Summary: Putting It All Together - Foundations for Designing User-Centered Systems - page 381

of view, a user model captures the capabilities and limitations on user perfor-
mance; from the software engineer's point of view, it would be used to illustrate
how the system could perform when operated by real users.
Models of users therefore serve as a summary repository of our knowledge of
users. It is important that this knowledge be captured in one place because it can
lead to emergent behaviors where there are interactions between the different
characteristics of users. The behavior of the models should be constrained in the
same sorts of ways that human behavior is constrained (memory limitations, and
so on).
We have seen throughout this topic that humans are less predictable, consistent,
and deterministic than computers. Defining a general, formal model of the human
(as part of the broader socio-technical system) is not currently possible. Instead, a
number of fragmentary and incomplete models of human information processing
behavior have been proposed by cognitive psychologists and reinforced by
empirical exploration. These models can be used to make predictions but do not
provide details about how the system should be designed. Although the models are
good at generating first-order effects, at lower levels of analysis their limitations
and inconsistencies become apparent. The models are therefore useful at pre-
dicting gross behavior, such as error-free, expert behavior on unit tasks.
The most accurate, detailed models have generally been developed for those
aspects of human performance that are easiest to test. Thus, models of the char-
acteristics of the senses are well established (particularly vision and hearing),
whereas there are very few models of some of the more intricate aspects of
cognition that can only be indirectly observed and are less well understood.
14.3.1 Unified Theories of Cognition
There have been several attempts to integrate all that we know about human
behavior into a single, unified theory of cognition (Newell 1990 ). The latest
attempts have been realized as cognitive architectures, although they might be
more properly described as human performance architectures because they deal
with more than just the cognitive aspects of human performance. These cognitive
architectures typically take the form of a computer programming language with
special capabilities and limitations representing a (small) subset of the known
capabilities and limitations of humans. Progress has been relatively slow, but the
cognitive architecture developers are working towards providing a single coherent
theory that ultimately accounts for—and predicts—human performance on all
tasks.
A unified theory of cognition effectively forms a single repository of useful
user-related information. This can be used in three main ways:
1. To help you remember theories and facts about users. You can use the schematic
of the Model Human Processor (MHP, described below), ACT-R (also described