Human-Centered Computing

Editor Blurb
Editor Blurb continued
Macrocognition
Email address

Gary Klein, Karol G. Ross, and Brian M. Moon, Klein Associates

Devorah E. Klein, Insight Product Development
Robert R. Hoffman, Institute for Human and Machine Cognition
Erik Hollnagel, Linköping University

I f we engineer complex cognitive systems on the basis of • Decisions are typically complex, often involving data
overload.
mistaken or inappropriate views of cognition, we can
• Decisions are often made under time pressure and
wind up designing systems that degrade performance rather involve high stakes and high risk.
than improve it. The results stemming from the application • Research participants are domain practitioners rather
than college students.
of any cognitive systems engineering methodology will be • Goals are sometimes ill-defined, and multiple goals
incomplete unless they include a description of the cogni- often conflict.
tion that is needed to accomplish the work. The concept of • Decisions must be made under conditions in which few
macrocognition is a way of describing cognitive work as it things can be controlled or manipulated; indeed, many
naturally occurs. key variables and their interactions are not even fully
understood.
Definition
Macrocognition is a term coined by Pietro Cacciabue In natural settings, domain practitioners rarely focus
and Erik Hollnagel to indicate a level of description of on microcognitive processes. Instead, they are concerned
the cognitive functions that are performed in natural with macrocognitive phenomena, as Table 1 shows.
(versus artificial laboratory) decision-making settings.1,2 These types of functions—detecting problems, manag-
Traditionally, cognitive researchers have conducted lab ing uncertainty, and so forth—are not usually studied in
experiments on topics such as puzzle solving, serial ver- laboratory settings. To some extent, they are emergent
sus parallel attentional mechanisms, and other standard phenomena. In addition to describing these types of phe-
laboratory paradigms for psychological research. We term nomena (the left-hand column) on a macrocognitive level,
these microcognition because they are aimed at investigat- we can also describe them on a microcognitive level.
ing the building blocks of cognition, the processes that we The two types of description are complementary. Each
believe are invariant and serve as the basis for all kinds of serves its own purpose, and together they might provide
thinking and perceiving. a broader and more comprehensive view than either by
In contrast, the methodology for macrocognition itself. We do not suggest that the investigation of macrocog-
focuses on the world outside the lab. This includes nitive phenomena will supercede or diminish the impor-
contexts designated by such terms as the “field setting,” tance of microcognition work—just that we need research
the “natural laboratory,” and the “real world.”3 Key features to better understand macrocognitive functions in order to
of cognition in naturalistic contexts include the following: improve cognitive engineering.
Another way in which the methodology for macrocog-
nition differs from that of microcognition deals with assump-
tions about cognition’s “building blocks.” Microperspec-
tives carry with them the notion of reductionism—that
explanations come from reduction to a set of basic func-
tions or components. Although we might want to reveal
specific causal sequences of various memory or atten-
tional mechanisms, this turns out to be difficult. When we
try to describe naturalistic decision making, we quickly
realize that it makes little sense to concoct hypothetical
Editors: Robert R. Hoffman, Patrick J. Hayes, and Kenneth M. Ford information processing flow diagrams believed to repre-
Institute for Human and Machine Cognition, University of West Florida sent causal sequences of mental operations, because they
rhoffman@ai.uwf.edu
end up looking like spaghetti graphs.

