mapped to entailments in model B. In the case of cognitive systems, model A will
be abstract and stored in the mind or the body of the cognitive agent and system B
will be part of its surrouding reality.
We must bear in mind, however, that models may vary widey in terms of pur-
pose, detail, completitude, implementation, etc. A model will represent only those
object traits that are relevant for the purpose of the model and this representation
may be not only not explicit, but fully fused with the model exploitation mecha-
nism.
2.3 Relations with other traits
Principle 1 grounds some common conceptions about cognitive systems; obviously
the most important is the question of representation. A cognitive system —by defi-
nition of cognition— necessarily represents other systems. Even more, these repre-
sentations must have deep isomorphisms with the represented objects so the cogni-
tive system can exploit formal entailments in its models to compute entailments in
the modelled object in order to maximise the utility of the interaction (more on
this in section 3). Paraphrasing what Conant and Ashby clearly stated Conant
and Ashby (1970) —every good regulator must contain a model of the system it
is controlling— we can say that every well performing cognitive system must con-
tain a model of the objects it is interacting with.
Many other core issues of cognitive systems are addressed by Principle 1. Two
quite fashionable these days are the questions of situatedness —cognition is neces-
sarily interactive with an external world— and embodiment —the necessary separa-
tion of the agent body from the rest as defined by the interaction boundary. Both
are duly addressed by the modeling perspective of Principle 1 even when they are
not as necessarily crisp as they may appear to roboticists because the model can
obviosly represent uncertainty and vagueness, hence being able to handle even
blurred bodies and fuzzy situations.
2.4 On model generation
Other so-called cognitive traits are left out of this picture of cognitive systems.
Model-based —cognitive— systems need not necessarily be learning systems —
even while learning will be a very common procedure for model generation. A
cognitive system may operate using a static model —coming from any source— as
long as it is considered valid. i.e. as long as the modeling relation with the external
object still holds.
Obviously, from the consideration of how the cognitive system becomes cogni-
tive or maintains its cognitive capability learning becomes crucial. Somehow the
models must be put there, in the mind of the cognitive system. In general —not
ASLab.org / Principles for Consciousness / A-2007-011 v 1.0 Final