/f/-53
e7 REQUIREMENTS
FOR PSYCHOLOGICAL TOWARDS ECOLOGICAL
MODELS TO SUPPORT TASK ANALYSIS
DESIGN:-
Alex Kirlik Georgia
Institute
of Technology
Abstract Cognitive engineering is largely concerned with creating environmental designs to support skillful and effective human activity. The goal of this chapter is to propose a set of necessary conditions for psychological models capable of supporting this enterprise. An analysis of the psychological nature of the design product is used to identify a set of constraints that models must meet if they can usefully guide design. It is concluded that cognitive engineering requires models with resources for describing the integrated human-environment system, and that these models must be capable of describing the activities underlying fluent and effective interaction. These features are required in order to be able to predict the cognitive activity that will be required given various design concepts, and to design systems that promote the acquistion of fluent, skilled behavior. These necessary conditions suggest that an ecological approach can provide valuable resources for psychological modeling to support design. Relying heavily on concepts from Brunswik's and Gibson's ecological theories, ecological task analysis is proposed as a framework in which to predict the types of cognitive activity required to achieve productive behavior, and to suggest how interfaces can be manipulated to alleviate certain types of cognitive demands. The framework is described in general terms, and illustrated with an example from our previous research on modeling skilled human-environment interaction.
Note: This paper is a draft of a chapter for a forthcoming book, The Ecology of Human-Machine Systems. J. M. Flach, P. Hancock, J. Caird, and K. J. Vicente (Eds.), Hillsdale NJ: Erlbaum. Please do not cite or distribute. Comments are welcome to:
[email protected]
This research Technology,
was supported by NASA Ames grant NAG2-656 Robert J. Shively, technical monitor. (NASA-CR-1903a_) PSYCHNLOGICAL
to the Georgia
Institute
of
ReqUIREMENTS F_R HOOrLS TO SUPPORT DESIGN:
T"_ARDS £COL_CICAL TASK Inst. of Tech.) 43 p
ANALYSIS
N92-25732
(Georgia c,3/53
onclas 0091207
INTRODUCTION Modernpsychologyjudgesits progressandproductsby a variety of number
of paradigms
community theory
gives
in current
no less that thirteen
when
you see one?"
computational,
evolutionary)
successful activity
even
thirteen
criteria
individual
research
science
efforts
in certain
formalization model
processes
have emerged
could
this divide
and conquer
paradigms
meets
resulting
array
paradigms Perhaps
its own goals,
but rather,
that comprise
anything,
will have
problems
which
constraints
square
increased
along
overall
number
meeting
of the journeys of constraints.
multiple
Knowing
the least expensive
and the one with the healthiest
menu
and success
able to integrate
efforts
is whether
diverging
between
of
the
in a way that will the many
research
or converging however,
paths.
that although
different
successful hills,
us. A coherent,
however,
do good
by decomposing
restaurant
is not particularly
condition
each of the many
may lead to individually
emerge
concern,
theory.
Rarely,
constraints
paralyzing
The eventual
we may find we have all climbed
the difficulty
path so that
a necessary
on whether
I am concerned,
criterion
most of
mechanisms
or evolution.
concern
are moving
where
is the primary
cognitive
our research
One primary
Research
set of otherwise
we are somehow
decomposed
science
search,
and in still others
and coherent
to a single scientific
to meet a large
themselves.
in town,
helpful
and if
useful
cognitive
solutions
to
the problem
via the
the one with the best
in allowing
one to find a good
meal at a fair price.
The purpose capable
adherence
involve
on whether
are judged
of merit.
realization
not so much
best left for time to decide.
when considered actually
hinges
we have
unification.
in others,
development
into useful
cognitive
this is a question
hillclimbing,
food,
whether
theoretical
strict and dogmatic
theory
products
by a diverse
that the proposed
human
of course,
programs
on any one search
computational
constraint
that the research
(e.g., experimental,
parallel
are suspended
a
do you tell a good cognitive
he mentions
a massively
Reviewing
suggests
"How
so that research
unencumbered
venture,
to wonder
allow for eventual
success
through
(1988)
only one of these standards
may be a demonstration
of research
It is natural
adherents
paradigms
the major
for a theoretical
criteria
thus resembles
can proceed
Claxton
to the question:
meet perhaps
for scientific
For example,
mathematical
answers
has enough
cognitive
Claxton's
psychology,
Each of the thirteen
if their products
in current
constraints.
cognitive
criteria.
of this chapter
of supporting
the design
i.e., cognitive
engineering
1986; Woods
and Roth,
is to identify
of environments
(Fischoff, 1988).
