HCI in the Global Knowledge-Based
Economy: Designing to Support Worker
Adaptation
KIM J. VICENTE
University of Toronto

Increasingly, people are being required to perform open-ended intellectual tasks that require
discretionary decision making. These demands require a relatively unique approach to the
design of computer-based support tools. A review of the characteristics associated with the
global knowledge-based economy strongly suggests that there will be an increasing need for
workers, managers, and organizations to adapt to change and novelty. This is equivalent to a
call for designing computer tools that foster continuous learning. There are reasons to believe
that the need to support adaptation and continuous learning will only increase. Thus, in the
new millennium, HCI should be concerned with explicitly designing for worker adaptation.
The cognitive work analysis framework is briefly described as a potential programmatic
approach to this practical design challenge.
Categories and Subject Descriptors: H.5.2 [Information Interfaces and Presentation]:
User Interfaces—User interface management systems (UIMS); H.5.3 [Information Interfaces
and Presentation]: Group and Organization Interfaces
General Terms: Human Factors
Additional Key Words and Phrases: Knowledge-based economy, adaptation, cognitive work
analysis

1. INTRODUCTION
Where is the field of HCI going in the new millennium? Any answer to this
question is bound to be a personal one, so I will begin by laying my
intellectual cards on the table. For the past 13 years, I have been conducting research on how to support workers in complex sociotechnical systems

This article is largely based on portions of a recently published research monograph [Vicente
1999]. This research was sponsored in part by a research grant from the Natural Sciences and
Engineering Research Council of Canada.
Author’s address: Cognitive Engineering Laboratory, Department of Mechanical and Industrial Engineering, University of Toronto, 5 King’s College Road, Toronto, Ontario M5S 3G8,
Canada; email: benfica@mie.utoronto.ca; http://www.mie.utoronto.ca/labs/cel/.
Permission to make digital / hard copy of part or all of this work for personal or classroom use
is granted without fee provided that the copies are not made or distributed for profit or
commercial advantage, the copyright notice, the title of the publication, and its date appear,
and notice is given that copying is by permission of the ACM, Inc. To copy otherwise, to
republish, to post on servers, or to redistribute to lists, requires prior specific permission
and / or a fee.
© 2000 ACM 1073-0516/00/0600 –0263 $5.00
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000, Pages 263–280.

264

•

Table I.

K. J. Vicente
Characteristics of Complex Sociotechnical Systems. See Vicente [1999] for a
detailed account.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)

large problem spaces
social
heterogeneous perspectives
distributed
dynamic
potentially high hazards
many coupled subsystems
significant use of automation
uncertain data
mediated interaction via computers
disturbances

by designing better computer-based tools (e.g., Vicente [1990; 1996; 1999]
and Vicente and Rasmussen [1990; 1992]). These application domains (e.g.,
process control plants, aviation cockpits, engineering design, and medicine)
are somewhat different from the domains with which most HCI researchers
have been concerned. More specifically, complex sociotechnical systems
tend to have many, although not all, of the characteristics listed in Table I.
And because different types of problems require different types of solution
methods, the work analysis techniques that are suitable for complex
sociotechnical systems need to be somewhat different than those in the
toolkit of most HCI researchers and designers. The theses of this article are
that problems of this type will become more prevalent in the new millennium, and that they will cause HCI to be increasingly concerned with
designing for worker adaptation. These points are best made by example.
2. CASE STUDY: HEDGE FUNDS IN AUGUST, 1998
In August of 1998, the world financial markets experienced a severe
setback. One of the interesting stories to emerge from this event was the
catastrophic losses experienced by hedge funds led by Wall Street’s “rocket
scientists.” These funds are based on very complex computer-aided trading
strategies and, thus, are a fascinating (if esoteric) example of HCI in
complex sociotechnical systems. The following account of what happened
with these hedge funds is based on the insightful article by Coy et al.
[1998].
What are Hedge Funds? Hedge funds are a high-tech form of financial
investment that relies very heavily on quantitative computer models to
make trading decisions. Hedge funds are rather unique because they are
purported to be “a clean, rational way to earn high returns with little risk”
[Coy et al. 1998, p. 116]. They are based on a sophisticated arbitrage
strategy that is intended to be “market-neutral,” meaning that the funds
are designed to make money regardless of whether prices are falling or
rising. Before August, 1998, hedge funds lived up to these claims. For
example, one hedge fund tripled in value between March of 1994 and
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

