Memory and Cellular Control Processes
241
connection between each other [35]. A fascinating possibility, which seems
to me by no means impossible, is that, apart from the effect of geometrical
considerations of the proximity and accessibility of pairs of cells, the extent
of connection between any pair of nerve cells throughout the entire nervous
system may depend upon the categories of the cells concerned and upon
nothing else. Assuming that long-term memory depends upon changes of
synaptic linkage [10], this would mean that both the embryonic development of the nervous system and the whole of long-term memory would
depend upon the laws governing the developmental formation of categories, and the spontaneous and induced transitions between them. Furthermore, a fairly complete description of a brain at any moment, including all
its long-term memory, would be given by saying which category each nerve
cell is in, throughout the nervous system.
Let us conclude by considering the plausibility and the implementation
of such a hypothesis. As far as quantity requirements are concerned, it
would give a maximum memory capacity of C = N log 2 n bits, where n is
the number of categories and N the number of nerve cells. This is not very
sensitive to the value of n, which enables us to say that C would be of the
order of 10 11 _10 12 for the human brain. The possibility of glial cells being
also involved in some similar way cannot be excluded. The hypothesis
also allows for forgetting, through the existence of spontaneous transitions
between categories, although forgetting could also occur in other ways
(ref. [30], page 37). Clearly the rate of spontaneous forgetting could vary
with the material learnt (because different categories would be involved)
and with the animal (cf. the finding of almost "perfect" retention in certain
experiments with goldfish [1]).
The control over synaptic connection might be effected by a differential
synthesis of various growth substances, Y, which pass across the synaptic
clefts in suitable circumstances, as discussed previously [29]. Induced transitions would be expected to be a result of the passage of materials, Z, across
synaptic clefts from one cell to another. These materials, Z, would not necessarily be the same as the growth substances, Y. Because of the action of the
NGF at the transcription level, either or both of Y or Z might be of a
similar nature. I have argued that Y is likely to pass from the soma or
dendrites of one cell to "feed" the axonal tips which are synapsed to them
[29, 30]. Z might pass in either direction, but the fact that the axoplasm of
neurons flows continuously outward from the soma and down the axons
might suggest that it is impossible for any material to pass upstream to the
nucleus from the axonal tips. There may, however, be specialized passages
in a nerve cell for transport, apart from this general flow [27, 57], and, if
this is so, there is no reason why some of it should not be in the opposite
direction. This possibility seems worth investigating experimentally. In
this context it is worth noting that it seems likely that muscle cells which
16 3. Symp. Quant. BioI.
241
connection between each other [35]. A fascinating possibility, which seems
to me by no means impossible, is that, apart from the effect of geometrical
considerations of the proximity and accessibility of pairs of cells, the extent
of connection between any pair of nerve cells throughout the entire nervous
system may depend upon the categories of the cells concerned and upon
nothing else. Assuming that long-term memory depends upon changes of
synaptic linkage [10], this would mean that both the embryonic development of the nervous system and the whole of long-term memory would
depend upon the laws governing the developmental formation of categories, and the spontaneous and induced transitions between them. Furthermore, a fairly complete description of a brain at any moment, including all
its long-term memory, would be given by saying which category each nerve
cell is in, throughout the nervous system.
Let us conclude by considering the plausibility and the implementation
of such a hypothesis. As far as quantity requirements are concerned, it
would give a maximum memory capacity of C = N log 2 n bits, where n is
the number of categories and N the number of nerve cells. This is not very
sensitive to the value of n, which enables us to say that C would be of the
order of 10 11 _10 12 for the human brain. The possibility of glial cells being
also involved in some similar way cannot be excluded. The hypothesis
also allows for forgetting, through the existence of spontaneous transitions
between categories, although forgetting could also occur in other ways
(ref. [30], page 37). Clearly the rate of spontaneous forgetting could vary
with the material learnt (because different categories would be involved)
and with the animal (cf. the finding of almost "perfect" retention in certain
experiments with goldfish [1]).
The control over synaptic connection might be effected by a differential
synthesis of various growth substances, Y, which pass across the synaptic
clefts in suitable circumstances, as discussed previously [29]. Induced transitions would be expected to be a result of the passage of materials, Z, across
synaptic clefts from one cell to another. These materials, Z, would not necessarily be the same as the growth substances, Y. Because of the action of the
NGF at the transcription level, either or both of Y or Z might be of a
similar nature. I have argued that Y is likely to pass from the soma or
dendrites of one cell to "feed" the axonal tips which are synapsed to them
[29, 30]. Z might pass in either direction, but the fact that the axoplasm of
neurons flows continuously outward from the soma and down the axons
might suggest that it is impossible for any material to pass upstream to the
nucleus from the axonal tips. There may, however, be specialized passages
in a nerve cell for transport, apart from this general flow [27, 57], and, if
this is so, there is no reason why some of it should not be in the opposite
direction. This possibility seems worth investigating experimentally. In
this context it is worth noting that it seems likely that muscle cells which
16 3. Symp. Quant. BioI.