a set of necessary
Slovic,
This effort
to promote
and Lichtenstein, is motivated
conditions skillful
for psychological
and effective
1978; Norman,
by my own limited
human
1986;
success
models activity,
Rasmussen, in attempting
2 to applytheproductsof cognitivescienceto cognitiveengineering.My experienceshaveled beto believethatthecentralproblemthatneedstobeovercometo maketheproductsof cognitivescience morerelevantto designis identifyinga moreproductivesetof dimensionsalongwhich modeling effortscanbedecomposed.We simplymustdecompose thecomplexproblemof cognitive modelingin orderto makeanyheadway.However,for cognitiveengineeringatleast,the decompositionmustbederivedfrom anoverallframeworkcapableof ensuringthattheresulting researchproductscanbereassembled into a coherenttheoryusefulfor design. A descriptionof a solutionto anyproblem,evenif expressedonly asa setof necessary conditions,playsa crucialrolein formulatinga problemdecompositionstrategycapableof ensuringthatthe subproblemsolutionscanbeeffectivelyintegrated.We haveto knowwherewe aregoingif we wantto get there. In termsof thepreviousanalogy,we haveto know thatour goal is a goodsquaremealata fair pricein orderto determinehowto decomposetheproblemof finding anappropriaterestaurant.A necessarysteptowarda moreapplicablecognitivescience,therefore, is a statementof the thesetof constraintsthatmustbemetif a psychologicalmodelis to support design.Theremaybeno goodreasonto expectthatthesetof constraintsthatmustbe satisfiedto supportdesignareidenticalto the setof constraintscognitivescientistsnormallyuseto guidetheir scientificexplorations.In fact,I will suggestbelowthatthenecessaryconditionsfor an acceptable psychologicalmodelin cognitivesciencearequitedifferentthanthenecessary conditionsfor a psychologicalmodelcapableof guidingdesign.Many of thedifficulties involvedwith trying to applycognitivesciencemodelingariseoutof this mismatch.Cognitivesciencehassimply decomposed its centralproblemin a mannerthatis very unfortunatefor thecognitiveengineer. As Carroll(1991)hasnotedin regardto thefailure of psychologyto meaningfullycontributeto understanding theproblemof human-computer interaction(HCI),therealizationthattheproducts of a basicsciencedo notprovideeffectiveresourcesfor applicationcanprovideimportantlessons for the basicscienceitself (alsoseeFlach, 1990b;Gibson,1967/1982;Neisser,1976). The solutionto theproblemof creatinga scientificbasisfor cognitiveengineeringis not merelyoneof improvingthe designer'saccessto researchfindings(e.g.,Meister, 1989),moving researchinto naturalisticor operationalcontexts(e.g.,Klein, 1989),or improvinggeneralizablityfrom experimentalresults(e.g.,Hammond,Hamm,andGrassia,1986),althougheachof thesegoalsis surelyimportant.Rather,I amconvincedthatthesolutionmustlie in a reformulationof the questionsposedby thebasicpsychologicalresearchitself: areformulationdrivenby an understanding of thepsychologicalnatureof thedesignproductandtheknowledgethatis required in orderto createit.
THE
PSYCHOLOGICAL
A standard modeling fixed
approach in cognitive
and to create a description
environment.
The designer,
environment
consistent
with the world. cognitive
environments. cognitive
solving
while the solution
the difference though,
between
of cognitive
function,
and software
just as it would
software
or ink in which
As Carroll
theories
however
psychological nonetheless principles understand
implicit
underlying
(1989)
way of
distinction
While
the scientist
not theories
creates
but merely rather
theories
environments:
than in psychological
view of what the environment
Although
a design
product
is, from
may be implemented
to view the relevant features
design.
is a lack of
restrictive
level at which
have noted,
of how the environment
each design influences
and fragmentary
features
of a psychological
in the design
its design.
Similarly,
you had better know be looked
to look for the relevant
of the VCR,
of that theory
in the
you had better
if you want to understand
at as an assemblage
structure
of the machine,
theory
are real. know
the (good) it is
engineering
If you want to
the operative
or predict
matter,
of
and skillfully
for example,
and electrical
underlying
of physical
think,
Although
VCR interface,
theories
an instantiation
behave,
may be.
of the VCR
the psychological
is actually
how people
of a (good)
of the inner workings
the functionality
product
these theories
the same sense that the physical
in the design
the VCR can of course
and theoretical
and environmental
in technological
to look for the relevant
is only implicit
or predict
problems
it is realized.
real in exactly
with the VCR,
which
be wrong
rudimentary
theory
behavior.
this is the wrong
and Campbell
the designer's
human
feasible
psychological
empirical
psychology
product.
that is best conceived
interact
as a search among possible
sophisticated
creates
that is
is the set of plausible
does not appear to be a faithful
of the design
an
as a search among possible
of the theoretical/applied
on an overly
of a design
and effectively
are symmetrical
of cognitive
that the designer
is based
and software,
product,
perform,
nature
of understanding
in hardware
dichotomy
the creation
in that
of creating
is the set of technologically
problem
These
relatively
behavior
challenge
skillfully
problem
requiring equally
the practices
it is assumed
But this perspective
the standpoint
theories
solution.