HCI in the Global Knowledge-Based Economy

•

265

December of 1997. Moreover, the fund had never lost more than 3% of its
value in any one month.
At first glance, hedge funds seem like the financial holy grail— excellent
returns with minimal risk. How is this possible? The computer models are
constructed using historical patterns based on very large amounts of data
from many years of market behavior. These historical patterns define a
referent for expected market behavior. When prices move outside of the
normal relationships defined by this referent, a signal is sent to make a
trade, the expectation being that prices will revert to their historical
patterns. In addition, other (hedge) trades are made to protect against the
anticipated risks that may accompany the initial trade. These investment
strategies usually take advantage of very small price discrepancies. As a
result, enormous investments are required to generate significant returns.
This, in turn, has led to an increasing need for borrowed money to be used
as leverage in the trades.
This brief description should make it clear that hedge funds are a very
sophisticated form of investment. The highly paid scientists that have
developed these models are frequently referred to as “rocket scientists” or
“quants” (short for quantitative), and they include two Nobel Laureates in
economics among their number.
What Happened to the Hedge Funds in August, 1998? Given their
success, hedge funds appeared to be impervious to economic disturbances.
However, in August of 1998, the claim that high returns could always be
obtained with minimal risk was shattered. For example, one hedge fund
lost 44% of its value (US $2 billion) in one month! For some, the losses were
worse than those experienced during the 1987 market crash. As one hedge
fund newsletter put it at the time, there is a “breakdown in market
structure...something highly unusual is happening” [Coy et al. 1998, p.
118]. Financial institutions using similar arbitrage strategies also lost a
great deal of money during the same period. For example, Smith Solomon
Barney Holdings lost US $300 million, and Merrill Lynch lost US $135
million.
Why Did It Happen? How could a seemingly iron-clad investment strategy lead to such catastrophic failure? Coy et al. [1998] identified a number
of related causes:
(a) the assumptions that were embedded in the computer models were
violated;
(b) there was a breakdown in the historical patterns;
(c) the computer models were “black box” models that were making automated decisions without the real-time input of seasoned traders;
(d) multiple, unanticipated events occurred all at the same time (e.g.,
Russia experienced a setback on its road to capitalism; the Asian
financial crisis worsened);
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

266

•

K. J. Vicente

(e) many people were making the same kind of bets, and money was being
borrowed heavily to finance the bets, thereby amplifying the losses.
As one financial analyst put it, “What occurred...was the financial world’s
equivalent of a ‘perfect storm’— everything went wrong at once” [Coy et al.
1998, p. 117].
Generalizing the Lessons Learned. What does any of this have to do
with HCI? After all, there is no mention of menus, mice, windows, navigation, or the WWW. If we define the field of HCI narrowly as being solely
concerned with people interacting with computers and usability as the only
criterion, then there is no lesson to be learned from this case study.
However, if we define HCI more broadly as also being concerned with
people interacting through computers to a complex world teeming with
novelty and change and usefulness as an important criterion, then there
are very important lessons to be learned from this case study.
In fact, for those who are familiar with the details of large-scale industrial disasters, such as the Three Mile Island nuclear power plant incident,
the lessons are very familiar ones (cf. Perrow [1984], Reason [1990], and
Leveson [1995]). Complex sociotechnical systems are open systems, meaning that they are subject to disturbances that are unfamiliar to workers
and that were not, and in some cases could not have been, anticipated by
designers. These unanticipated events can range from the catastrophic—
such as the multiple plant failures at Three Mile Island or the multiple
financial disturbances during August of 1998 —to the mundane. As an
example of the latter, Norros [1996] conducted a field study of flexible
manufacturing systems, and found that workers had to cope with an
average of three disturbances per hour (i.e., events during which the system
was functioning in ways that were not anticipated by designers). Because
these events are unanticipated, the procedures or automation that designers provide will— by definition—not be directly applicable in these cases.
Therefore, to deal with these disturbances effectively, workers must use
their expertise and ingenuity to improvise a novel solution. In complex
sociotechnical systems, the primary value of having people in the system is
precisely to play this adaptive role. Workers must adapt on-line in real
time to disturbances that have not been, or cannot be, foreseen by system
designers.
Playing the role of adaptive problem solver to cope with unanticipated
events is a challenge. If we expect workers to play such a role effectively,
then we should provide them with the appropriate support rather than just
expect them to play this role on their own, unaided. Specifically, we need to
design computer tools that are tailored to help workers perform open-ended
intellectual tasks that involve discretionary decision making. In the case
study above, this would involve providing financial analysts with tools that
would allow them to take advantage of their domain knowledge to improvise a solution to an unfamiliar and unanticipated set of events like the one
that occurred during August of 1998.
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

HCI in the Global Knowledge-Based Economy

•

267

Note that this conclusion generalizes well beyond the details of this
particular case study. I believe that, in the future, HCI will have to be
increasingly concerned with systematically designing tools that deliberately support adaptation. The rationale for this claim is described next.
3. THE GLOBAL KNOWLEDGE-BASED ECONOMY AND THE DEMAND
FOR ADAPTATION
Everywhere we look in the media, we see a number of buzzwords appearing
with increasing frequency, such as: innovation, knowledge worker, wealth
creation, flexible manufacturing, free trade, and global village. What do
these terms have in common, and what is their significance for HCI? These
terms are symptomatic of a fundamental shift toward a global knowledgebased economy that will lead to an increasing demand for workers, managers, and organizations to adapt in the future.
3.1 The Global Knowledge-Based Economy
There are a number of trends that have acted together to transform
qualitatively the nature of contemporary industry and economics. These
changes, in turn, pose a new set of requirements for success in many
workplaces.
Brzustowski [1998], the President of the Natural Sciences and Engineering Research Council of Canada, has identified the following contributors:
—The market is full of products based on recent advances in science and
technology
—These products are made all over the world, regardless of brand name
—High quality is expected in the market, and is frequently being achieved
—Today’s new high-tech products become tomorrow’s commodity products
—The market for medium and low-tech goods is fiercely competitive
—Commodity prices are low, and distance is not a factor even for bulk
materials
—Everybody has to compete on price, so productivity improvement is key
—Changes in demand and market conditions can be rapid and unpredictable
—New knowledge appears in new goods and services at an increasing pace.
According to Brzustowski, these factors have led to a global knowledgebased economy that is qualitatively different from the economy of the past.
The speed and connectedness of the global knowledge-based economy is
well illustrated by Motorola’s manufacturing of electronic pagers [Morone
1993]. Seventeen minutes after an order for a pager is received from
anywhere in the United States, a bar code is placed on a blank circuit board
in a factory. Within 2 hours of the order, a finished product is shipped, even
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

