REPRESENTATION OF EVENTS IN NERVE NETS 5
The McCulloch-Pitts assumptions give a nerve net the character of
a digital automaton, as contrasted to an analog mechanism in the sense fa
miliar in connection with computing machines. Some physiological processes
of control seem to be analog. Just as in mathematics continuous processes
can be approximated by discrete ones, analog mechanisms can be approxjjnated
in their effect by digital ones. Nevertheless, the analog or partly analog
controls may for some purposes be the simplest and most economical.
An assumption of the present mathematical theory is that there
are no errors in the functioning of neurons. Of course this is unrealistic
both for living neurons and for the corresponding units of a mechanical auto
maton. It is the natural procedure, however, to begin with a theory of what
happens assuming no malfunctioning. Indeed in our theory we may represent
the occurrence of an event by the firing of a single neuron. Biologically
it is implausible that important information should be represented in an
organism in this way. But by suitable duplication and interlacing of cir
cuits, one could then expect to secure the same results with small probabil
ity of failure in nets con, trueted of fallible neurons.
Finally, we repe it that we are investigating McCulloch-Pitts nerve
nets only partly for theii own sake as providing a simplified model of nervous
activity, but also as an illustration of the general theory of automata, in
cluding robots, computing machines and the like. What a finite automaton
can and cannot do is thought to be of some mathematical interest intrinsic
ally, and may also contribute to better understanding of problems which arise
on the practical level.
PART I: NERVE NETS
3. McCulloch-Pitts Nerve Nets
Under the assumptions of McCulloch and Pitts [19^3 ], a nerve cell
or neuron consists of a body or soma, whence nerve fibers (axons) lead to
one or more endbulbs.
A nerve net is an arrangement of a finite number of neurons in
which each endbulb of any neuron is adjacent to (impinges on) the soma of
not more than one neuron (the same or another); the separating gap is a
synapse. Each endbulb is either excitatory or inhibitory (not both).
We call the neurons (zero or more) on which no endbulbs impinge
input neurons; the others, inner neurons.
At equally separated moments of time (which we take as the integers
on a time scale, the same for all neurons in a given net), each neuron of
the net is either firing or not firing (being quiet). For an input neuron,
the firing or non-firing at any moment t is determined by conditions out
side the net. One can suppose each is impinged on by a sensory receptor
organ, which under suitable conditions in the environment causes the neuron
to fire at time t. For an inner neuron, the condition for firing at time t