What Is Memory that It May Have Hindsight and Foresight as well? - Understanding Understanding: Essays on Cybernetics and Cognition - page 118

three steps back in time, and so on, until we arrive at the “birth-date state”
Y 0 of the system:

( ) .

Y

= F X,X ,X ,X ,...,Y 0 ¢ ¢¢ ¢¢¢

The remarkable feature of this expression is that it clearly shows the depen-
dence of the present output of this system on the history of the previous
inputs, rather than on just its present input or to put this into more poetic
terms, this system's present actions depend on its past experiences.
Two features should be noted here. First, no storage of representations
of past events—save for those traveling through the delay loop—take place
here. Reference to the past is completely taken care of by the specific func-
tion F that is operative. F is, so to say, the “hypothesis” that predicts from
previous cases future actions. Physiologically, F is determined by the func-
tional organization of the cellular aggregate (A,A + ). Second, an external
observer who wishes to predict the behavior of this system in terms of its
input-output or stimulus-response pattern

= ()

YfX,

and who has no access to its internal structure may soon find to his dismay
that he is unable to determine the elusive function “f”, for after each experi-
mental session the system behaves differently unless—by lucky circum-
stances—he finds a repeated sequence of inputs that will given him—by the
very nature of that particular F—repeatedly a corresponding sequence of
outputs. In the former case, this experimenter will in disgust turn away with
the remark “unpredictable!”; in the latter case he will say in delight: “I
taught it something!” and may turn around to develop a theory of memory.
Although such recursive function elements exhibit some interesting
properties, in the restricted and isolated way in which I have discussed
them, they are as yet incapable of responding to OK-signals (+), HANDS-
OFF-signals (-), and any other signals that report eigen-states of affairs, or,
in general, to self-referential information. Assume such information were
available. The question is now to which of the elements in the three-element
recursive function computer must this information go in order to modify
its modus operandi in accordance with a desired eigen-state configuration?
It seems to me that the question already carries its own answer: if such
change is necessary at all, then the only effective way to modify the general
properties of this computer is to change its “hypothesis” by which it com-
putes future states from past experience, i.e., the recursive function F 1 which
was operative until this moment must be altered to become, say, F 2 , and
perhaps later F 3 ,F 4 and so on, in order to achieve the properties that are
commensurate with the system's eigen-state configurations. In other words,
F itself has to be treated as a variable, as an element in a range of functions
F(F), whose particular value F i is determined by the eigen-states. Physio-
logically this means that the recurrent fibers that carry self-referential infor-
mation are synapsing with cells in the (A,A + ) aggregate so as to change it