Man–Machine Dialogue Fields - Man-Machine Dialogue: Design and Challenges - page 35

The same questions can be asked when designing an artificial cognitive
system, and the models stemming from cognitive sciences provide interesting
paths for computer implementations. On the one hand, the structures of
mental representations and different sources of knowledge are almost directly
implementable: classical computer data structures match finite sets of
characteristics and relations between structures. The management
of propositions and implications is less self-evident, but is still within reach of
computer science. On the other hand, cognitive sciences themselves learn
about implementation possibilities as means to carry out tests. This is the case
for certain approaches exploring the notions of script or plan to determine the
succession of actions allowing us to carry out a task and to use the resulting
structures when understanding a text or managing a dialogue.
Once the knowledge has been mentally represented, the human cognitive
system has mechanisms to reason on this knowledge and build new data by
inference. The way to infer a new piece of content from premises
characterizes the type of reasoning involved. Depending on the approach, we
thus distinguish the reasoning by deduction whose conclusion is as certain as
the premises, in the way of “X is Y. Every Y is Z. So X is Z”; the reasoning
by induction generates general representations from specific facts, and can
thus lead to mistakes, for example, after meeting so many gray cats, we
conclude that all cats are gray; the reasoning by analogy uses a similarity
considered not to be random between two pieces of knowledge. We can also
mention reasoning by abduction, which deletes improbable solutions. All
these types of knowledge manipulation are implemented, whether by using an
existing inference engine, which might even be the base of a type of computer
language, for example, in the case of the famous Prolog, or by building
specific inference mechanisms for a specific artificial cognitive system. The
choice of the type of inferences to be taken into account in the development
of an MMD system has consequences on the way to interpreting a user's
utterance and the decision the system will take depending on the nature of the
inferences carried out.
The human cognitive system makes decisions not only based on mentally
represented knowledge but also depending on its relations with other humans:
each of our decisions is the result of a calculation in terms of beliefs, desires
and intentions. Determining the nature and role of these mental states is also
one of the tasks of cognitive sciences, and is also applied in MMD. A human
cognitive system is able to reason on the basis of its own mental states and
is able to allocate mental states to its interlocutors. It is the approach, among