for the perpetuation
for the psychological
a mix of hardware
a person's
goal being
for the designer the scientist's
PRODUCT
space for the scientist
to a given task, and the designer's
appreciation
terms.
space
subtlety and difficulty,
One reason,
underlying
how people
terms, the solution
In this sense a theoretical/applied
portraying
activities
underlying
to obtain a given cognitive
of comparable
DESIGN
is to hold a task environment
with the ultimate
We can thus characterize
environments
methods.
behavior,
with the principles
"solutions"
THE
on the other hand, is faced with the opposite
In problem
theories
OF
psychology
of the cognitive
to elicit a desired
maximally
NATURE
physical
human
its design
interaction
as well.
this is the wrong
either for understanding
While level at
electromechanical
4 functionor for understanding userinteraction.Boththe electricalengineerandcognitiveengineer structurethephysicalmatterusingorganizingprinciplesderivedfrom theorieswithin their own disciplines.It is theadequacyof thesetheories,ratherthananyfactssolely aboutthephysical form of themachine,thatdeterminewhethertheVCR will play, andwhethertheusercanplay it. Eachinstanceof humaninteractionwith anyartifactis thusapsychologicalexperimenttesting theassumptions embodiedin theenvironmentaldesign(of.Wise's (1985)construalof an architecturaldesignin termsof scientifichypotheses).Althoughit may befashionablewithin the cognitiveengineeringcommunityto bemoanhowlittle guidancemodernpsychologyprovidesthe designer,thepsychologicalnatureof the designproductis inescapable.The correctresponseto thecurrentandunfortunatelackof applicablepsychologicalresearchis not toattempttodo psychology-freedesign(sincethis is impossible-- the designwill not beapsychologicalbut insteadreflect thedesigner's"folk" psychologicaltheory),butratherto askwhatkindsof psychologicalmodelsareneededto supportcognitiveengineering,andto beginthe long range empiricalandtheoreticalwork necessary to realizethem. This is not to imply, of course,thatcognitiveengineersshouldwait for a moreapplicable psychologyto emergebeforemakingdesigncommitments(this wouldbehopelesslynaive),or shouldturn all attentionto modelingandawayfromthe designof prototypesandexpanding technologicalopportunities.It maybe,asBraitenberg(1984)hassuggested, thathumancapacities for synthesisfar exceedcapacitiesfor analysis,in thesensethatourcreativeproductsreflect a degreeof implicit or tacitknowledgethatis far moreelaborateandrich thantheknowledgewecan explicitly stateandformalize. Thedirectmanipulationinterfacewasnot deducedin anyinteresting sensefrom psychologicaltheory-- in fact, a casecouldbemadethatwe still haveno unified psychologicaltheorythatwouldpredicttheprofoundsuperiorityof directmanipulationover commandinterfacesfor varioustasks.It just sohappenedthatin thiscasethedesignerhadan implicit understanding of howpeoplenaturallyinteractwith theworld thatwascloserto thetruth thananyexplicit andformalpsychologicaltheoryavailableatthattime. While cognitive psychologistsmaybe ableto identifywhy command-lineinterfacesareinefficient in variousways, it wasthe designersandnot psychologicalresearchthatpointedtowardenvironmentsthatsupport moreefficientinteraction,a finding thatshouldprobablyinform psychologicalresearchitself. Thereis a catch,however,to this design-as-research strategy.Assuminga particularprototype of a designconceptis successful,anyusefulgeneralizations thatemergefrom creatingthe prototypewill beat thelevel of thepsychologicalassumptions underlyingthedesign,ratherthanat thelevel of theparticulartechnologiesusedto implementthedesign.To returnto apoint made earlier,thehardwareandsoftwareimplementationis the wronglevel to look for therelevant
featuresof thedesignproduct. Especiallyin HCI, a vastamountof researcheffort hasbeen expendedtrying to answerquestionscomparingvariousinterfacetechnologies,for example, designoptionssuchas scrollingwindows,hypermedia,etc. This researchis of dubiousvalue(see alsoVicente,this volume),sincethe"it depends"answersproducedby sucheffortswill only lead to a neverendingseriesof technology-specific designprinciples,ratherthana stableandgenerative theoreticalaccountof human-environment interaction that can guide design in novel situations. To profit make
from the design-as-research
explicit
the psychological
strategy,
assumptions
system.
A successful
system
demonstrates
implicit
psychological
theory
underlying
to articulate
the theoretical
traditional
experimental
be more efficient surely
than analysis
rides on whether
were validated
strategy
must include
implicit
psychological
of the prototype.
for this purpose.
essential
argued
to that faced
guiding
and behavior
accounts
of cognitive task).
There
designer
and scientist
the other
reasons
heuristic
guidance
A search
is, however, search
over
task) is no more
in opposite
cognitive
models.
will be necessary
guidance
to the cognitive over environments
cognitive
itself
aid to design
an applied
by particular between
by discussing
activities.
will have different
faced
comparable
mode
of cognitive
than is a search
for
environments
(the
these two problems.