MAY/JUNE 2003 1094-7167/03/$17.00 © 2003 IEEE 81

Published by the IEEE Computer Society
Table 1. Important macrocognitive phenomena and traditional microcognitive lab research. ing, sensemaking, and problem detection.
For example, a research program on deci-
Macrocognition phenomena of concern Parallel traditional microcognition topics sion making started by investigating the
to domain practitioners of concern to cognitive scientists strategies used by experienced firefighters.5
Planning and problem detection Puzzle solving This research program used accounts of
Using leverage points to construct options Strategies for searching problem spaces critical incidents to propose a new model of
Attention management Serial versus parallel processing models decision, called the Recognition-Primed
Uncertainty management Estimating probabilities or uncertainty values Decision model. The RPD model tried to
explain how experienced decision makers
Explaining cognitive phenomena by Why study macrocognition? could generate effective courses of action
decomposing or reducing them to hypo- Some will object to postulating a distinc- without having to consider more than a
thetical building blocks might not always tion between micro- and macrocognition. single option. Normative models of deci-
be necessary. If anything, supplementary If both levels address cognitive processes, sion making, such as utility theory, dictate
explanatory concepts come from above why introduce new terms and a new dis- that “good” decision making involves spec-
rather than from below—for example, tinction? One reason is that without it, most ifying all the action alternatives, all the
feedback/feedforward, self-organization, researchers would likely continue experi- possible outcomes, and their likelihoods,
equilibrium, and so on. Macrocognitive mentation on microcognition and ignore and evaluating all the alternatives for their
functions can be considered as perspec- macrocognition. Second, the study of costs and benefits. The RPD model postu-
tives, but not in the sense that the constit- macrocognition might require a different lates that we can use pattern matching to
uent functions are necessarily elements, approach to research. Third, we believe categorize a situation, so that the recogni-
or elementary in any way. And they are that the field of microcognition will also tion of familiarity (case type) evokes a
rarely like the “basic” cognitive functions benefit by being contextualized by macro- recognition of the typical way to respond.
of microcognition. It is more like the phe- cognitive functions. Furthermore, experienced decision makers
nomenon you often find in a functional Macrocognition comprises the mental can evaluate a single course of action by
analysis—that is, that function A is a pre- activities that must be successfully accom- mentally simulating it rather than by delib-
condition for function B, and function B plished to perform a task or achieve a goal. eratively comparing it to other options.
is in turn a precondition for function A. Other somewhat related terms have been After considerable research on recogni-
In that sense, each one encompasses the used in this regard, such as situated cogni- tion-primed decision making, we realized
other, but one is not more elementary than tion and extended cognition.4 These terms that the model was basically a combination
the other. Each description has a value in describe the fact that macrocognitive of three decision heuristics that had already
itself, and the fact that multiple descrip- functions are generally performed in col- been well-studied from the microcognition
tions exist only reflects that you can look laboration—by a team working in a natural perspective: availability and representative-
at something from different viewpoints situation, and usually in conjunction with ness to identify the typical course of action,
and different levels. The linkages we should computational artifacts. The emphasis in and the simulation heuristic to evaluate the
look for are therefore dynamic ones that macrocognition is on cognitive functions, course of action.6 Therefore, in this case it
can explain how functions or behaviors and teams can perform these. Thus, we can was possible to trace the macrocognitive
can emerge and interact. study how the barriers to effective problem phenomenon back to hypothetical micro-
To some extent, macrocognitive phe- detection might be different for individuals cognitive components. However, several
nomena take place over longer time peri- than for teams. Macrocognitive functions decades of research on the availability, repre-
ods than microcognitive phenomena, but can be performed using information tech- sentativeness, and simulation heuristics
the distinction is not time-linked. Some nology, or without any technology at all, had not led to a discovery of recognitional
macrocognitive phenomena happen very and we can study how technology helps us decision making. That is why we see the
quickly, and some aspects of microcogni- past some barriers but introduces others. We macrocognitive functions as emergent. We
tion, such as puzzle solving, can be drawn prefer the term macrocognition because in discover them by investigating cognition in
out. Macrocognition often involves ill- addition to broadening the focus to include field settings rather than by continually pur-
defined goals, whereas microcognitive the team and technology context, it also suing explanations of lab findings.
tasks usually have well-defined goals. broadens the level of description of the
As researchers learn more about macro- cognitive functions themselves. General A variety of macrocognitive
cognition, they are likely to clarify its rela- approaches such as situated cognition are functions
tionship to microcognition. However, the important for explaining why cognitive Our current list of the major macrocog-
two levels might not line up neatly. Micro- functions must be studied in natural con- nitive functions appears in the center of
cognitive research has posited a set of dis- texts, but they only point to the need to dis- Figure 1.5,7–12 The circle around the primary
tinctions (for instance, the difference cover and understand the macrocognitive functions shows a range of supporting
between memory and inference) that might functions that operate in natural contexts. macrocognitive processes.13–19 We do not
not be useful as we study macrocognition. Furthermore, one of a macrocognition include them as primary functions because
The study of macrocognitive functions will framework’s intended functions is to decision makers, at least those we have
introduce new distinctions that will have to encourage the development of descriptive studied, do not carry out these processes as
be evaluated on their own merits. models of processes such as decision mak- an end in itself but rather as a means for

