From Pointing to Pondering
Stuart K. Card and Thomas P. Moran
Xerox Palo Alto Research Center
technical understanding of the user
himself and of the nature of human-computer interaction
- the scientific base of user technology, is necessary in order to
- understand why interaction techniques are (or are not) successful,
- to help us invent new techniques,
- and to pave the way for machines that aid humans in performing significant intellectual tasks
Our focus is on what
we have learned about users in our years of studying them
and how we see our findings relating to the original visions
of the personal workstation.
- The Vision of an Applied User Psychology
1974 PARC
Xerox’s idea was to draw concepts from cognitive psychology and artificial intelligence to create an applied cognitive science of the user
Applied Information-processing Psychology Project (AIP)
Worked with Allen Newell
Substantial payoff to designing systems with detailed understanding of the way the human must process the information attendant thereto
- Emerging psychology of cognitive behavior
- Central activities within scope of emerging theory
- Computer sci is one-sided in treatment of phenomena (no effort in understanding user)
HCI: involving two active agents each capable of initiating an exchange of information involves human cognitive processing
Wanted to create science of the user rooted in cognitive theory and have practical application (conceptual tools to think abt key characteristics of user and calculational tools to take account of user’s behavior)
Abbreviated history
The cognitive interface: The user has certain
characteristics as an information processor,
such as a limited working memory, that
together with the goals he is trying to achieve
determine his behavior.
The conceptual #interface: The computer
system is also a complex information
processing mechanism, and the the user
needs to have some kind of mental model of
it in order to effectively interact with it.
The task interface." Systems are designed to
help their users do tasks, not only small,
routine tasks, but also larger, difficult
intellectual tasks that are the object of the
grand visions of personal workstations.
- The Physical Interface: Pointing
we wanted to
understand the reasons why the results came out the way
they did. We therefore made mathematical models of each
device and tested them against the data until we found
models that fit. The model for the mouse was particularly
instructive. The mouse was best modeled by a version of
Fitts's Law:
Movement Time = Constant + .1 log2(D/S + .05) sec, where D is the distance the hand moves to the target and S
is the target width
this is the same law that describes movement
time for the hand alone, with about the same constant of
proportionality
the limiting factor in moving the mouse is not in the mouse, but in the eye-hand coordinate system itself
wanted to build theory:
model precise enough to let designers perform back-of-the-envelooe calculations and identified key constraints in the design space for pointing devices
- The Cognitive Interface: Cognitive Skill
The Model Human Processor
An interesting result of the mouse study was the way
an evaluation of pointing devices led to a consideration of
human information processing characteristics
1982 Model Human Processor inspired by the processors, memories, switches (PMS) notation of Bell and Newell for describing the architecture of the user
- three processors
- four memories
- 19 parameters of these
- 10 principles of operation
preserving the illusion of causality
MHP can be used to compute predictions about human performance
Text Editing and Cognitive Skill
MHP also contains Herbert Simon’s bounded rationality principle:
A person acts so as to attain his goals through
rational action, given the structure of the task; his inputs of information, and boutided by limitations
upon his knowledge and processing ability. (Card,
Moran, & Newell, 1983).
behavior
could be modeled as a search through a space of states of
the problem, a problem space
assumption: we would find users
searching through problem spaces to accomplish goals,
trying various solutions, backing off and taking other tacks
when they ran into trouble.
Two early decisions:
Although we had come
to PARC initially with the intention of studying computer
programming, we decided once we arrived that there were
strategic advantages to studying text editing. A second,
tactical, decision was to work with expert subjects rather
than novices, in order to have more stable behavior to
analyze.
This wasn't problem solving; we came to call it cognitive
skill and set out to build models to characterize and predict
this mode of behavior.
The result was a class of models in which
the user's cognitive structure consists of four components:
(1) a set of familiar Goals that the user would recognize
when faced with a specific task; (2) a set of primitive
Operators (actions) that the user was skilled at performing
and could deploy whenever necessary; (3) a set of Methods,
consisting of "compiled" sequences of subgoals and
operators, that the user could use to attain his goals; and (4)
a set of Selection rules that enable the user to choose among
competing methods for goals. We call a model specified by
these components a GOMS model
together, these components constitute the user’s cognitive skills for performing tasks (no need for problem-solving strategies)
practical question
related to the applicability of our models was whether the
additional work of constructing a fine-grained model was
worth the effort
We were
surprised to discover that fine-grained modcls did not yield
a worthwhile or even a significant increase in prediction
accuracy
even
the crude models seemed to capture and predict behavior
fairly well. These properties suggested that the GOMS
model could be turned into the kind of engineering tool
that a designer could use to model and predict skilled user
behavior in computer-mediated tasks.
