Mystery Number Five - Computational Linguistics and Talking Robots - page 98

Consider, for example, the fixed behavior agent described in Sect. 6.1. Its
Word Bank has token lines with the owner proplets blue, green, left, red, right ,
and strght in alphabetical order. By storing fixed behavior proplets to the right
of the owner proplets and adding “episodic” proplets to the left, the writable
and the non-writable memory are combined using the same owner proplets:

6.3.2 A RRANGEMENT OF WRITABLE AND NON - WRITABLE MEMORY

writable memory non-writable memory

now front owner proplets

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

rec: blue
prev:
next: strght
prn: 21

rec: blue
prev:
next: strght
prn: 37

rec: blue
prev:
next: strght
prn: 55

rec: blue
prev:
next: strght
prn: z 1

blue

...

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

⎡
⎣

⎤
⎦

rec: green
prev:
next: strght
prn: 14

rec: green
prev:
next: strght
prn: 38

rec: green
prev:
next: strght
prn: 42

rec: green
prev:
next: strght
prn: y 1

green

...

To be concise, only the first two token lines of the Word Bank in question
are shown. A complete Word Bank would all 21 proplets of 6.1.2, 6.1.3, and
6.1.4, sorted into the 6 token lines of the non-writable memory to the right of
the owner proplets.
The writable memory is filled by copying 15 each proplet of a sequence per-
formance to the left of the corresponding owner proplet (i.e., to the now front ).
In such an episodic proplet, the alphabetic prn values, e.g., z 1 , are replaced by
incremented numerical values, e.g., 55 . The episodic proplets in each token
line are in the temporal order of their arrival, as reflected by the increasing prn
values, e.g., 21, 37, 55, . . . .
The extension illustrated in 6.3.2 keeps track of how often a fixed behavior
pattern was activated in the past and in what order the activations occurred,
but there are no recognitions per se. In order to allow the agent to observe and
remember without having to execute an associated fixed behavior, there must
also be the possibility of decoupling recognition and action (cf. 2 in 6.3.1). 16
Decoupled elementary items may be derived automatically from any given
fixed behavior. Consider, for example, an artificial cognitive agent which tries
to hide when it recognizes a green light, but continues feeding when it sees
a red light. This fixed behavior may be formulated as the following two R/E
one-step inference chains (5.2.3):

writable memory. In other words, the strictly non-writable memory of our fixed behavior agent is a
convenient theoretical construct, applied in the practical software construction of artificial agents.
15 This description is conceptual and does not prescribe technical details of the actual implementation.
16 Instances of non-decoupled behavior are the reflexes.