Computation in Neural Nets - Understanding Understanding: Essays on Cybernetics and Cognition - page 54

FIGURE 17. Primitive nerve nets: (a) independent effector; (b) receptor-effector
system; (c) receptor-computer-effector system.
are the foundation for the emergence of neural systems with the complex-
ity of a mammalian brain. Fig. 17a shows symbolically the “independent
effector” (muscle cell in a sponge) that translates directly a general “stim-
ulus” into action—contraction in most cases. The first step from detection
to discrimination is accomplished by separating detection and action and
localizing these functions in different elements (fig. 17b). This permits the
development of specific sensors responsive to certain stimuli only (light,
chemistry, touch, etc.). The final step in preparing the tripartite architectural
organization of the nervous system—detector, computer, effector—is sug-
gested in fig. 17c, where an intermediate ganglion cell acts as a primordial
nucleus for what is to become the information processing interface between
detection and action.
Although an array of such simply organized units as in fig. 17b appears
not to have the properties which we would expect from a neural net, for
there is no direct connection from neuron to neuron, these systems still
deserve to be called interaction nets from a general point of view, because
a particular state in one unit—say a contraction—may influence the state
of its neighbors via the mechanical properties of the medium in which they
are embedded. That such “mixed nets” are capable of highly organized
behavior may be illustrated with the beautiful observations of Kinosita
(1941) on the kinetics of the spines of the sea-urchin.
When a localized stimulus is given to the body surface of a sea-urchin,
the spines around the stimulated spot respond so as to lean towards the
stimulated spot, and this response diminishes rapidly with distance from the
locus of stimulation (fig. 18). The first thought that comes to mind is to
assume an anastomosing plexus of interacting nerve cells which transmit
the information of this perturbation over an appropriate region to cause
contraction of the muscle fibers attached to the spines (Üxküll, 1896).
However, Kinosita was able to demonstrate that there are no fiber to fiber
connections, only proximate fiber-muscle connections, hence each receptor
pair has to operate according to local information of the deformation of its
surroundings caused by deformations of the more distant regions.