The
models
that different
while
types
of
As a result, the theories
properties
Let us examine
might differ
the
problem
a particular
endeavor
in the two cases.
engineer.
the
that the problem
a theoretical
to promote
difference
search
where
and communicated.
therefore,
heuristic
across
tested,
phase
be expected,
best provide
reasoning
the design-as-research
It should
to direct
that
to traditional
over environmental
to the scientist
support
using
hypotheses
one reasons
guidance
reasoning
such demonstrations
directions;
heuristic
best support
from
DESIGN
that are promoted
best provide
would
may
by an analysis
explicit,
but is instead
an important
-- synthesis
and also by showing
for environments
and behavior
progress
no alternative
is a necessary
product,
by
is probably
followed
the researcher
be demanded
impede
program
the possibly
(as would
conclusions
SUPPORT
theory
forcing
the psychological
is made
TO
application
(the designer's activity
phase
synthesis
of the design
is not one of mere
activity
scientist's
nature
by the scientist.
design
to
of the prototype unless
identify
A research
that a good psychological
psychological
by the designer
There
both an initial synthesis theory
Although
may actually
can subsequently
the researcher
than its own success,
is articulated.
methodology)
MODELING I have
other
prior to environmental
the researcher
methodology
nothing
upon
most to the success
-- the hope for generalizable
by the success
experimental
that contributed
the design
assumptions psychology
then, it is incumbent
than the theories how the theories
from the theories
that would
that that
that best
6 Much of ourcurrentunderstanding of cognitive-levelhuman-environment interactionconsists of a setof somewhatindependent triples,eachof which provides a psychologicalmodelof howapersonmight achievea particularbehaviorin a specified environment.Whenthedifficult but importantjob of integratingthisknowledgeinto coherent theoryis attempted,theseeffortstypically focuson integratingacrosstheprocessandbehavior dimensionsratherthanacrosstheenvironmentdimension.The resultis thatunderstanding the environmentalcontributionto behavioris a largelyignoredcomponentin the theoreticalunification. Oneapproach,for example,to achievingtheoreticalunificationof this setof triplesis to integrate acrossthebehaviordimension.Theresultsherearepowerful,typically hybrid "cognitive architectures"(CardandNewell, 1989).Thesegeneralpurposecognitiveframeworkshavethe processingresourcesto producea wide varietyof behaviors,from simplemotorresponsesto complexproblemsolvingandplanning.Yet anotherapproachis to integrateprimarily acrossthe processdimensionin anattemptto showthatthefunctionalityof a wide varietyof existingmodels canbesubsumedundera singleprocessmodelingformalism. Cognitivemodelsdemonstrating howsymbolicprocessingtechniquescanbeimplementedusingneuralnetworkor connectionist formalismsaregoodexamplesof partialtheoreticalunificationalongtheprocessdimension. Theoreticalintegrationsalongtheenvironmentdimension,however,arehardlyeverattempted but arecritically neededto supportthecross-environmental reasoninginherentin design. It should not comeasa surprisethatmostcognitivepsychologistsarenot overlyconcernedwith this typeof theoreticalunification,sincean acceptable scientificproductis a modelof behaviorin a specified environment,andrarelyis reasoningbackwardsfromcognitivetheoryto environmentrequired. Exceptperhapsin experimentaldesignitself,rarelyis the cognitivepsychologistforcedto reason acrossenvironmentsin orderto activatespecifiedcognitivemodes.Significantexceptions(i.e., attemptsattheoreticalintegrationacrossenvironmentalinfluenceson cognitionandbehavior)are Rasmussen's (1986)theoryof multi-levelenvironmentalrepresentation asreflectedin the "abstractionhierarchy,"andHammond,Hamm,Grassia,andPearson's(1987)effortsto obtaina rich enoughsetof environmentalandtaskdescriptorssothatthecognitivemodeunderlying judgmentbehavior(e.g.,analytical,intuitive) canbepredictedandpromotedthrough environmentalmanipulation.Only aunifiedtheoryof theenvironmentalinfluencesoncognition canguidethedesigner'ssearchfor environmentsto activatespecifiedcognitiveprocesses. But whatwould a unifiedtheoryof theenvironmentlook like, andwhattypesof guidance wouldit provide?Whatwouldbeintegratedin sucha theorywould bethediversesetof knowledgeof whatthepsychologicallyrelevantaspectsof theenvironmentmight be,for the purposeof trying to understandor predicthumanbehaviorandperformance.Onemust,for
example,determinewhenit will beappropriatetounderstandtheenvironmentin termsof stimuli andreinforcementsasin behaviorism;cues,criterionandfeedbackasin modelsof judgment; options,chancenodes,choicenodesandprobabilitydistributionsasin decisiontheory;initial states,goal statesandoperators as in Newell and Simon's (1972)problem solving theory; affordances
or consu'aints
and differential Each
equations
of the above
environmental behavior
of environmental
appropriate
will do justice
in complex
selecting
an environmental
and behavior
the cognitive
engineer
can operate
concept,
activity
expressed
maximaIIy
engineering
activities
reasoning
knowledge process
is expressed
scientific design
study. in studies
effectively
experiments
of cognition
of knowledge
necessary
experimental
psychology
the fundamental findings,
capable
difficulty
it may approach
to guide research
models
When
of
the problem
that will assist
the central
problem
in which
will be to create mode
of cognitive
for both of these cognitive that support
the designer's
but perhaps
oddly,
of cognitive activating
activity,
knowledge
guiding
mode
design
is that the knowledge needed
of cognitive
Much
needed
the knowledge activity
this
but rather
those cognitive
the designer
to reason
is quite similar
needs goes beyond
to have actually
designed
for
experimental
underlying
1987, Meister,
in the
activities
has been made of the inability
(e.g., Rouse,
of
in predicting
In other cases
a specified
Guidance
successful,
design.