82 computer.org/intelligent IEEE INTELLIGENT SYSTEMS

achieving the primary functions listed. domains such as fire-
ng
This distinction is as much for pragmatic fighting, critical-care Maintaini common g
in g rou
as for theoretical purposes: to highlight nursing, and military d nd
boar
those functions that repeatedly emerge as decision making and ry De
to

ve
s
ends in themselves across a variety of proj- is tempered by the Naturalistic decision making

nd

lop
na
ects in various domains. features of the

ing
latio

me
Additional macrocognitive functions domain.5

imu
Sensemaking / Situation assessment

ntal
and supporting processes will eventually be • Experienced people

Mental s

models
added to this set; some of the functions in rely more heavily on
the figure might be subsumed into others recognitional strate- Planning
as researchers make new discoveries. For gies. When people are
Adaptation / replanning

ement
Attentio
instance, we have not included situation just learning about a
awareness7 in Figure 1 because it is a state domain, their ap-

anag
nm
rather than a process; it arises through sense- proach tends to be Problem detection

ty m
ana
making and situation assessment. Basically, more analytic and

gem

ain
we are less concerned with presenting an deliberative. Coordination

ert
en
official list than with encouraging research • If people have any

nc
t
Tu U
at the macrocognitive level of description. experience in a rni
ng
c tion
leve
We considered trying to diagram the domain, the first rage of a
points into courses
relationships between the different func- option they generate
tions and supporting processes in the is usually plausible
format of processing diagrams—the cur- (and certainly not Figure 1. Macrocognitive functions and supporting processes for
rency of cognitive science—but decided random). individuals, teams, and information technologies.
that such a representation is still premature. • People typically
In most natural settings, the decision maker evaluate options
must accomplish most or all of these func- using mental simulation rather than ana- nature of the empirical world, continually
tions, often at the same time. A macrocog- lytical comparison. revising conceptions of it and remaining
nitive function such as problem detection • As people gain experience, they spend flexible in methods of discovery and analysis.
can be an end in itself for a mission such more time examining the situation and less In the case of complex cognitive systems, the
as intensive-care nursing or intelligence on contrasting the options, whereas novices naturalist probes the world in which people
analysis, or it can be a means toward an spend more time contrasting options and actually live and work and the emerging situ-
end of command and control replanning. less on comprehending the situation. ations in which they find themselves. The
Mental simulation and storybuilding are approach becomes most salient when con-
typical strategies for sensemaking but are Many of the accounts researchers have trasted with attempts to abstract or simulate a
also supporting strategies for naturalistic provided of macrocognitive functions and piece of the empirical world, as is typical in
decision making. A mental model of a situ- processes are preliminary and tentative. laboratory studies, or to substitute a preset
ation must be developed for decision mak- Nevertheless, they are the best descriptions image of it, as in many information process-
ing, sensemaking, effective planning and currently available—because macrocogni- ing accounts of cognition.
replanning, coordination, adaptation, and tive processes have received so little atten- The naturalistic approach could yield an
replanning. In other words, everything can tion. That is a major reason for calling out empirical basis for macrocognition. Yet, when
be connected to everything. This makes macrocognition as a distinct framework. someone proposes it to the research commu-
any attempt at depicting a flow diagram We must study these types of functions nity as an investigative approach, standard
either ad hoc or useless because cognition, and processes, even though they do not fit methodological objections are often raised:
as it occurs in the world, can’t be “frozen.” neatly into controlled experiments. We Naturalism does not follow the experimental
Some of the functions that Figure 1 must find ways to conduct cognitive field paradigm, it (therefore) lacks rigor, the proce-
depicts have been studied to a level of research that can improve our understand- dures are (therefore) soft, and the results are
specificity that enables the creation of spe- ing of the functions and processes encoun- (therefore) not generalizable. From our van-
cific models, whereas others are still in tered at the macrocognition level. tage point, these objections are wrong, a clear
the early stages of modeling. An example case of methodolatry. Many grand figures of
of a specific model is the RPD model, A natural science research science exemplify the naturalist at work—
mentioned earlier, which has generated approach Charles Darwin, Jean Piaget, Galileo Galilei.
several empirical generalizations about We propose that the naturalistic perspec- It would be nonsense to say that Darwin con-
lawful relationships: tive is appropriate for studying macrocog- tributed nothing to science because he did
nition.20,21 Naturalists develop theories, not formulate his theory of evolution as a
• People make most decisions using re- concepts, and methods by observing and consequence of a series of lab experiments.
cognitional strategies, fewer decisions interacting with the world. Research for the Nor would it make sense to criticize Galileo
by comparing options analytically. This naturalist is a process—not a single, prede- because he did not try to hold constant certain
generalization is based on studies in fined procedure. The naturalist digs out the variables in the nighttime sky. Leading natu-

