Humans in Work System Environment - Fundamentals and Assessment Tools for Occupational Ergonomics - page 97

must be taken into account (Luczak and Schlick, 2000, 2001). First, it is required to model human infor-
mation processing in complex task settings. That means the model must be able to represent input-
output transformations, which significantly extend Sternberg's (1969) classical stimulus-response
approaches. Therefore, the theoretical construct of a “mental model” must be considered, because
tasks of production planning and diagnosis often require an appropriate reasoning space for the oper-
ator. Second, it is required to represent concurrent threads of reasoning and, therefore, cope with
aspects of mental resource allocation. Therefore, a symbiosis of stage- and resource-oriented models
of human information processing is preferred (Kahnemann, 1973). Third, there is a need to model differ-
ent time bases of human information processing, because some tasks, for example, process control, need
synchronous information processing, while other tasks, for example, production planning, are comple-
tely self-paced. Fourth, especially synchronous human information processing involves different sensory
modalities, which must be considered. Fifth, the autonomy of the manufacturing system requires to
model aspects of human learning and adaptation. Sixth, human errors play an important role for
performance prediction and evaluation of system reliability should be taken into account. Seventh, a
strict modeling formalism is required to anticipate time consumption of relevant cognitive functions.
Eighth, it is required to represent cognitive stress and strain, because equal performance levels may be
assessed differently with regard to “mental costs.” Ninth, it is required to integrate formal aspects of
human communication in terms of semiotic modeling, so that different levels of information exchange
between human and machine can be differentiated (physical, syntactical, semantical, pragmatical).
In addition to these application-driven requirements, two utility-driven requirements must be taken
into account: first, it is useful to have software tools for supporting cognitive analysis, modeling, and
evaluation. Second, widespread models are preferred, because they ease data collection and comparative
assessments.

5.5.5.2 Preset Catalog of Cognitive Models
It is necessary to distinguish between normative and explicative approaches of cognitive modeling.
Normative models provide an answer to the question “what should be?,” that means the model structure
is developed with a system of design goals in mind. These models usually stem from the domain of
cognitive engineering (Rasmussen, 1986) or the domain of naturalistic decision-making (Zsambok
and Klein, 1997).
The following normative models were considered:

1. Caccibue (1998): cognitive simulation model — COSIMO
2. Card et al. (1983): model human information processor, which is the basis for the well-known
approach of goals, operators, methods, and selection rules — GOMS (e.g., Kieras, 1997)
3. Rasmussen (1983, 1986): “model of skills, rules, knowledge, signals, signs, and symbols” — SRK

In contrast, explicative models provide an answer to the question “what is the cause?.” That means the
model structure is developed through inductive reasoning steps regarding cause-and-effect relationships.
Therefore, explicative models have their roots in cognitive science or artificial intelligence research
(Newell, 1990).
The following explicative models were considered:
1. Anderson (1993): ACT-R theory
2. Newell (1990, 1992): unified theories of cognition — UCT

5.5.5.3 Model Evaluation by Multi-Criteria Value Functions
The goal of further investigation was to select the model with the best goodness-of-fit concerning the
fulfillment level of the cited requirements. Therefore, a decision theoretic approach with multi-criteria
value functions was preferred (Eisenf¨hr and Weber, 1994). In a first step, the fulfillment level of the
requirements was ranked model-by-model using a three-level ordinal scale (requirement fully fulfilled

¼

value of 1, requirement fulfilled partially value of 0.7, requirement not fulfilled value of 0). In a

¼ ¼