to guide
the knowledge
faces
of a
behavior
as the selection
the type of understanding
a particular system
productive
engineer
design
of such a theory,
of successfully
that approaches
ability to promote
The reflection
models.
in existing
models.
to which
frameworks
and
and each way is suggestive
that promotes
of a task.
of
of cognition
1986).
will promote.
That is, it is the often tacit and unformalized
over environmental
experimenter's
ways,
cycle,
model,
have the rudiments
not so much
state variables
representation
forms
(Rasmussen,
candidate
in the design
with the demands
environmental
may already
of designing
earlier
different
the cognitive
design
can only come from theoretical
over alternative
Psychology
cases,
the environmental
system
No single
in fact, can be viewed
for an existing
as an environmental
consistent
the many
of the environment
In some
description
activity
has its place.
in various
activity,
theory,
and so on.
interaction
the structure
descriptions.
ecological
description
is manifest
A large part of design
the cognitive
design
theory,
to understanding
in behavior
environmental
(1979)
human-environment
that best describes
must be sensitive.
control
of environmental
structure
model
as in Gibson's
as in manual
forms
structure
observed
different
on action
1984),
the to the type of basic but perhaps
the experimental
these experiments.
a
ISSUES Cognitive
engineering
IN
thus demands
theoretical
basis, and the resulting
the models
used to describe
predicting
environmentally
concerned
as much
Gibson,
psychology,
of attention
techniques
internal
cognitive
situated
behavior
1991).
activity.
A cognitive
cognitive
at the types
model.
There
with a strong formal,
psychology
design
activity
model
and precise
capable
will therefore
of models
does the environmental
cognitive
modeling
must be as explicit,
and of supporting
one looks
rarely
as does the internal
models
as with internal
When
however,
MODELING
for environmental
environmental
with the environment
1979; Anderson,
cognitive
ENVIRONMENTAL
of
have to be
(e.g., Brunswik, produced
receive
1952;
by current
close
are at least three reasons
as
to the amount
for this state of
affairs. First, experimental purposes
of control,
environmental
cognitive
literature).
internal
questions
representation
used by expert
of problem
case that researchers
solving working
environmental can remain
unconfound
the environmental
environmental
manipulations
various
performance.
cognitive
To evaluate such as problem
what problems memory
implicit
the open problem
conditions
within
(but see Larkin
with the result within
a given
from the internal
engineering
activities
will be activated,
solving,
design, decision
In the following
will be, given I will discuss
arise due to the need to represent
various
1987).
are forced
program.
constraints
for cognitive
will have to be solved,
demands
research
these
what decisions design
in the types of
representation
upon
rarely
is it the
to reason
across
widely
There
may be no pressure
on cognition
the
Finally,
environmental
and behavior
to
when
range.
is to determine
under what environmental
and required,
and working will have
for effective
memory,
to be made,
for a particular
two types of constraints
on acceptable
and external
start so far
about
concepts
both internal
be the case that
the issue is not only to understand making,
science
environment
accounts
problem
that assumptions
are made over a very narrow
a candidate
(e.g., differences
and Simon,
paradigm
the cognitive
it may indeed
solvers),
model
in the
of the external
While
of the external
an established
within
the description
problem
any influences
in most papers
in research
such an approach
and novice
activity
conditions,
descriptions
processes
using
-- the environmental
model
of the environment.
that they fail to capture
efficiency
and especially
for the
and impoverished
description
psychological
is even made between
can be answered
representations
downstream
Second,
their environments
simplistic
of the stimulus
of the internal
often no distinction
interesting
Thus,
the length
of the description
and the subject's
varying
(compare
experimental
orientation,
often feel the need to simplify
and are thus able to get by with highly
models
with the length
internal
psychologists
influences
task cognitive
but also to determine and what
working
task. psychological on cognition
models
that
and behavior.
