DETERMINATION OF NEURAL CONNECTIONS
215
in the limb-moving segments of the spinal cord. Studies on the functional
determination of these segments have revealed that the time when
thoracic segments grafted into the place of brachial segments are no
longer able to move the limb, coincides with loss of the capacity to produce a greater number of motoneurons (Straznicky and Székely, 1966).
The morphological character and the density of motoneurons (and probably those of the internuncials playing upon them), together with a
somatotopic localization, may create the anatomical basis for the organization of a cell assembly capable of emitting an output pattern necessary for limb movement, without attributing an instrumental role to
selective connections. An attempt to construct such a network will be
shown in a later paper.
This speculation requires the existence of constancy of the morphological characteristics of neurons situated within a given nerve center.
However, the enormous variety in morphological details of neurons is
contrary to any attempt to construct a neural network organized on
the basis of the morphological characteristics of components. In other
words, the neuron is morphologically unreliable in building up a network of specific function on a "morphological specificity" basis. This
directly leads to the problem, first raised by von Neumann (1956),
of constructing reliable networks out of nonreliable components. Making
use of the probabilistic logic introduced by von Neumann for such problems, McCulloch (1963) succeeded in showing that an error-free calculation of logical function can be attained useing only unreliable neurons,
if there are more than two. Calculation can detect the presence or absence of an output from the network, and the fallibility of McCulloch's
models lies in the uncontrollable changes of the threshold of the neurons.
In an anastomosing net with a large number of components, the fallibility can be surprisingly big and the function of the network is still
reliable. If the threshhold is brought to a fixed value, it can easily be
replaced by the amount of excitation or by the number of synapses.
McCulloch came to the conclusion that crude statistical specifications
can ensure error-free calculation despite gross perturbation of threshold,
of excitation, and even of local synapsis. It seems that "crude statistical
specifications" can, nevertheless, be ensured by the determination of
the morphological characters of neurons.
A tentative assumption, therefore, for the program of future studies
is that, in a given cell assembly composed of elements with welldefined shape and form, the morphological character of the components
may lead to a structural organization in some nerve center of characteristic function. If, with our increasing knowledge about the structural
215
in the limb-moving segments of the spinal cord. Studies on the functional
determination of these segments have revealed that the time when
thoracic segments grafted into the place of brachial segments are no
longer able to move the limb, coincides with loss of the capacity to produce a greater number of motoneurons (Straznicky and Székely, 1966).
The morphological character and the density of motoneurons (and probably those of the internuncials playing upon them), together with a
somatotopic localization, may create the anatomical basis for the organization of a cell assembly capable of emitting an output pattern necessary for limb movement, without attributing an instrumental role to
selective connections. An attempt to construct such a network will be
shown in a later paper.
This speculation requires the existence of constancy of the morphological characteristics of neurons situated within a given nerve center.
However, the enormous variety in morphological details of neurons is
contrary to any attempt to construct a neural network organized on
the basis of the morphological characteristics of components. In other
words, the neuron is morphologically unreliable in building up a network of specific function on a "morphological specificity" basis. This
directly leads to the problem, first raised by von Neumann (1956),
of constructing reliable networks out of nonreliable components. Making
use of the probabilistic logic introduced by von Neumann for such problems, McCulloch (1963) succeeded in showing that an error-free calculation of logical function can be attained useing only unreliable neurons,
if there are more than two. Calculation can detect the presence or absence of an output from the network, and the fallibility of McCulloch's
models lies in the uncontrollable changes of the threshold of the neurons.
In an anastomosing net with a large number of components, the fallibility can be surprisingly big and the function of the network is still
reliable. If the threshhold is brought to a fixed value, it can easily be
replaced by the amount of excitation or by the number of synapses.
McCulloch came to the conclusion that crude statistical specifications
can ensure error-free calculation despite gross perturbation of threshold,
of excitation, and even of local synapsis. It seems that "crude statistical
specifications" can, nevertheless, be ensured by the determination of
the morphological characters of neurons.
A tentative assumption, therefore, for the program of future studies
is that, in a given cell assembly composed of elements with welldefined shape and form, the morphological character of the components
may lead to a structural organization in some nerve center of characteristic function. If, with our increasing knowledge about the structural