268

•

K. J. Vicente

if the lot size is one. Davis and Meyer [1998] provide additional evidence for
the qualitative changes represented by the global knowledge-based economy. What implications do these changes have for success in the workplace and, thus, for HCI?
3.2 The Future Demand for Adaptation
I believe that the trend toward a global knowledge-based economy will
increase the need for adaptation. Workers, managers, and organizations
will all have to become more flexible and adaptive than in the past. There
are a number of arguments from a diverse set of sources that can be put
forth in support of this claim.
Workers. The U.S. National Academy of Science/National Research
Council Committee on Human Factors issued a report a few years ago,
documenting what it believed were the most fertile areas for human factors
research in the next few decades [Nickerson 1995]. The report was written
by a diverse group of experts, spanning the entire range of human factors,
from physical to psychological to social-organizational aspects. One of the
important points made in the report is that computerization will continue
to change the nature of work. “Technological change means changes in job
requirements. The ability to satisfy changing, and not entirely predictable,
job requirements in a complex, culturally diverse, and constantly evolving
environment will require a literate workforce that has good problem solving
skills and learning skills” [Nickerson 1995, p. 22]. Consequently, “a critical
aspect of industrial competitiveness will be the ability to adapt quickly to
rapid technological developments and constantly changing market conditions” [Nickerson 1995, p. 42]. Therefore, in the future, the work force will
have to be more versatile and adaptive than in the past.
In his recent monograph on the interaction between engineering and
society, Pool [1997] reaches essentially the same conclusion but from a
somewhat different path. By reviewing the influence that society has had
on technology, Pool noted that there is an increasing social need to “do
more with less.” This trend will only increase with the demands imposed by
the global knowledge-based economy. Because of the premium that is put
on increasing efficiency and providing new functionality, engineering systems have become—and will continue to become—more complex. This
increase in complexity has had an unintended result, namely a commensurate increase in unanticipated events. In other words, as systems become
more complex, they become more open. Unanticipated disturbances are
bound to occur. Thus, there is a greater need for worker and organizational
adaptation than in the past. By inference, we can expect that the requirement to support adaptation will increase accordingly.
Cannon-Bowers et al. [1997] describe an excellent example of the trend
identified by Pool [1997]. Because of a tremendous reduction in operating
budget, the U.S. Navy is under pressure to greatly reduce costs. As a result,
the Navy has set the goal of reducing the manning level on a future
generation of ships from the current level of 350 workers to an envisioned
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

HCI in the Global Knowledge-Based Economy

•

269

level of 95 workers. Clearly, this is a very ambitious target, but the
magnitude of the design problem is compounded by the fact that the
missions that such ships are expected to play in the future will increase in
complexity. There will be more missions. They will be more varied in
nature, and their nature will be highly unpredictable. Therefore, to meet
the challenge to do “more with less” effectively, there will be a greater need
to design tools that help workers perform open-ended intellectual tasks
that involve discretionary decision making.
Managers. Adaptation is relevant, not just to workers, but to managers
as well, as evidenced by Morone’s [1993] fascinating field studies. He
studied corporate general managers that have been successful in turning
their companies (Motorola, Corning, and General Electric) into global
leaders in high-technology markets. Such markets are particularly affected
by the symptoms accompanying the trend toward a global knowledge-based
economy, exhibiting a high degree of volatility and uncertainty. Morone
learned that successful managers adapted to all of this turbulent change to
keep their companies competitive. As one manager put it, “You need a
strategic intent— but within that context, you have to be totally opportunistic.... You can’t know what’s around the next corner, so construct an
organization that is able to adapt” [Morone 1993, p. 119]. In another part of
his book, Morone observed that some of the successful businesses he
studied “followed the general course that had been hoped for, but the
specific form they took, the specific market and technological developments
to which they had to respond as they followed that general course, were
not, and could not have been, anticipated” [Morone 1993, p. 190]. Thus, yet
again we see the need to adapt to unanticipated events. Finally, another
manager pointed out: “A lot of people think of product development as
involving a lot of planning, but... the key is learning, and an organization’s
ability to learn” [Morone 1993, p. 224]. The high-tech markets studied by
Morone have been particularly volatile in the past, and show every sign of
continuing to be so. However, as the trends identified by Brzustowski
[1998] affect different sectors of the economy, we can expect to see the same
strong need for managers to adapt to uncertainty and novelty in other
industries as well.
Organizations. Adaptation is also relevant at an organizational level. In
his national bestseller, Senge [1990] discussed the importance of organizational learning in the global knowledge-based economy. In many cases,
traditional ways of governing and managing have become outdated and are
breaking down because of the changes documented by Brzustowski [1998].
The static, hierarchical organizational structures that have dominated in
the past are no longer as appropriate, given the current pace of change.
According to Senge, one path that modern organizations can adopt in order
to succeed and survive in this environment is to learn to accept, embrace,
and seek change. This, in turn, would require a decentralized organizational structure whose individuals are committed to continuous learning
and adaptation to novelty.
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