--
Thefirst setof constraintsarestructural.I will arguethatthe a model'sstructuremustbecapable of representingbothcognitiveandenvironmentalorganizationin a single,unifiedformat;i.e., that theappropriateunit of analysisandmodelingmustbe thehuman-environment system,ratherthan thehumanalone. Thesecond set of constraints concern the content of acceptable models. I will suggest
that cognitive
understanding
engineering
fluent,
skilled
that rationalize
detached
create
which
a design
cognitively-intensive features target
human
intellectual promotes
control
activity.
fluent
In most cases
and skilled
of behavior.
and also to identify
the causes
of error-prone
the Integrated
Human-Environment attempts
such as steering
electromechanical
to model
control
feedback
control
system
in order
designed
so that control
demands
language
for environmental
described
in terms
outputs
(heading).
case the input As Flach
The human
notes,
schematic
level of description,
invariance
at the organismic
These
engineers
much
human
transfer
adjust
to changes
noted,
the ability of the human
suggest
"that
control
system
in the dynamics
'the human designer
(McRuer
in order
limits.
that relates
system
controller
familiar
be) to
promotes
to define
the
a design
systems.
Control
theory
the human
(steering
could
in similar
to system
terms.
the human
transfer
control shares
function.
Rather,
of the controlled
as empirical
the human system.
to the environmental
function;
is a scientific
and interminable
and Jex, 1967) were only successful system,
rather
quest." once
the search
than in human
system.
which
to
(1954)
was so great
as to
can lure the
Subsequent
for invariance behavior
that the
appeared
and Taylor
function
(p. 1752)
At this
indicated
function
will-o'-the-wisp
i.e.,
this goal of finding
As Birmingham transfer
command.
behavior.
results
transfer
In this
be a steering
than at the level of the organism-environment however,
be
can be
adjustments)
would
modeling
as a
has a weIl specified
automobile)
manual
of
so that vehicles
(airplane,
inputs
control
with the design
and limits
during
manual
to modeling
was to discover
psychological
function'
models
that capture
concerned
and the output
current
to the level of the human-machine
which
in current
can be described
to response
into a fruitless
activity
"plant"
stimuli
to adjust
transfer
than a design
performance
capabilities
these
were in for a rude awakening
was no single
rather
Engineers
The controlled
level, rather
goal is (or should
interaction
of the automobile
relating
the design
their attention
human
the goal in this endeavor
of the function
than environmental
cognitive
turned
as a feedback
might be the heading
(1990a)
a description
to assess
function
rather
models
skilled
an aircraft.
were within
of a transfer
that assist in
System
systems
modeling.
models
environmental
human-machine
a car or flying
feedback
activity,
We require
effective,
behavior,
shifted
with the world,
that promote
One of the earliest
attempts
interaction
of environments
Modeling
there
is most in need of environmental
alone.
modeling in behavior The
10 crossovermodelof humanmanualcontrolbehaviordevelopedby McRuerandhiscolleaguesis a statementof behavioralinvarianceat thelevelof thehuman-environment system. Why shouldthis finding concerninghumanperceptual-motor behaviorinform our discussionof cognitive-level to expect
human-environment
that cognitive-level
perceptual-motor
flexible,
situation,
In fact, there
and the one pursued as an integrated
for behavioral
expressing transfer
cognitive
function
are a variety
interaction
representations
used to model
less constrained,
model
formalisms
controller
is
to this
both human
and
as a tool in the
capable
in the same language,
both the manual
than is
and thus
response
is to describe
requires
on behavior
structure
that just the opposite
human-environment
such a strategy
constraints
is even
researchers,
unit, and to use this unified Pursuing
to believe
Note also that the correct
control
to be little reason
to environmental
of reasons
interaction.
and external
is that there appears
with the world
by these manual
invariance.
both intemal
The answer
will be any less adaptive
than is perceptual-motor
environment search
behavior
activity.
that case; i.e., that human more
interaction?
of
such as the
and the controlled
system. Although
rarely
level behavior. (Brunswik,
The Lens
1952;
environment. cognitive
used, this approach
on judgment
performance
book
resources
toward
external
environmental
perform
a particular
When
structure
combine
environment descriptions
systems, modeling already
indicate
of how both internal
of understanding
Bennett,
and
for
and Joyce,
framework
and the
consistencies
mechanism
modeling.
1988).
also represents
This framework
"computational"
the processes
approach
being
constraints
that will be engaged
proposed
analyses
However,
for describing understanding of skilled
of environmentally
1988) activity
work remains
will be possible,
to the production
In a pair of penetrating
the kinds
and Holtzblatt,
it is clear that much
a clear shift toward
mutually
some
Analysis system
to determine
as to a number
(e.g., see Brehmer
human-environment
judge
both the environmental
without
the Rational
judgement
and to
task.