The Keystroke-Level Model
idealized prediction problem:
Given a task (possibly involving several
subtasks), lhe command language of a system,
the motor skill parameters of the user, the
response time parameters of the system, and
the method used for the task, predict how long
an ~xpert user will take to execute the task
using the system, providing he uses the method
without error.
Each action of the method is
described in terms of the operators of the model, then a
time for the method is computed.
understanding how well the schemes would work for
expert users (which most users would be for most of their
time on the system) was expensive, because a long time
would have to be spent training the users. The solution was
to run experiments for novices and to use the
Keystroke-Level Model for predicting expert performance.
Each think-execute chunk of behavior is a unit task.
The unit task structure of cognitive skill is interesting
because the performance limitations of the user show
through the purely rational organization of his behavior.
limitations in the user's
working memory
If the information in
working memory should reach a level higher than the
working memory capacity, then user performance will
stiffer, usually manifested by the user committing errors.
Unit-task behavior is at present a theoretical notion
based on empirical observation
“cognitive rhythm” important feature of user behavior
The Psychology of HCI
Calculational models: relating these to the classical literature in cognitive
psychology and human factors
the Model Human Processor captured the relevant psychological
literature in the terms of a unified, approximative model.
This model also provided a foundation for our other
Models
The other big missing piece for us was to understand
the relationship between the cognitive skill we had
discovered and the classical notion of problem solving in
cognitive psychology.
Here we built a theory of the
behavioral continuum between problem solving and
cognitive skill and showed how practice on a task would
gradually chan~ge problem solving behavior into skilled
behavior. This is all active area of research in cognitive
psychology today
- The Conceptual Interface: Mental Models
expert users: procedural knowledge
the expert
users have some sort of mental model of what is happening
inside the computer Chow-it-works" knowledge). The “user’s model” conceptual model the user can have of the system
user’s conceptual model is distinct from but related to deisgner’s implementation model
The user's model provides an integrated package of
knowledge that allows the user to predict what the system
will do if certain commands are executed, to predict the
state of the system after the commands have been executed,
to plan methods for novel tasks, and to deal with odd error
situations (by characterizing the system's state according to
the model, then choosing operations necessary to leave that
state).
Early Encounters with Conceptual Models
It is clear that users attempt to make sense--by
building mental models--of the behavior of a system as they use it. If a simple model is not explicitly or implicitly
provided, users formulate their own myths about how the
system works.
if
the user is to understand the system, the system has to be
designed with an explicit conceptual model that is easy
enough for the user to learn. We call this the intended user's
model
User Interface Design Methodology
PARC ,and the Systems Development Division (SDD),
which was created to develop office system products based
on the research at PARC
The committee decided not to try to design an actual
interface, but to propose a methodology
(1) analyze what tasks
the user will want to do and the steps they go through to
accomplish the tasks;
(2) design an intended user's model in
terms of which the tasks may be cast; (
3) design a command
language to make that model work; and
(4) design an
information display to reflect the operations of the system
in terms of the conceptual model.
we recommended that the designer should lay
out an intended user's model before designing the
command language and the information display
Empirical Studies
The goal was to see whether we could find some
kind of mental models buried in the user's knowledge
no difference
between the two groups in both the routine and complex
problems, but the model group performed much better on
the invention tasks. The most surprising result was that
even some of the no-model group were able to perform
some of the invention tasks
in the
routine and complex problems, the behavior was almost all
skilled method execution: the subjects had been taught
what to do, and they did it; even for the complex problems
it wasn't difficult to knit together the methods they had
learned for solving the problems.
We found that the most critical (although not very time
consuming) problem solving was in the task space, where
the user analyzed the given task into subtasks and delegated
them to the model space or the methods space. The main
difference between the users who had a model and those
who didn't was that they had different problem spaces in
which to work
The model space was an
effective problem space, within which the solution to the
invention problems could be found; the method space was
not particularly effective, but it was sufficient to allow some
non-model users to stumble onto solutions to some of the
invention problems (often, much to their surprise).
Model-based
problem solving appears to be very mentally intensive, so
users avoid it if they can apply cognitive skills. But, if users
don't have appropriate methods available, they will retreat
to some kind of problem solving. In these cases, a good
conceptual model provides an effective problem space in
which to work.
Thus, system designers should think of a conceptual
model of a system as not just a simple view of a complex
system, but as a problem space through which users can
search for solutions to a variety of novel problems
the intended user’s model should be closely related to the kinds of tasks the
users arc likely to do, and the users should be provided with
heuristics for moving through the model space.
