It is impossible to believe that neuronal connectivity is fixed and
identical in every individual, Accidents of cell destruction would be inevitable,
even if we could believe that the life-long development of a brain was immutable.
The variation in the mere size of brain parts and cell number among individuals
tells us the same story. Therefore, variation in connection pattern must be
taken as an inevitable component of the developmental pattern of a brain. Since
this morphological development is concurrent with learning, it must play some
part in it; if nothing else, the memory mechanism must accommodate itself to the
immense variety of patterns of channéls. Even if there were some alternative
mechanism of information storage the operating behavior of the system would
surely depend on variations in the connection pattern, and to that extent, the
latter must inevitably play a substantial part in the behavioral development,
i.e. the memory of the organism. We have now only to ask whether this inter-
connection pattern is influenced by the life-long experience of the animal. It
is important to remember that the individual does not actually remember any part
of the real world. We.are entitied-te- infer only-that what is remembered is a —
pattern of effective behavior which, hopefully, may be an adaptive response to
past or anticipated sensory data, and which may include the regeneration of
simulations of past experience. A glance at a shelf of books may enable us
later to utter, be it silent or public, a fragmentary reconstruction of a few
bits of description of that bookshelf. No conceivable technique of information

storage would enable the books themselves to be accurately reproduced.+

1 Some remark along these lines may be absolutely essential, because I sometimes
feel that the people who insist on much larger dimensions of information storage
have completely missed some elementary aspects of the physiological psychology of
perception, I don't know whether I have stated it very well here.

J. Lederberg
1/14/65
If we adopt the second approach to this manuscript I would try to put
together a couple of paragraphs that would relate the problem of the develop-
ment of the interconnection system, since we have now placed the burden of
memory on this, to the general problem of development and how this would then
suggest the kind of analytical tools that would be most appropriate to further

progress.

J. Lederberg
1/14/65
One motive that has impelled speculation on more complex mechanisms than
neuronal interconnection has been the concern whether this could generate a high
enough level of information storage. With neither a precise quantitative estimate
of the requirements nor explicit model of the cerebral system, it is not easy to
pass judgment on their correspondence with one another. However, we can make the
following rough and ready calculations.

(1) A crude model of interconnection. Let a typical neurone have a domain
of 10,000 neighbors with which it can potentially interact. If,in fact, it

pat naa hoe re

establishes 100 synapses, the choice of audience ‘corresponds to about 800 bits

A
of information per neurone, i.e. about 8 x i0!? bits for the entire brain. The

calculation follows: each synapse represents a choice of one among 104 or 213-3
possibilities, i.e. carries 13.3 bits. However, the 100 synapses might be
arranged in 100! alternative combinations, corresponding to 525 bits (2°45 = 100!).
We have then 1330 ~ 525 = 805 bits as the information content of the unordered
set of synapses from a single neurone.

This calculation may be, if anything, conservative in its postulate of
the richness of synapses with which it will vary in proportion. The domain of
potential audience doubtless varies even more widely throughout the brain. The
information value of a neurone will go roughly as a logarithm of the size of
the domain.

(2) To estimate the information-storage requirement, we note that the
human sensory channels can hardly sustain a continued average rate of 50 bits
per second. A century is 3.15 x 10? seconds, giving a generous figure of
1.6 x 10!! bits of information input per lifetime, but still this is only a

fiftieth of the calculated capacity. On the most demanding assumptions of

retention and infallibility of memory this still leaves ample room for redundant
representation for storage of instinctual programs and for the computing
elements which must process this information. Further, it takes no account

of additional morphological dimensions of information, e.g. the length,
diameter and speed of each conductor, not to mention any specific qualities of

their terminations.

J. Lederberg

1/14/65
et

Concerning the Woodrow analogy, this is perfectly appropriate in the
context of this lecture. I don't think I would give quite so much currency
to those developments in a more studious presentation. The memistor
is of course an intriguing example of an analog device with memory, but if
only because it is an analog device, I think it is not the most appropriate
example to give a great deal of stress to. Im fact, it seems to me that a
latching relay with a tired spring, or subject to a noisy environment, would
be just as apt an example.

While the neurone must be regarded as a digital device, I suppose there
is an analog aspect in view of its asynchrony. We could very well imagine that
a good deal of information is enabled by variations in conductor length and speed
with consequent variations in the synchrony of firing of the synapses at the
target. This would then be added, of course, to the very familiar cell-to-cell
interrelationships in the nets. We certainly still need a clearer picture, and
I suspect it will be asimple one, of the way in which a bit of information is
actually stored, and more particularly the way in which this bit is appropriately
related to the sensory experience. From this point of view, I found £hak the
other way of looking at it, that is the question of how to store the effector
pattern makes the job appear somewhat simpler, but we still have the problem of
how to match an effector subroutine(e.g. a simulation of a past impression) to
the original sensation. This would obviously entail an enormous amount of compu-
tation, and it has been my hunch for a long time that sleep, and especially
dreaming, were the incidents of this process. If this is true, and if several
other elements of the general theory presented here are true, there should be
some definite biosynthetic correlates of sleep, which as far as I know, have
never been looked at and which we are now discussing as the next order of

business here.