What Is Memory that It May Have Hindsight and Foresight as well? - Understanding Understanding: Essays on Cybernetics and Cognition - page 121

isolated functionally: it is the interaction of all the processes here involved
that “life” the information from the input signal and translate it into action
meaningful for this tile.
Nevertheless, if forced to interpret some of this tile's functional compo-
nents in terms of those conceptual components I would reluctantly give the
following breakdown: (i) Perception is accomplished by the elements that
establish self-referential relations in the spatio-temporal configurations of
stimuli and responses; (ii) Memory is represented by the particular modus
operandi of the central computer whose gross functional organization is
determined and redetermined by evaluation of eigen-states or relations;
(iii) Inference in this tile appears on three levels, depending on the type of
functions that are in range F and on the type of processes one wishes to
focus on. Adductive inference is operative in the cumulative absorption of
comparisons of past external and internal experiences that give rise to the
functional organization of the central computer. Inductive or deductive
inferences are computed by the central system concurrently with any
new signal, the inferential mode being solely dependent on strings of earlier
failures or successes and of some of this tile's internal dispositions to “dis-
regard” false inductions or to take them “seriously” by converting to more
stringent logical deductions.
I shall now conclude my thesis with only a brief report on some proper-
ties as they may be relevant to this topic of aggregates of such tiles or
“tessellations” as they are usually referred to in the literature. John von
Neumann was the first to realize the high computational potential of
these structures in his studies of self-reproducing automata, 21 and later
Löfgren applied similar principles to the problem of self-repair. 22 We
use these, however, in connection with problems of self-reference and self-
representation.
Two features of cognitive tiles permit them to mate with other tiles: one
is its inconspicuous element T which translates into a universal “internal
language” whatever the “output language may be; the other one is its essen-
tial character as a “through-put” element. Consequently, one may assemble
these tiles into a tessellation as suggested in Figure 7, each cross, white or
black, corresponding to a single tile, while each square in a cross represents
the corresponding functional element as suggested in the earlier Figure 6.
Information exchange between tiles can take place on all interfaces,
however, under observance of transmission rules implicit in the flow
diagram of Figure 6. For instance, one tile may incorporate into its own
delay loop preprocessed information from an adjacent tile, but eigen-state
information of one tile cannot retroactively modify the operations of a
“left” tile, although it can—via its own output—modify that of a “right” tile,
and so on.
When in operation, this system shifts kaleido-scopically from one par-
ticular configuration of cooperating sets of adjacent tiles to other con-
figurations, in an ever changing dynamic mode, giving the impression of