Information Technology Reference
In-Depth Information
The simplest situation will often be chosen, and safety constraints are there-
fore not always respected.
3.
Opportunistic
: the next action reflects the salient features of the current
context. Only a little planning or anticipation is involved, perhaps because
the context is not clearly understood by the agent or because the situation is
chaotic. Opportunistic control is a heuristic that is applied when the knowl-
edge mismatch is large, either due to inexperience, lack of formal knowledge,
or an unusual state of the environment.
4.
Scrambled
: the next action is in practice unpredictable or random. Such a
performance is typically the case when people act in panic, when cognition
is effectively paralyzed and there is accordingly little or no correspondence
between the situation and the actions.
3
Our Approach
Several researchers developed applications for modeling virtual operators reason-
ing [13] [3]. Only a few works propose to build systems incorporating behaviors
based on cognitive and error models [2] [14] [15]. But this generation of behaviors
is not founded on models issued from cognitive science, it is based on informatics
foundation.
We find an application of the CREAM model in El Kecha ıandDespreswork
[1]. They used it to know what errors the trainee is making and furthermore
what causes lead to these errors. Woods [16] developed an application, based on
Reason, reproducing a simplified functioning model of a central nuclear with a
cognitive simulator to reproduce human operator actions. They do not aim to
generate errors but to explain the errors done by a virtual operator controlled
by a real human.
To characterize the deviations that virtual operators may sometimes display,
we were inspired by the notion of borderline tolerated conditions of use (BTCU)
from studies in ergonomics and human reliability [9]. This notion highlights
the implicit individual and social regulations in the working environment which
may lead to compromises in the use of tools and the performance of tasks.
Some tasks are carried out only partially or not at all, because of a lack of
time due to compromises made between safety and production. This concept is
complementary to other elements linked to the individual, such as consciousness
of risks, the effects of tiredness, or temporal pressures on performance.
In our approach, we combine COCOM, CREAM and BTCU. Starting from
these models we propose new mechanisms and algorithms to simulate human
decisional processes and generate human behavior-based errors. A challenge is
to implement these models to produce the expected flexible, contextual and
erroneous behaviors in both normal and constrained working conditions.
We integrate these models into a more global agent cognitive architecture,
MASVERP
1
, based on Belief-Desire-Intention [12] in order to produce behaviors
in response to agent goals and perception of the environment. According to its
1
Multi-Agent System for Virtual Environment for Risk Prevention.
