352x350x4) Neurons in the Spinal cord / Nervous System
  • Three Neuron Types
    • Normal - fires neighbors on stimulus
    • Inhibitory - retards neighbors from firing on stimulus
    • Nociceptor - pain receptors
  • Structural Organelles
    • Cortex - learning and memory
    • Lateral Geniculate Nucleus - visual system inputs
    • Motor Cortex - coordination of musculature
    • Amygdala - fast pain response
  • Software to Real World Interfaces
    • Visual - input from time based 2D images in sequence (video or SOM output)
    • TCP/IP - muscles to set IP addresses / Port combinations open or closed
    • FFT - manipulate 2D FFT
    • Statistical - manipulate sets of statistics on a defined problem domain


  • Neuron - A FrANN Neuron is a data structure with numerous biologically inspired parameters (neuron transmitter level, firing potential, NT capacity, etc.)
  • Software determines the change of state of each neuron in the digital mind as quickly as possible using a GPU based architecture
  • Learning occurs as changes in the state information of each neuron