Mystery Number Five - Computational Linguistics and Talking Robots - page 116

In this way, an artificial agent may learn an open number of words by “contex-
tual definition” - where “contextual” means the (nonlanguage) context of use
(and not the language-level co-text).
Modeling the adaptability of cognition in DBS results in a substantial degree
of unpredictability in the artificial agent's behavior seen from the outside. 30
For example, (i) the contents in an agent's Word Bank are a rich source of far-
reaching personal associations which are inaccessible to an outside observer,
(ii) the agent's recent history with its various sources of stress may create
short-term sensitivities which are hard to predict, (iii) the more constant lean-
ings of the agent may bring about unexpected preferences, and (iv) the agent
may have unusual, individual strategies to realize those preferences.
Corresponding to this unpredictability of an artificial agent seen from the
outside is the agent's constant adjustment on the inside to maintain a state of
balance. Any parameter values deviating from the “desired” value are recorded
by the DBS system and used as triggers to compute compensatory actions. At
first, such reactive behavior may be hard to predict from the outside. Over
time, however, patterns will emerge in many areas of behavior. They are the
result of an agent-internal formation of schemata. A hierarchy of schemata
results in levels of abstraction, which each have their own set of inferences.
Individual learning by experience results in more and more predictable be-
havior. On the one hand, this is necessary as a process of “fitting in.” On the
other hand, there is the possibility of redundant habits. To interrupt the inter-
minable glide into fruitless routine, a regular, general cleanup of the system
may be installed to revise cognition into more productive behavior, based on
cost-benefit analysis. The continuous interaction between habitual behavior
and innovative behavior is facilitated by their uniform coding as sets of pro-
plets and their time-linear processing with LA-grammars.
Innovative behavior at various levels of abstraction is constantly demanded
from the cognitive agent due to the unpredictability of the external and in-
ternal environments. In DBS, innovative behavior is based on the automatic
derivation and application of inferences.
30 In humans, customs of politeness buffer the inherent uncertainty about another's state of mind. Po-
liteness must be realized also in the artificial agent with language. Given that politeness is ritualized
(fixed) behavior, this should not be too difficult.