Theoretical Studies: Task Mapping
theoretical analysis of conceptual
models to show where they fit into the overall structure of
the user interface
The Command Language Grammar
formalism (Moran) shows how models relate to
the task domain, the command language, and the detailed
user-computer interactions.
conceptual model provides the user with a link between his
task domain and the syntax of the interactive dialogue
called a surrogate model
task-action mappings
The properties of radically different calculator
designs, such as algebraic versus stack calculators, could be
best understood by an analysis of how well calculation tasks
could be directly mapped into the actions available on the
calculators. bypassed surrogate models
we have proposed a
calculus, called ETIT analysis, for task mapping:
The "fit" of a system to a task domain can be
assessed by enumerating rules for reformulating
system-independent task descriptions ("external tasks") into
system-specific task descriptions C'internal tasks").
Rule-based system description techniques, look promising as a way provide system designers with
calculational techniques for predicting the learnability and
"guessability" of systems.
- The Task Interface: Pondering Ideas
the role of a science of the user in
the future development of the personal workstation
The challenge is to create systems that, through
intimate cognitive interaction with users, aid them in
structuring and manipulating their ideas.
The key to building such systems is to find ways by
which a user can act on his ideas as objects, just as current
text editors allow him to act on words as objects. externalize the ideas
two problems:
- representation (exploit AI and cognitive sci)
-
manipulation (exploit interactive computer graphics)
a shift in our research
strategy from studying users of existing systems to studying
users of new systems that we ourselves build, which enables
us to understand the nature of the tasks and the limits of
users and systems for dealing with them
Representation: Idea Structuring
MT: example is Scrivener
The key research issue here is to help the users develop
explicit mental models of idea structures, so they can see
them, play with them, and evaluate them. This requires the
invention of representations for externalizing ideas and idea
structures.
NoteCards
Open system
we are now observing a population of
idiosyncratic, exploratory users
Manipulation: Idea Browsing
how a computer
system can compensate for human cognitive limitations
The display can be used as an external memory to augment
the user's internal working memory
But the problem is that these techniques, which work
well with with a few dozen objects, do not scale well up to
the several hundred or even thousands of objects
Window Working Set Model analyzes access to screen objects in a manner analogous
to the analysis of virtual memory operating systems.
Informally, this model suggests that screen space itself is the
key constraint and that at some point as the number of, say,
overlapped windows required increases, user performance
will decrease in a sudden and non-linear way, sending the
user into the window version of thrashing.
Advanced graphics systems, however, open up a new
set of possibilities.
The visual movement techniques is coupled with other
retrieval techniques for allowing the user to focus on a
limited number of items at a time (as required by our
understanding of user's processing capacity) while retaining
rapid access to a large number of items (as required by our
understanding of the requirements of complex intellectual
tasks).
- Conclusions
At the physical interface level, we have discovered that
user performance with pointing devices is constrained by
the information-processing capacity of the user. We have
learned the quantitative law describing this constraint and
have determined that certain devices, such as the mouse, are
at the performance limits allowed by this law.
At the cognitive level, we have learned that routine
human-computer interaction, such as text-editing, does not
involve problem solving, but rather cognitive skills based on
the execution of known methods. We can see the
information-processing constraints of the user show
through this skilled performance in the unit task of users.
We have characterized cognitive skill to the extent of
developing an engineering model for use by user-interface
designers.
At the conceptual level, we have learned that usersoften have mental models of the systems they use, and that
such models enable performance of novel tasks. System
conceptual models provide the basis for users to acquire
mental models, and thus are an important basis for system
design. But mental models are cognitively intensive, and
users will avoid them by attempting to map directly from
their tasks to the actions required in a system. Theories of
task mapping are just beginning to emerge.
We believe that the most
interesting problems are at the task level: understanding the
nature of complex intellectual tasks and finding ways to
build idea-structuring tools, both representation tools for
structuring ideas and display tools for browsing ideas
Computer as Rorschach
Sherry Turkle
looking at the computer as a
metaphor and as a projective medium, and suggesting that
this subjective side of the computer presence is highly
relevent to understanding issues concerning computation
and public life
ethnographic investigation of the relationships that people
form with computers and with each other in the social
worlds that grow up around the machines.
ideas about computers
carry feelings about political and personal issues
But in
addition, the expert enters into relationships with computers
which can give concreteness and coherence to political
and private concerns far removed from the world of computation.
In particular, the act of programming can be an
expressive activity for working through personal issues
relating to control and mastery
there is a subjective side to people's
relationships with technology
the
elusiveness of computational process and of simple descriptions
of the computer's essential nature undermine
such consensus and make the computer an exemplary
"constructed object," a cultural object which different
people and groups of people can apprehend with very
different descriptions and invest with very different attributes.
The Subjective Computer
The fact
that the computer touches on a sphere--intelligence--that
man has long thought to be uniquely his, means that even
popular discourse about computers can raise tense questions
about what is man and what is machine.