270

•

K. J. Vicente

More recently, Davis and Meyer [1998] have identified a similar set of
corporate requirements. Because the pace of change has accelerated, the
rules that guided corporate decision making in the past are no longer as
reliable. As a result, companies might have to give up on the idea of stable
solutions to business problems. Instead, they may have to move from
relying on prediction, planning, and foresight to building in flexibility,
speed, and self-organization. In short, “an enterprise must adapt to its
environment....It needs to be every bit as adaptable as the economy in
which it participates” [Davis and Meyer, 1998, p. 114].
Summary. The global knowledge-based economy will continue to transform the landscape of modern work. Changing and unpredictable circumstances will be the norm. As a result, there is an increasing need for
workers, managers, and organizations to become more flexible and adaptive than in the past. These trends suggest that the requirement to design
for adaptation will only increase in the future.
3.3 The Relationship between Adaptation and Learning
Along with the buzzwords identified earlier, we also frequently find another set, such as: life-long learning, continuous improvement, and learning organization. How does continuous learning fit into the picture I have
been drawing?
So far, I have argued that the global knowledge-based economy has
created a strong need for adaptation to change. In essence, this is equivalent to a call for learning to learn. By adapting to disturbances, workers
are, in effect, engaging in opportunities for learning. Hirschhorn made this
point in the context of process control: “Each time operators diagnose a
novel situation, they become learners, reconstructing and reconfiguring
their knowledge” [Hirschhorn 1984, p. 95]. The connection between adaptation and learning goes well beyond process control, however. Related
conclusions have been reached in other research areas, such as psychology,
control theory, and cognitive science (e.g., Narendra [1986], Gibson [1991],
Johannson [1993], and Norros [1996]).
Nowhere is the robust relationship between adaptation to novelty, action
variability, and learning opportunities as well thought out as in the study
of human motor control in ecological psychology, thanks to the seminal
work of Nicholai Bernstein. This relationship was clearly expressed in a
book written approximately 50 years ago but that has only been published
much more recently [Bernstein 1996]. Bernstein was concerned with a very
different problem, so his terminology is different than that used here. For
instance, he uses the term “dexterity” to refer to the capability to find “a
motor solution for any situation and in any condition” [Bernstein 1996, p.
21]. In the terms of this article, he is referring to the capability to adapt to
unanticipated demands.
According to Bernstein, the need for dexterity (i.e., adaptation) increases
under the following conditions: (a) the problem to be solved becomes more
complex; (b) the problem to be solved becomes more variable; (c) the
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

HCI in the Global Knowledge-Based Economy

•

271

number of unique and unexpected problems that need to be solved in
real-time increases. Although the content is obviously different for motor
control, these generic characteristics are surprisingly similar to those that
are associated with the global knowledge-based economy and complex
engineering design projects. Demands on workers are becoming more
complex; those problems take different forms, rarely repeating themselves;
and, there is an increasing need to deal with novel, unanticipated situations in a timely fashion. Just as these characteristics lead to an increase in
the need for dexterity in motor control, they also lead to an increase in the
need for adaptation in complex sociotechnical systems.
These similarities establish the connection between motor dexterity and
worker adaptation. How does the relationship with learning fit in? The
answer to this question can be found in the rationale behind Bernstein’s
beautiful phrase, “repetition without repetition” [Bernstein 1996, p. 204,
emphasis in original]. Allowing workers to play the role of adaptive actors
means that they will repeatedly have to generate a solution to a problem,
rather than following a prepackaged procedural solution. As Bernstein
pointed out in the context of motor control, “during a correctly organized
exercise, a student is repeating many times, not the means for solving a
given motor problem, but the process of its solution, the changing and
improving of the means” [Bernstein 1996, p. 205, emphasis in original].
But because workers are solving the problem anew each time, opportunities for learning are created. This process is explained by this extended
quotation:
Repetitions of a movement or action are necessary in order to solve a motor
problem many times (better and better) and to find the best ways of solving it.
Repetitive solutions of a problem are also necessary because, in natural
conditions, external conditions never repeat themselves and the course of the
movement is never ideally reproduced. Consequently, it is necessary to gain
experience relevant to all various modifications of a task, primarily, to all the
impressions that underlie the sensory corrections of a movement. This experience is necessary for the animal not to be confused by future modifications of
the task and external conditions, no matter how small they are, and to be able
to adapt rapidly [Bernstein 1996, p. 176, emphasis in original].

Bernstein’s statements can be generalized to the case of workers in the
global knowledge-based economy. Here too, conditions rarely repeat themselves precisely, so workers should gain experience with solving problems
under a wide variety of initial conditions. Supporting workers to be
adaptive problem solvers accommodates this need for learning because it
recognizes the situated nature of action [Suchman 1987]. By accommodating context-conditioned variability in action (cf. Turvey et al. [1978]), we
can help workers gain valuable experience so that they are not confused if
they have to perform the same task in a different way in the future because
of a change in context. In other words, designing for adaptation is equivalent to designing for continuous learning.
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

272

•

K. J. Vicente

3.4 Recap: How Much Have Things Changed?
In this section, I have argued that the characteristics of the global knowledge-based economy put a premium on adaptation and continuous learning,
and that these trends will increase. The novelty of these work demands is
perhaps best illustrated by contrasting them to the Tayloristic approach
that was prevalent early in the 20th century [Taylor 1911]. The following
quotations from a recent biography of Frederick Taylor make the point in a
stark fashion:
The control of work must be taken from the men who did it and placed in the
hands of a new breed of planners and thinkers. These men would think
everything through beforehand. The workmen— elements of production to be
studied, manipulated, and controlled—were to do as they were told [Kanigel
1997, p. 371].
The work itself might be no more physically demanding, but somehow, by day’s
end, it felt as if it were. Going strictly by somebody else’s say-so, rigidly
following directions, doing it by the clock, made Taylor’s brand of work
distasteful. You had to do it in the one best way prescribed for you and not in
your old, idiosyncratic, if perhaps less efficient way [Kanigel 1997, pp. 209 –
210].