MAY/JUNE 2003 computer.org/intelligent 83

ralists created rigorous observation methods, functions we want to support. The macrocog- Proc. 2001 Command and Control Research
made valuable discoveries, and tested their nition framework is intended to clarify what and Technology Symp. (CD-ROM), Naval
hypotheses, leading to the conclusion that these functions are, so that we can do a better Postgraduate School, 2001.
“more discoveries have arisen from intense job of studying and supporting them. 11. G. Klein et al., “Features of Problem Detec-
observation of very limited material than from tion,” Proc. Human Factors and Ergonomics
statistics applied to large groups.”22 Soc. 43rd Ann. Meeting, vol. 1, Human Fac-
The naturalistic perspective qualifies as tors and Ergonomics Soc., 1999, pp. 133–137.
being scientific in the best meaning of that Acknowledgments 12. K. Weick, Sensemaking in Organizations,
term. The long-held view that the study of Sage Publications, 1995.
We thank David Woods, Rob Hutton, Jeffrey
cognition must adhere to tightly controlled Sanchez-Burks, and Amelia Armstrong for help-
studies using experimental methods would ful discussions and criticisms of the ideas pre- 13. D. Gopher, “The Skill of Attention Control:
only serve to limit us in our attempt to study sented in this article. Acquisition and Execution of Attention
The preparation of this article was supported Strategies,” Attention and Performance XIV:
and describe macrocognitive functions. Synergies in Experimental Psychology, Arti-
Our focus must turn now to formulating by the Advanced Decision Architectures Collab-
orative Technology Alliance sponsored by the ficial Intelligence and Cognitive Neuro-
criteria for evaluating naturalistic studies, US Army Research Laboratory under Coopera- science, D.E. Meyer and S. Kornblum, eds.,
as other disciplines have done.23 Our call tive Agreement DAAD19-01-2-0009. MIT Press, 1992.
for more macrocognition research is also a 14. G. Klein et al., Cognitive Wavelength: The Role
call for this research community to develop of Common Ground in Distributed Replanning,
the science of understanding human cogni- References tech. report AFRL-HE-WP-TR-2001-0029,
tion in natural settings. Wright-Patterson Air Force Research Labora-
1. P.C. Cacciabue and E. Hollnagel, “Simulation tory, 2000.
of Cognition: Applications,” Expertise and
Technology: Cognition and Human-Com- 15. G. Klein and S. Wolf, “The Role of Leverage

