Mystery Number Five - Computational Linguistics and Talking Robots - page 99

6.3.3 T WO SIMPLE R/E ONE - STEP INFERENCE CHAINS

R/E [rec: red square] K cm/exec [act: feed] K+1
R/E [rec: green circle] K cm/exec [act: hide] K+1
Automatic decoupling consists of disassembling the stimulus-response pairs
of the agent's fixed behavior into their elementary parts. For example, the dis-
assembling of 6.3.3 results in the following elements, each a separate and man-
ageable task of software and hardware engineering:

6.3.4 D ECOUPLED RECOGNITIONS AND ACTIONS

rec: red act: hide

rec: green act: feed

rec: square
rec: circle
Any execution of the fixed behavior inferences 6.3.3 is copied into the agent's
writable memory in the form of decoupled recognitions and actions as defined
in 6.3.4. Thereby, they receive incremented numerical prn values, resulting in
their reconnection in terms of their temporal sequence (coordination).
Let us assume that the combinations red square and green circle are
knowns, while the alternative combinations green square and red circle are
unknowns (cf. 3 in 6.3.1). When faced with such unknowns, two kinds of
actions are available to the agent, namely act: feed and act: hide . Without
additional assumptions, the choice between these options is random.
For appraisal (cf. 4 in 6.3.1), let us add two more decoupled internal recog-
nitions to 6.3.4, namely [rec: good] and [rec: bad] . When the agent is faced
with an unknown, e.g., red circle , there are the following possibilities:

6.3.5 P OSSIBLE CONSTELLATIONS WHEN FACED WITH AN UNKNOWN

1 rec: red circle act: hide rec: good
2 rec: red circle act: hide rec: bad
3 rec: red circle act: feed rec: good
4 rec: red circle act: feed rec: bad
For example, if the agent recognizes a red circle, chooses to hide, and evaluates
the decision as bad (because it interrupts feeding unnecessarily), as in 2 of
6.3.5, the next 17 encounter of a red circle will not be at random, based on the
following general inference rule:

17 In real life, agents usually do not change their behavior after the first negative experience. This makes
functional sense insofar as instances of the same behavior pattern may call forth different conse-
quences from the external environment. However, the worse the consequence of an initial random
behavior, the more likely the avoidance of that behavior in a second encounter.