Biomedical Engineering Reference
In-Depth Information
motivations states, and they also exhibit the cor-
responding 3D animation sequence.
If motivation parameters, like thresholds or
motivation growth rate, vary, it may be possible
to obtain different emergent behaviors, simulat-
ing different personalities in an easy way. For
instance, a calm character can be simulated with
higher threshold than anxious ones, because it
will take more time for it to achieve the discharge
zone. Then, the probability of a frustration at-
titude decreases. Different
m
values may reflect
the various degrees of importance for motivations
between different characters. The choice of a
discharge action is another way to express varia-
tion among personalities. Aggressive characters
tend to express frustration by violent actions,
while more inhibited ones tend to do it by subtle
movements of impatience.
All this flexibility, though, is not computation-
ally complex as it was expected to be. Indeed, the
more complex part in a decision-making process
is that involving environment information search
and analysis, but the present architecture restricts
this processing through motivation filtering. Since
reactive agents (the simplest form of agents) are
responsible for motivation processing, the prun-
ing of the whole action selection computation is
significant. The hierarchical cognitive structure
that propagates activity through very specific
rules rapidly reaches the final decision, making
the processing even more efficient. Although
motivational graphs use symbolic rules, it seems
reasonable that they can be implemented by us-
ing other techniques, if they are more adequate
to certain specific tasks. As a future work,
motivation-oriented constructivist learning and
cultural influences can be included in the cogni-
tive structure.
behavior and cognition; (b) artificial mind models
and architectures; (c) application fields.
Artificial Intelligence has always been about
simulating human behavior and cognition through
computational models, as it is well known. But
until now only fragments of partial and isolated
cognitive functions and properties have been
actually implemented. But the development and
convergence of many technologies, along with the
progress in complex system formalism, cogni-
tive modeling, and many other multidisciplinary
subjects, is finally giving the opportunity to at-
tempt to integrate these multiples partial models
coming from many sources. Integration means
building an artificial mind capable to coordinate
and perform in a consistent and harmonious way
behaviors so different as feeling, reasoning, danc-
ing, eating, creating, etc. It is time to put together
the work done by AI researchers till now. It is not
an easy goal to pursue. But we have to remember
that the whole does not function the same way
as merely the sum of the parts. So, integrating
would be the actual and effective way of testing
partial models.
Models and architecture still have difficulties
in incorporating dynamics to the system. Although
psychological and pedagogical researches and
practices have shown the superiority of construc-
tivism over behaviorism, this has not been deeply
approached by AI models, yet. Also, until now,
it has not been possible to implement the way
people influence each other from the point of view
of their cognitive and knowledge structure and
reasoning processing. Cultural issues are still to
be approached by AI techniques.
Applications of Virtual Humans with an
integrated Artificial Mind would expand the
possibilities of complex simulations. There are
areas, like serious games, that have already cre-
ated education, training, cognitive ergonomics
research, or management applications, but they
have to unfold and cover much more aspects
than what has been done. Maybe the path to be
FUTURE RESEARCH DIRECTIONS
There are three main areas where future perspec-
tives can be discussed: (a) understanding of human
Search WWH ::
Custom Search