236
J. S. GRIFFITH
plausible on account of its greater economy and the lack of any evidence for
the first. Also, experiments on rates of information acquisition and transmission in man are consistent with it. With the first assumption, the human
memory capacity has been estimated as 10 16 -10 19 bits [28, 55] and with the
second, lOB_lOll bits [28, 46, 54].
These estimates are very approximate and depend upon assumptions
which can reasonably be questioned. Have they any value? I believe they
have, because they can be compared with theoretical calculations of capacity derived from various possible mechanisms for memory and used to
give some guidance as to the plausibility of those mechanisms. For example,
if memory is stored in nerve cells and their connections, then if the first
assumption is correct there are 10 6 -10 9 bits per nerve cell. This would
create insuperable difficulty for a theory which asserted that changes of
threshold were responsible for memory, and considerable difficulty for one
which assigned it to changes in synaptic connection. On the other hand,
with the second assumption, we need only 10- 2 -10 bits per cell, which
would be acceptable on either theory.
It has been suggested that we should introduce a unit of memory, to
be called the mnemon, which is to be the "minimal physical change in the
nervous system that encodes one memory" ([11], see also [58]). One should
beware of the assumption, implicit in this definition, that there is a minimal
unit. If memory involves the accumulation of some substance, X, or synaptic growth, there may be no minimum above the level of a single molecule,
which might be ridiculously small in comparison with the smallest amount
of storage normally achievable in practice. In this context, one should note
that in spite of its "quantized" sound, a bit is not necessarily the smallest
amount of information that can be stored [36].
c) Transfer Experiments and "RNA Theories of Memory"
Many experiments have been reported which claim transfer of learning
between animals (see, e.g., [33,49, SOD and some of these also claim support
for an "RNA theory of memory". We shall not enter here into the complicated question of the extent to which these claims are justified [9, 31,34],
but shall make a few theoretical comments on the problem of their possible
interpretation (see also [15, 28]).
We should replace vague talk about "RNA theories of memory" with
some definite hypothesis, probably one of the following three. First,
memory is stored in a coded form in the nucleotide sequence of speciallysynthesized RNA, which is not transcribed from the inherited DNA. There
is no evidence for this, and it must surely be regarded as highly improbable.
Second, the physical representation of memory is in the form of the quantities
of one or more types of mRNA present in each of the cells (neurons pro-
J. S. GRIFFITH
plausible on account of its greater economy and the lack of any evidence for
the first. Also, experiments on rates of information acquisition and transmission in man are consistent with it. With the first assumption, the human
memory capacity has been estimated as 10 16 -10 19 bits [28, 55] and with the
second, lOB_lOll bits [28, 46, 54].
These estimates are very approximate and depend upon assumptions
which can reasonably be questioned. Have they any value? I believe they
have, because they can be compared with theoretical calculations of capacity derived from various possible mechanisms for memory and used to
give some guidance as to the plausibility of those mechanisms. For example,
if memory is stored in nerve cells and their connections, then if the first
assumption is correct there are 10 6 -10 9 bits per nerve cell. This would
create insuperable difficulty for a theory which asserted that changes of
threshold were responsible for memory, and considerable difficulty for one
which assigned it to changes in synaptic connection. On the other hand,
with the second assumption, we need only 10- 2 -10 bits per cell, which
would be acceptable on either theory.
It has been suggested that we should introduce a unit of memory, to
be called the mnemon, which is to be the "minimal physical change in the
nervous system that encodes one memory" ([11], see also [58]). One should
beware of the assumption, implicit in this definition, that there is a minimal
unit. If memory involves the accumulation of some substance, X, or synaptic growth, there may be no minimum above the level of a single molecule,
which might be ridiculously small in comparison with the smallest amount
of storage normally achievable in practice. In this context, one should note
that in spite of its "quantized" sound, a bit is not necessarily the smallest
amount of information that can be stored [36].
c) Transfer Experiments and "RNA Theories of Memory"
Many experiments have been reported which claim transfer of learning
between animals (see, e.g., [33,49, SOD and some of these also claim support
for an "RNA theory of memory". We shall not enter here into the complicated question of the extent to which these claims are justified [9, 31,34],
but shall make a few theoretical comments on the problem of their possible
interpretation (see also [15, 28]).
We should replace vague talk about "RNA theories of memory" with
some definite hypothesis, probably one of the following three. First,
memory is stored in a coded form in the nucleotide sequence of speciallysynthesized RNA, which is not transcribed from the inherited DNA. There
is no evidence for this, and it must surely be regarded as highly improbable.
Second, the physical representation of memory is in the form of the quantities
of one or more types of mRNA present in each of the cells (neurons pro-