Clearly, times have changed considerably since Taylor’s days. A more
flexible approach to work analysis is needed to meet the needs of the global
knowledge-based economy.
4. COGNITIVE WORK ANALYSIS: A POTENTIAL PROGRAMMATIC
APPROACH
So far, I have described a phenomenon (the global knowledge-based economy) and an associated objective (designing for adaptation, or equivalently,
continuous learning). In this section, I will briefly describe a potential
programmatic approach for achieving that objective. Cognitive work analysis (CWA [Rasmussen et al. 1994; Vicente 1999]) is a work analysis
framework that can be used to create computer-based tools to support
worker adaptation and continuous learning. CWA was developed since the
1960s by researchers at Risø National Laboratory in Roskilde, Denmark
(see Vicente [1998; 2000] for an introduction to, and historical review of,
this intellectual lineage), and its concepts have been tailored to the
properties listed in Table I.
4.1 A Constraint-Based Approach
If a work analysis framework is going to support continuous learning and
adaptation to novelty and change, then it must be flexible enough to
support variability in action that is sensitive to local, contextual details
that cannot be, or have not been, anticipated. CWA accomplishes this by
adopting a constraint-based approach. Vicente [1999] provides a detailed
description of this approach, but the basic idea is illustrated generically in
Figure 1. Various layers of goal-relevant constraint can be identified as
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

HCI in the Global Knowledge-Based Economy

•

273

Space of Action
Possibilities

Trajectory
2
Trajectory
1

Fig. 1. The constraint-based approach. Designers identify constraints on action that can be
embedded in a computer-based tool. Workers then have the flexibility to adapt within the
remaining space of possibilities. Although not shown in the figure, the constraint boundaries,
and thus the available degrees of freedom, are situation-dependent and therefore dynamic.
Adapted from Vicente [1999].

being relevant to a particular design problem (see the following subsection
for a more detailed description of the categories of constraint adopted by
CWA). These constraints can be integrated to create a dynamic constraint
boundary. The resulting constraint space, illustrated in Figure 1, defines a
set of action possibilities. These possibilities represent the relevant degrees
of freedom for productive action. Of course, the constraint boundaries, and
thus the constraint space, will change as a function of the context (e.g., the
task being performed, the strategy being adopted, the competencies of the
actor).
By identifying these goal-relevant constraints, designers can generate a
set of information requirements that can be used to design a computerbased tool. That tool would provide workers with feedback and decision
support as to the constraints that need to be respected, but it would not
identify a particular path or trajectory through the constraint space. It
would be up to workers to decide which trajectory (i.e., set of actions) to
take for a particular set of circumstances. This constraint-based approach
thereby provides workers with continual opportunities for learning and the
flexibility to adapt within the space of relevant action possibilities. In one
circumstance, workers may select one trajectory, whereas in another circumstance, they may have to select a different trajectory to achieve the
same task goals. Thus, the constraint space provides the flexibility required
to support situated action [Suchman 1987]. In addition, the same worker
can choose different trajectories, even when the circumstances remain the
same. As a result, the constraint space is also flexible enough to support
the intrinsic variability frequently observed in human action. Finally,
different workers can choose different trajectories to achieve the same
outcome in different ways. Therefore, the constraint space is also flexible
enough to support individual differences between workers.
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

274

•

K. J. Vicente

Control
Task 2

Strategy A
Soc-Org
Worker

Control
Task 1

Strategy B

Fig. 2. The CWA framework is an example of a constraint-based approach that is comprised
of five layers: work domain, control tasks, strategies, social-organizational, and worker
competencies. These relationships are logically nested with a progressive reduction of degrees
of freedom. When the constraints are integrated, the result is a constraint space like the one
in Figure 1. Adapted from Vicente [1999].

In summary, the constraint-based approach illustrated in Figure 1 allows
workers to respond to unanticipated contingencies and to follow their
subjective preferences (by choosing a different trajectory through the
constraint space), while at the same time, satisfying the demands of the job
(by staying within the constraint boundaries). For an example of an
interface and a decision support system designed according to this philosophy, see Vicente [1996] and Guerlain et al. [1999], respectively.
4.2 Five Layers of Constraint
The CWA framework is an example of a constraint-based approach that is
comprised of the five layers of constraint. The first layer of constraint is the
work domain that is a map of the environment to be acted upon. The second
layer of constraint is the set of control tasks that represents what needs to
be done to the work domain. The third layer of constraint is the set of
strategies that represents the various processes by which action can be
effectively carried out. The fourth layer of constraint is the social-organizational structure that represents how the preceding set of demands is
allocated among actors, as well as how those actors can productively
organize and coordinate themselves. Finally, the fifth layer of constraint is
the set of worker competencies that represent the capabilities that are
required for success.
Figure 2 illustrates a hypothetical example showing how these five layers
of constraint are logically nested. The size of each set in this diagram
represents the productive degrees of freedom for actors, so large sets
represent many relevant possibilities for action whereas small sets repreACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