1987; Whiteside,
human-machine
to behavior
cognitive-
of human
of both the human
and has been enlightening
describing
the question
we turn to the problem
(Suchman,
contribute
integrated
to address
description
likely not have surfaced
(1990)
to understanding
and analysis
tool in understanding
contribution
by Anderson
step in the direction provides
is a unified
abilities,
out the environmental
The recent
1955)
that would
applied
for the description
As such, it has been a fruitful
in judgment
systems
framework
1956; Hammond,
constraints
partialing
Model
has been successfully
dynamic
typical
"situated" of behavior
to be done before the kinds human
in modem
an integrated
human-
of psychological
interaction
how both environmental
with technological and cognitive
structure
behavior.
of the cognitive-level
ecology
of human-machine
systems,
a
11 Hutchins(1987; 1991)
has suggested
apart from the external
devices
that human
cognition
in the environment
In ship navigation
checklists,
and calculators
can sometimes
be used in lieu of memorial,
operations;
and in modem
aircraft
for memory
of intended
the cockpit.
In such environments,
and environment. describing
both internal
of information
processing
environment
as external
The importance of modem
and limitations, human work
cognition
some
there
possible
mathematical
simply
not be possible,
theoretical
routine
of intimacy hypothesis behavior.
that intensive As Norman
With
thus be difficult
and perhaps
(in preparation) intellect,
amount
activity
than by increased
exploitation
a disembodied
a considerable
rather
Skilled
of environmental
terms.
We have
cognition
of the
be underestimated. cognitive
abilities
be possible
does even
for
the scientist
is out there,
in the diagram
of external
analog
cognitive
measurements,
thoughts
itself.
props
of knowledge,
--
and technical
of this kind would reached
The locus
by the mind
of a process
to any single
like
part of the internal-
(pp. 378-379).
an airplane,
force,
to understanding
driving
is often accompanied
detachment, structure
by a heightened
an observation plays
from the world,
intelligent
lots of deep planning
the
a car, or performing
that leads to the
a key role in productive
has noted: isolated
in terms
the kind of synthesis
in even greater
such as flying
requires
has noted.
to support
to attribute
up such a system.
activity,
(1991)
itself.
systems
of human
or "conclusion"
representation
cannot
But rarely
these things,
the end-state
tasks we find in daily life. with the world
histograms,
both person
and the like.
over how it can even
the synthesis
Without
also apply,
skilled
schema
within
as seen in the description
picture
upon a huge variety plots,
because
that makes
would
underlying
curves,
by the external
network
dismal
as Donald
wiring
heavily
would
tools,
across
of human
and behavior
achievements.
diagram;
at a theory.
synthesis
The same comments
many
cognitive
notations,
-- to arrive
mechanisms
in its modem
depends
directly
a rather
burden
structures
compatible
representations,
aids to thought paints
is no internal
jargon
external
problem
by a scientific
The theoretician
is driven
external
memory
these integrated
the world,
of us in a state of wonder
from external
For example, made
back upon
research
to have resulted
in isolation
turned
and
of the cognitive
is distributed
in mutually
cognitive
with notepads,
procedural,
to external
function
functions
to perform
1991), much
allocated
be understood
interaction
that gave rise to a description
of these environmental
leaving
(Hutchins,
cognitive
cognitive
has been
psychological
human
then, that understanding
metaphor
memories,
for example,
has been
the entire
and external
the computer
1987),
speeds
It is not surprising,
come full circle:
Much
and current
cannot
that have been designed
functions.
computational
(Hutchins,
and behavior
behavior
and decision
requires making,
the level
12 andefficientmemorystorageandretrieval.Whentheintellectis tightly coupledto theworld,decisionmakingandactioncantakeplacewithin thecontextestablished by thephysicalenvironment,wherethestructurescanoftenaid asadistributed intelligence,taking some of the memory and computational burden off the human. (Chapter
10, p. 6)
The human-environment internal
cognitive
environmental through
system
activity
design,
many
that promotes fluent
fluent
as opposed
existing
for effective
aircraft
an "opaque
veil,"
human
example.
and Bohlman
theory
would
cognitively
intensive
modeling
are observed
however, interaction problem
better
and airspace
problems solvers
(Woods
or decision
evolved.
processing work
theories,
it seems
safe flight.
It is natural
What
the situation
commercial as
an active
that it is appropriate
to
could
to view such attempts to create
1988), and training
has yet to be proven
accounts
of training
of activities guide
the
methods
with suspicion,
their own set of human-machine
to make people
effective.
