Summary: Putting It All Together - Foundations for Designing User-Centered Systems - page 384

14.3.2.3 Explicit Process Models: Model Human Processor
Card et al. ( 1980 , 1983 ) believed applying information processing psychology
should be based on task analysis, calculation, and approximation. Their Model
Human Processor (MHP) offers an early, simple, integrated description of psy-
chological knowledge about error-free human performance relevant to HCI and
system design. It was one of the first attempts to get away from the proliferation
of descriptions developed to account for different psychological observations, and
to provide a unified description of users. It started to create a quantitative
methodology including average times. The MHP was an oversimplification, but it
did provide a kind of prototype model that could be used as the basis for
discussions.
The MHP was made up of a set of memories and processors, together with a
set of principles, including the ''principles of operation,'' which describe how the
components functioned together. There were three interacting subsystems: the
perceptual system, the motor system, and the cognitive system, each of which
had their own memories and processors. These are shown schematically
in Fig. 14.2 . A detailed description of how the MHP would perform a task
is achieved using either a KLM analysis or a GOMS analysis, described in
Chap. 11 . Examples of MHP analyses are shown in Table 14.1 , and two more are
on the topic's web site.
Long term memory essentially does not decay (d = ?) and has infinite
capacity (l = ?). Working memory has a visual store and an auditory store.
Memories in both stores decay, but at different rates, and the sizes are different.
Working memory outside the stores has a capacity limitation. The perceptual
processor consists of sensors and associated buffer memories, the most important
being a Visual Image Store, and an Auditory Image Store to hold the output of
the sensory system while it is being symbolically coded. The cognitive pro-
cessor receives symbolically coded information from the sensory image store
into its working memory, and uses information previously stored in long-term
memory to decide how to respond. The motor processor carries out the selected
response.
Time predictions are generated by analyzing a task into the constituent oper-
ations that are executed by the subsystems. Then average times are associated with
these operations based on the selected band of performance: Fastman, Slowman,
and Middleman, which allows predictions to be made along the central and
extreme points of the behavioral continuum of fast to slow users.
The MHP assumes highly idealized behavior (e.g., a single strategy to solve a
task), and has trouble representing errors. The latter is important. Errors in text
editing, for example, have been shown to account for 35% of expert performance
time and 80% of the variability in that time (Landauer 1987 , p. 151). Although the
representations of the perceptual, cognitive, and motor subsystems were weak, the
MHP did demonstrate the feasibility of the general idea and inspired later work on
information processing cognitive architectures realized as computer programs,
such as Soar (Newell 1990 ) and ACT-R (Anderson 2007 ).