HCI in the Global Knowledge-Based Economy

•

275

sent fewer relevant possibilities for action. The outer boundary represents
the first phase of CWA, work domain analysis, and shows what the
controlled system is capable of doing. This level is a fundamental bedrock
of constraint on the actions of any actor. No matter what control task is
being pursued, what strategy has been adopted, what social-organizational
structure is in place, or what the competencies of the workers are, there are
certain constraints on action that are imposed by the functional structure
of the system being acted upon. For example, pilots cannot use engines for
functions that they are not capable of achieving. Thus, the work domain
delimits the productive degrees of freedom that are available for action.
The second phase, control task analysis, inherits the constraints of the
first phase but adds additional constraints as well. For the sake of clarity,
only two hypothetical control tasks are illustrated in the example in Figure
2. Although the work domain provides a large number of total degrees of
freedom, when actors are pursuing a particular control task, only a subset
of those degrees of freedom are usually relevant. For example, when pilots
are navigating at cruising altitude, the constraints associated with the
landing gear and brakes are usually not relevant. Furthermore, there are
new constraints that must be respected above and beyond those imposed by
the work domain. For example, for some control tasks, it is important that
certain actions be performed before others. This constraint is a property of
the control task, not the work domain. As shown in Figure 2, the net result
is a reduction in the relevant degrees of freedom. When workers are
pursuing a particular control task, only certain actions are meaningful and
require consideration. It is for this reason that the sets depicting control
tasks 1 and 2 in Figure 2 are nested within the set for the entire work
domain.
The third phase of CWA is depicted in the example in Figure 2 by two
hypothetical strategies, A and B, for control task 2 (strategies for control
task 1 are not shown for the sake of clarity). The strategy phase also
inherits the constraints associated with previous phases of analysis. After
all, a strategy cannot make a work domain do something that it is not
capable of doing. This is why the two sets for strategies A and B in Figure
2 are subsets of the work domain set. In addition, a strategy must also
work within the constraints associated with its corresponding control task;
otherwise it will not reliably achieve the required task goals. It is for this
reason that strategies A and B in Figure 2 are nested inside the set for
control task 2. The strategies phase also introduces new constraints of its
own, however. The control task level merely identifies the degrees of
freedom associated with achieving a particular goal. There are conceivably
many different ways in which control task 2, for instance, can be performed. All of these processes are encompassed by the control task 2 set in
Figure 2. When a particular strategy for performing the control task is
identified, some degrees of freedom are usually not relevant because they
are only required for other strategies. A specific strategy imposes a certain
flow or process that adds constraints on top of those that are imposed by
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

276

•

K. J. Vicente

merely achieving the desired outcome. It is for this reason that strategies A
and B are nested within the set for control task 2 in Figure 2.
The fourth phase of CWA, social-organizational analysis, follows a similar pattern. It too inherits the constraints imposed by previous phases of
work analysis, and it too adds a new layer of constraint. There are multiple
organizational structures that could conceivably be adopted for any one
strategy. To take a very simple example, a strategy may be performed by
one worker alone, by two workers in a collaborative manner, by one worker
supervising another, or by a worker supervising automation. In each of
these cases the same strategy is being adopted; the same control task is
being pursued; and the same work domain is being acted upon. Nevertheless, these different organizational architectures have different constraints
associated with them. A worker executing the strategy alone will likely
draw on a different, although probably overlapping, set of relevant actions
than two workers executing the strategy in a cooperative fashion. Thus, a
particular social-organizational structure represents a further narrowing of
degrees of freedom. This logic explains why the social-organizational set in
Figure 2 is nested within the set for strategy A (analogous constraints for
strategy B are not shown for the sake of clarity).
Finally, the fifth phase of worker competencies reduces the degrees of
freedom even further. There are certain things that people are simply not
capable of doing. Consequently, particular ways of working are not feasible.
For example, some activities require too much working memory load, too
much time, too much knowledge, or too much computational effort for
people to perform. These constraints are specifically associated with workers’ competencies, not with any of the other preceding phases of analysis
alone. This final narrowing down of degrees of freedom is illustrated in
Figure 2 by the worker competency set, which is nested within the
social-organizational set for strategy A.
4.3 Modeling Tools and Design Implications
The CWA framework is also comprised of modeling tools that can be used to
identify each layer of constraint [Rasmussen et al. 1994; Vicente 1999]. For
example, the abstraction hierarchy can be used to conduct a work domain
analysis (layer 1); the decision ladder can be used to conduct a control task
analysis (layer 2); and the skills, rules, and knowledge taxonomy can be
used to conduct a work competencies analysis (layer 5). These modeling
tools are used to create models for particular design problems.
As shown in Table II, each of these models is linked to a particular class
of design interventions. The list is merely intended to be illustrative, not
definitive or exhaustive. Beginning with the work domain, analyzing the
system being controlled provides a great deal of insight into what information is required to understand its state. This analysis, in turn, has
important implications for the design of sensors and models [Reising and
Sanderson 1996]. The work domain analysis also reveals the functional
structure of the system being controlled. These insights can then be used to
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

HCI in the Global Knowledge-Based Economy
Table II.

