Mystery Number Four - Computational Linguistics and Talking Robots - page 69

5. Mystery Number Four:
Autonomous Control
The fourth mystery is how to design autonomous control as an upscaling from
the cycle of communication by utilizing the conceptual and computational
constructs of DBS described so far. The overall goal of autonomous control
is to enable survival in the agent's ecological niche. For this, control must
guide behavior by connecting the agent's recognition to meaningful action,
including language interpretation (recognition) and production (action).
5.1 Pinball Machine Model of Cognition
According to Bernard (1865) and Wiener (1948), the task of autonomous con-
trol is to maintain the agent in a continuous state of balance (equilibrium,
homeostasis) 1 vis à vis constantly changing external and internal environ-
ments. DBS utilizes this task as the main motor driving the agent's cognitive
machine. To maintain balance, the agent's cognition continuously produces
blueprints for appropriate actions, with non-action as a limiting case.
It follows that a large part of autonomous control is retrieval. Blueprints for
action are derived by relating current content to past experiences stored in the
agent's memory. The search for suitable action models consists of (i) retrieving
contents which match the agent's current concern to various degrees and at
various levels of abstraction, and (ii) evaluating their outcome.
The better the available models fit the current task, the less the agent must
rely on trial and error. This applies also to behavior not directly related to
survival: human desires such as those for power, love, belonging, freedom,
and fun may be subsumed under the balance principle by treating them as part
of the agent's internal environment - like hunger.
While making relevant contents available in real-time is a necessary condi-
tion for maintaining the agent's balance, it is not sufficient. This may be shown
by the following comparison. The topics in a private library and the proplets in
1 In more recent research, Herrmann (2003) uses the balance principle as the difference between the
agent's ist-Zustand (as-is-state) and soll-Zustand (as-should-be-state) to trigger language production.