T he more we learn about macrocogni-

tion, the better should be the applications.
puter Cooperation, J.M. Hoc, P.C. Cacciabue,
and E. Hollnagel, eds., Lawrence Erlbaum
Associates, 1995, pp. 55–73.
Points in Option Generation,” IEEE Trans. Sys-
tems, Man and Cybernetics: Applications and
Reviews, vol. 28, no. 1, 1998, pp. 157–160.
We should be able to design better ways 2. D.E. Klein, H.A. Klein, and G. Klein, 16. G.A. Klein and B.W. Crandall, “The Role of
for using information technology, better “Macrocognition: Linking Cognitive Psy- Mental Simulation in Naturalistic Decision
interfaces, and better training programs. chology and Cognitive Ergonomics,” Proc. Making,” Local Applications of the Ecologi-
We should be able to discover strategies for 5th Int’l Conf. Human Interactions with Com- cal Approach to Human-Machine Systems,
plex Systems, Univ. of Illinois at Urbana- vol. 2, P. Hancock et al., eds., Lawrence Erl-
enabling operators to control complex and Champaign, 2000, pp. 173–177. baum Associates, 1995, pp. 324–358.
highly dynamic systems, especially sys-
tems operated in distributed environments. 3. R.R. Hoffman and K.A. Deffenbacher, “An 17. R. Lipshitz and O. Strauss, “How Decision-
Researchers have empirically demon- Analysis of the Relations of Basic and Makers Cope with Uncertainty,” Proc.
strated that a range of cognitive functions Applied Science,” Ecological Psychology, Human Factors and Ergonomics Soc. 40th
vol. 2, no. 3, 1993, pp. 309–315. Ann. Meeting, vol. 1, Human Factors and
and processes are central in complex cog- Ergonomics Soc., 1996, pp. 189–192.
nitive systems, but these functions and 4. E. Hollnagel, “Extended Cognition and the
processes have received little or no interest Future of Ergonomics,” Theoretical Issues in 18. W.B. Rouse and N.M. Morris, “On Looking
from the pertinent research communities. Ergonomics Science, vol. 2, no. 3, 2001, pp. into the Black Box: Prospects and Limits on
309–315. the Search for Mental Models,” Psychological
To a great extent, they are emergent phe-
Bull., vol. 100, no. 3, Nov. 1986, pp. 349–363.
nomena—only obvious once researchers 5. G. Klein, Sources of Power: How People
begin to investigate performance in natural Make Decisions, MIT Press, 1998. 19. J.F. Schmitt and G. Klein, “Fighting in the Fog:
contexts. The systems we would design to Dealing with Battlefield Uncertainty,” Marine
support decision making would be very 6. D. Kahneman, P. Slovic, and A. Tversky, eds., Corps Gazette, vol. 80, Aug. 1996, pp. 62–69.
Judgment under Uncertainty: Heuristics and
different if we defined decision making as Biases, Cambridge Univ. Press, 1982. 20. H. Blumer, Symbolic Interactionism, Univ. of
the process of multiattribute utility analysis California at Berkeley, 1969.
or the collection of biases that must be con- 7. M.R. Endsley, B. Bolte, and D.G. Jones,
tinually corrected. Designing for Situation Awareness: An 21. B.M. Moon, “Naturalistic Decision Making:
Approach to Human-Centered Design, Tay- Establishing a Naturalistic Perspective in Judg-
Researchers can probably be more effec- lor and Francis, 2003. ment and Decision Making Research,”
tive working as naturalists to capture and Advanced Decision Architectures Collaborative
study macrocognitive functions than by try- 8. G. Klein, “Features of Team Coordination,” Technology Alliance cooperative agreement
ing to impose an experimental structure. New Trends in Cooperative Activities, M. DAAD19-01-2-0009, US Army Research Lab-
Furthermore, macrocognitive functions are McNeese, M.R. Endsley, and E. Salas, eds., oratory, 2002, http://arlada.info/uploads/62/286/
Human Factors and Ergonomics Soc., 2001, KLEIN_Report_-_101402.doc.
linked; attempts to study individual processes pp. 68–95.
in isolation from the others will probably 22. W.I.B. Beveridge, The Art of Scientific Inves-
result in distortions. 9. G. Klein and T.E. Miller, “Distributed Plan- tigation, Vintage Books, 1957.
As we develop better tools and methods ning Teams,” Int’l J. Cognitive Ergonomics,
vol. 3, no. 3, 1999, pp. 203–222. 23. L. Athens, “Scientific Criteria for Evaluating
for cognitive systems engineering, we will
Qualitative Studies,” Studies in Symbolic
have to gain a clearer sense of the cognitive 10. G. Klein and L. Pierce, “Adaptive Teams,” Interaction, vol. 5, 1984, pp. 259–268.