•

277

Relationships between the Five Phases of CWA and Various Classes of Systems
Design Interventions (from Vicente [1999])

(1) Work Domain

What information should be measured? (sensors)
What information should be derived? (models)
How should information by organized? (database)

(2) Control Tasks

What goals must be pursued, and what are the constraints on
those goals? (procedures or automation)
What information and relations are relevant for particular
classes of situations? (context-sensitive interface)

(3) Strategies

What frames of reference are useful? (dialog modes)
What control mechanisms are useful? (process flow)

(4) Social-Organizational

What are the responsibilities of all of the actors? (role
allocation)
How should actors communicate with each other?
(organizational structure)

(5) Worker Competencies

What knowledge, rules, and skills do workers need to have?
(selection, training, and interface form)

design a database that keeps track of the relationships between variables,
providing a coherent, integrated, and global representation of the information contained therein.
The control task analysis deals not with data structures, but with control
structures. The goals that need to be satisfied for certain classes of
situations, and the constraints on the achievement of those goals, are
identified here. This knowledge can then be used to design either constraint-based procedures that guide workers in achieving those goals, or
automation that achieves those goals autonomously or semiautonomously.
In addition, this analysis will also identify what variables and relations in
the work domain may be relevant for certain classes of situations (see
Figure 2). Those insights can be used to design context-sensitive interface
mechanisms that present workers with the right information at the right
time [Woods 1991].
The strategies analysis deals not just with what needs to be done but also
with how it is to be done. Each strategy is a different frame of reference for
pursuing control task goals, each with its unique flow and process requirements. Thus, identifying what strategies can be used for each control task
provides some insight into what type of human-computer dialog modes
should be designed. Ideally, each mode should be tailored to the unique
requirements of each strategy (e.g., Pejtersen [1992]). The strategies analysis also reveals the generative mechanisms (i.e., rules or algorithms)
constituting each strategy, which, in turn, helps specify the process flow for
each dialog mode.
The social-organizational analysis deals with two very important and
challenging classes of design interventions. Given the knowledge uncovered
in previous phases, analysts can decide what the responsibilities of the
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

278

•

K. J. Vicente

various actors are, including workers, designers, and automation. These
role allocation decisions define the job content of the various actors. In
addition, analysts should also determine how the various actors can effectively communicate with each other. That analysis will help identify the
authority and coordination patterns that constitute a viable organizational
structure.
The final phase shown in Table II is the analysis of worker competencies.
Since the work demands have been thoroughly analyzed by this point, the
knowledge, rules, and skills that workers must have to function effectively
can be determined. This analysis will help develop a set of specifications for
worker training and selection (if relevant). In addition, this analysis will
also provide some insight into how information should be presented to
workers because some competencies may not be triggered unless information is presented in particular forms [Vicente and Rasmussen 1992].
Although the list in Table II is not definitive, it shows how—right from
the very start—CWA is deliberately geared toward uncovering implications
for systems design.
4.4 Summary
The CWA framework is intended to be a programmatic approach to one
large class of HCI problems: how to design computer-based tools that help
workers perform open-ended intellectual tasks that require discretionary
decision making. These tasks require continuous learning and adaptation
to novelty. CWA tries to support these demands by adopting a flexible
constraint-based approach, by identifying five categories of goal-relevant
constraints, by providing modeling tools that can be used to analyze each
layer of constraint, and by linking each layer of constraint to a particular
category of systems design interventions.

5. THE FUTURE: WHAT CAN WE BE SURE OF?
Predicting the future is always an uncertain endeavor, but given the
analysis presented in this article, there are a few things that appear to be
indisputable. As complex sociotechnical systems become more open, change
will become the norm, not the exception. Therefore, to be competitive in the
global knowledge-based economy, there will be an increasing demand for
workers, managers, and organizations to be flexible and adaptive. At the
same time, there will be an accompanying need for learning to learn.
Accordingly, computer-based tools should be deliberately and systematically designed to help workers effectively fulfill these challenging needs.
The CWA framework is intended to be a programmatic approach to this
practical problem. But regardless of whether or not CWA itself succeeds in
achieving its objectives, HCI in the new millennium should be, and will be,
increasingly concerned with systematically designing for worker adaptation.
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

