Using Misunderstanding and Discussion in Dialog as a Knowledge Acquisition or Enhancement Procecss - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 77

Data here may be facts, implications or equivalences.
So the first teacher's goal is to teach a lesson and the first student's goal is to learn this lesson. In
our course position, this goal is essential for retaining the taught lesson. This is the commonest and the
most natural goal, the one that seems to appear most frequently.
Student's initiative. In this second case we face another type of learning scheme we'll designate by
'the curious student case' as we have here an enterprising entity, who has its own motivations for getting
new knowledge from the 'teacher'. Indeed, our agents can be curious, as natural agents are sometimes,
and can wish to deepen their learning beyond the basic lesson.
We have identified three student's main goals for this case:
2. Increasing the KB connexity
3. Widening the predicates base
4. Understanding why some formulas imply others
Human goals vs. artificial goals. Each of these four goals (1-4) may just be considered in terms of
adjunction of new formulas, however we have defined them accordingly to human-like learning goals,
in order to be as close as possible to a natural situation. Each goal respectively corresponds to the ac-
complishment of the following human-like goals:
1. Teaching/learning pieces of knowledge through a lesson
2. Making more links between pieces of knowledge the 'student' already knows
3. Knowing new entities which check properties the 'student' already knows
4. Understanding why some properties are deductible from others, i.e. is there another way to under-
stand them?
KB connexity preservation goal. Each of these goals should fulfill a common goal: Preserving the
connexity property of the KB. This is a student's typical goal. We estimate this goal important as it aims
at preserving the KB consistency, in a human meaning: A human cannot use a new piece of knowledge
that is not linked to his/her own ones, so when s/he faces such a new data, s/he'll usually try to get data
that makes the junction between the new one and the already known.
For example, a human 'student' who ears about the concept of rectangle for the first time cannot do
anything with it and so should ask the 'teacher' what a rectangle is. Let us assume the 'student' knows
what a quadrilateral is and what the properties 'diagonals intersect in their middle' and 'diagonals length
are equals' means. The 'teacher' could then explain a rectangle by a quadrilateral which fulfills these
two properties. After this explanation, the rectangle should make sense to the 'student' as s/he can
understand this concept with his/her own ones: There is a link between them. We'll detail this example
in the next section.
In other words, by understanding , we mean 'not increasing the KB components number': The 'stu-
dent' understands a data that is linked to at least one component of its KB. By definition, we consider
that a 'student' agent knows a formula if it owns it. If the taught data is not linked to any component,
the 'student' have to inform the 'teacher' of its misunderstanding as we'll see in the next section.
KB consistency preservation goal. Here is another student's typical common goal. The 'student'
has to keep its KB consistent in order to prevent reasoning conflicts. For example, it cannot own the
two next formulas at the same time: ( P → Q ) and ( P →¬( Q )). When learning new data, it will then have