84 computer.org/intelligent IEEE INTELLIGENT SYSTEMS

 T h e A u t h o r s
Gary Klein is founder Brian M. Moon is a Robert R. Hoffman is a research scientist at
and chief scientist of research associate at the University of West Florida’s Institute for
Klein Associates and Klein Associates. Con- Human and Machine Cognition and a faculty
author of Sources of tact him at Klein Asso- associate in the Department of Psychology.
Power: How People ciates, 1750 Commerce Contact him at the IHMC, 40 Alcaniz St., Pen-
Make Decisions (MIT Center Blvd. N., Fair- sacola, FL 32501; rhoffman@ai.uwf.edu.
Press, 1998) and Intu- born, OH 45324; brian@
ition at Work (Double- decisionmaking.com.
day, 2003). Contact Erik Hollnagel is full
him at Klein Associates, 1750 Commerce professor of human-
Center Blvd. N., Fairborn, OH 45324; gary@ Devorah E. Klein machine interaction at
decisionmaking.com. works at Insight Product Linköping University
Development, designing and coeditor in chief of
studies and conducting Cognition, Technology
Karol G. Ross is a user research for the de- & Work. Contact him at
senior research asso- velopment of medical, Linköpings Universitet,
ciate at Klein Associ- industrial, and consumer LIU/IKP/IAV, S-581 83
ates. Contact her at products. Contact her at Linköping, Sweden; eriho@ida.liu.se.
Klein Associates, Insight Product Devel-
10901 133rd Cir. opment, 23 Bradford St., Concord, MA 01742;
N., Champlin, MN dklein@insightpd.com.
55316; karol@
decisionmaking.com.

Advertiser/Product Index
May/June 2003

NEXT ISSUE
Advertising Sales Offices
July/August 2003: Currents in AI
In artificial intelligence’s early days, vision formed a cornerstone of the Sandy Brown
discipline. More than 20 years ago, owing largely to David Marr’s work
10662 Los Vaqueros Circle, Los Alamitos, CA
and suggestions, the direction of computer vision research shifted toward
90720-1314; phone +1 714 821 8380; fax +1 714
building a better foundation for low-level vision. Computer vision has
now become a thriving subdiscipline, with much of the low-level
821 4010; sbrown@computer.org.
foundations in place, and running on computer hardware that is adequate
Advertising Contact: Debbie Sims, 10662 Los
to support significant real-time vision. Computer vision is ready to
Vaqueros Circle, Los Alamitos, CA 90720-1314;
resume its significant role in the pursuit of AI. This special issue will
feature articles that demonstrate how intelligent vision is contributing
phone +1 714 821 8380; fax +1 714 821 4010;
to AI today. It will also suggest future directions of computer vision dsims@computer.org.
research as a contributing influence in AI.

For production information, and conference and classified advertising, contact Debbie Sims, IEEE Intelligent Systems, 10662 Los Vaqueros Circle,
Los Alamitos, CA 90720-1314; phone +1 714 821 8380; fax +1 714 821 4010; dsims@computer.org; http://computer.org.

MAY/JUNE 2003 computer.org/intelligent 85