HCI in the Global Knowledge-Based Economy

•

279

ACKNOWLEDGMENTS

I would like to thank Jack Carroll for encouraging me to write this article,
and the three reviewers for their exceptionally incisive and constructive
comments.
REFERENCES
BERNSTEIN, N. A. 1996. On dexterity and its development. In Dexterity and Its Development,
M. L. Latash and M. T. Turvey, Eds. Lawrence Erlbaum Assoc. Inc., Hillsdale, NJ, 1–244.
BRZUSTOWSKI, T. 1998. Engineering design and innovation, what’s the connection?. Clarice
Chalmers’ Design Lecture. Department of Mechanical and Industrial Engineering, University of Toronto, Toronto, Canada.
CANNON-BOWERS, J. A., BOST, R., HAMBURGER, T., CRISP, H., OSGA, G., AND PERRY, A. 1997.
Achieving affordability through human systems integration. In Proceedings of the 3rd
Annual Naval Aviation Systems Engineering Supportability Symposium (Arlington, VA),
COY, P., WOOLLEY, S., SPIRO, L. N., AND GLASGALL, W. 1998. Failed wizards of Wall Street.
Bus. Week, Sept. 21, 114 –119.
DAVIS, S. AND MEYER, C. 1998. Blur: The Speed of Change in the Connected Economy.
Addison-Wesley, Reading, MA.
GIBSON, E. J. 1991. An Odyssey in Learning and Perception. MIT Press, Cambridge, MA.
GUERLAIN, S. A., SMITH, P. J., OBRADOVICH, J. H., RUDMANN, S., STROHM, P., SMITH, J. W.,
SVIRBELY, J., AND SACHS, L. 1999. Interactive critiquing as a form of decision support: An
empirical evaluation. Hum. Factors 41, 1, 72– 89.
HIRSCHHORN, L. 1984. Beyond Mechanization: Work and Technology in a Postindustrial Age.
MIT Press, Cambridge, MA.
JOHANNSON, R. 1993. System Modeling and Identification. Prentice-Hall, Englewood Cliffs,
NJ.
KANIGEL, R. 1997. The One Best Way: Frederick Winslow Taylor and the Enigma of Efficiency.
Viking, New York, NY.
LEVESON, N. G. 1995. Safeware: System Safety and Computers. Addison-Wesley Longman
Publ. Co., Inc., Reading, MA.
MORONE, J. G. 1993. Winning in High-Tech Markets: The Role of General Management.
Harvard Business School Press, Boston, MA.
NARENDRA, K. S. 1986. Adaptive and Learning Systems: Theory and Applications. Plenum
Press, New York, NY.
NICKERSON, R. S. 1995. Emerging Needs and Opportunities for Human Factors Research.
National Academy Press, Washington, DC.
NORROS, L. 1996. System disturbances as springboard for development of operators’
expertise. In Cognition and Communication at Work, Y. Engeström and D. Middleton, Eds.
Cambridge University Press, New York, NY, 159 –176.
PEJTERSEN, A. M. 1992. The Book House: An icon based database system for fiction retrieval
in public libraries. In The Marketing of Library and Information Services 2, B. Cronin, Ed.
ASLIB, The Association for Information Management, London, UK, 572–591.
PERROW, C. 1984. Normal Accidents: Living with High Risk Technologies. Basic Books, Inc.,
New York, NY.
POOL, R. 1997. Beyond Engineering: How Society Shapes Technology. Oxford University
Press, Inc., New York, NY.
RASMUSSEN, J. 1986. Information Processing and Human-Machine Interaction: An Approach
to Cognitive Engineering. North-Holland Publishing Co., Amsterdam, The Netherlands.
RASMUSSEN, J., PEJTERSEN, A. M., AND GOODSTEIN, L. P. 1994. Cognitive Systems Engineering.
Wiley series in systems engineering. John Wiley and Sons, Inc., New York, NY.
REASON, J. 1990. Human Error. Cambridge University Press, New York, NY.
REISING, D. V. AND SANDERSON, P. M. 1996. Work domain analysis of a pasteurization plant:
Building an abstraction hierarchy representation. In Proceedings of the Human Factors and
Ergonomics Society 40th Annual Meeting (Santa Monica, CA), Human Factors and Ergonomics Society, Inc., Santa Monica, CA, 293–297.
ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.

280

•

K. J. Vicente

SENGE, P. M. 1990. The Fifth Discipline: The Art and Practice of the Learning Organization.
Doubleday, New York, NY.
SUCHMAN, L. A. 1987. Plans and Situated Actions: The Problem of Human-Machine
Communication. Cambridge University Press, New York, NY.
TAYLOR, F. W. 1911. The Principles of Scientific Management. Harper and Row Publishers,
Inc., New York, NY.
TURVEY, M. T., SHAW, R. E., AND MACE, W. 1978. Issues in the theory of action: Degrees of
freedom, coordinative structures and coalitions. In Attention and Performance VII, J.
Requin, Ed. Lawrence Erlbaum Assoc. Inc., Hillsdale, NJ, 557–595.
VICENTE, K. J. 1990. Coherence- and correspondence-driven work domains: Implications for
systems design. Behav. Inf. Tech. 9, 6, 493–502.
VICENTE, K. J. 1996. Improving dynamic decision making in complex systems through
ecological interface design: A research overview. Syst. Dyn. Rev. 12, 251–279.
VICENTE, K. J. 1998. An evolutionary perspective on the growth of cognitive engineering: The
Risø genotype. Ergonomics 41, 2, 156 –159.
VICENTE, K. J. 1999. Cognitive Work Analysis: Toward Safe, Productive, and Healthy
Computer-Based Work. Lawrence Erlbaum Assoc. Inc., Hillsdale, NJ.
VICENTE, K. J. 2000. Cognitive engineering research at Risø from 1962–1979. In Human/
Technology Interaction in Complex Systems, E. Salas, Ed. JAI Press, Inc., Greenwich,
CT. In Press.
VICENTE, K. J. AND RASMUSSEN, J. 1990. The ecology of human-machine systems II: Mediating
“direct perception” in complex work domains. Ecol. Psychol. 2, 3, 207–250.
VICENTE, K. J. AND RASMUSSEN, J. 1992. Ecological interface design: Theoretical foundations.
IEEE Trans. Syst. Man Cybern. SMC-22, 4, 589 – 606.
WOODS, D. D. 1991. The cognitive engineering of problem representations. In HumanComputer Interaction in Complex Systems, J. Alty and G. Weir, Eds. Academic Press Ltd.,
London, UK, 169 –188.

ACM Transactions on Computer-Human Interaction, Vol. 7, No. 2, June 2000.