Because
and mental
presumably
tasks, or the design
aids have the potential
problem?
generation,
that better
Such accounts
to think that
kind of psychological
awareness
hypothesis
only natural
in their theoretical
and Roth,
of
adaptation.
the cockpit
I find it most unsettling
inference,
interaction.
thinkers.
the acquisition
in the modern
is so great
is necessary
effective
of maintaining
displays
In
are quite unlike
As a result,
1991)
that
and described.
has gone so far as to describe
for remedying
systems,
makers
of environments
the operators
and Woods,
to ensure
such as inductive
solving
an environmental
of their situations.
necessary
flight crews
to create
cognitive
that the difficulty
a living constructing
in current
since problem
abilities
from cockpit
"theories"
activities
theoretical
in which
(1988)
suggests
the most leverage
of aids to assist
make crews
of
can be engineered
need to be identified
that complex
(e.g., Sarter
such as these are the key to enhancing design
activity
the features
behavior
the reason
Edwards
are sometimes
provide
intensive
(1979)
as constructing
abilities
the
as a function
and the end state is one of only partially
awareness"
As one who tries to make theoretical
behavior,
psychological
is impeded
of the aircraft
of crews
skilled
systems,
in which
is a prime
understanding speak
and effective
of "situation
cockpit
cognitive
is often (but not always)
is that the environments
of behavior
The problem
to be predicted
to allow
with the Worm
engineer
to cognitively
performance
modes
behavior
how necessary
Interaction
human-machine
those environments fluent
of Fluent
the goal of the cognitive
support
for productive
and modeling
manipulation.
The Need for Models
design
necessary
and also to identify
environmental
Since
must serve as the unit of analysis
more "rational"
to
13 The alternativesolution,of course,is to designenvironmentsmoreconsistentlywith the principlesunderlyingskilled,dynamichumaninteractionwith theworld. Pursuingthis strategy, however,requirestechniquesfor environmentalmodelingcapableof representingthe featuresof taskenvironmentsthatbothpromoteandinhibit theacquisitionof fluentmodesof behavior. Only modelsof productive,skilledbehaviorcanprovidetheresourcesfor a task-analyticapproach capableof identifyingfeaturesof anenvironmentaldesignthatareinconsistentwith theprinciples underlyingskilledactivity. Theproblemof identifyingdemandsfor complexcognitiveactivities suchasproblemsolving,planning,decisionmakingposedby a givenenvironmentaldesignmost requiresmodelsof skilled, fluentbehavior,not making
rational
action
models
considerations
Due to their roots
than they are with necessity to describe
and perhaps
all, behavior
process
and often prescribe
such as search
comparative evidence
considerations. behavior
space,
planning,
or artificial
when
The great
or decision
the result
these
performance
inference,
requires
models
Nearly
any,
or the
function.
of these rational
sorts of complex
is
cognitively-intensive
or utility
the sufficiency
of such models
of situations.
of some
to a goal structure
intelligence,
with sufficiency
appeal
hypothetico-deductive
to bear on limiting
identifying
theory
in a huge variety
as being
with respect
for successful
cognitive capable
No empirical methods
activities
of indicating
for
will when
such
are not necessary.
My observations hypothesis
of skilled
human
that cognitively-intensive
i.e., when
effective
thus requires
solutions
will not be available,
skill.
I realize
solutions
to cognitive
However,
intuitions
experiments
the cleverness
in complex for action
solutions approaches
capable
mainly
of behavior of subjects
on laboratory carefully
task demands.
in operational
findings designed
Predicting
when effective
may be skewed
in the graphics
of perception-
counter-intuitive. by the fact that
the availability
my own intuitions
demands
perception-action
opt for perception-action
as being
to preclude
cognitive
by a theory
will typically
even in the laboratory
who are able to short-circuit
off the whir of a disk drive or an aberration
are used only as a last resort;
available.
psychologist
Although
settings,
have led me to the working
can only be provided
that skilled performers
are typically
for meeting
selection
of defining
tasks may strike the cognitive based
systems
are not readily
and this knowledge
that the claim
on cognition
shortcuts
observations
modeling
behavior methods
perception-action
demands
cuing
a problem
of options
However,
be necessary
activities
through
ever be brought
selection.
actually
action
solving,
above are more concerned
can be rationalized
evaluation could
in either economic
such as those mentioned
their ability
action
of problem
activities.
Why is this the case?
action
models
of perception-
are largely
based
I am continually
for complex software.
cognitive I have
on
amazed activity
ceased
to be
at by
14 suprisedandfrustratedby suchcleverness,andhavebegunto view thetendency
toward
perceptual
is no doubt,
selection
however,
of action
as a fundamental
that more empirical
experiments
must provide
and in more
environmental
however, everyday
work
settings.
observations
of the behavior
local area grill.
My interest
in this behavioral
appeared
observational
to possess studies
environment
many
can be readily or by initiating
intervention
and control
In the environment and ovens,
preparation
items within
this type of skilled
to be taking
contemplative
What cook's
First, there
behavior
allows
some
activity.
arose
human-machine
working
because
I have made
busy rush periods
skilled
I have observed
based on the external
selection
based
internal
representation
performance
in my more
at a
in this
limited
24 hour access
And by making
to this
a well timed
limited)
itself automatically and is always
of automated
capacity
action
though,
of intimacy
for
of unpredictable
and in parallel, veridical
maintenance."
(also
serves
capable
Nado,
and
to engage
in
dynamically,
events.
is visible.
Note that the The most
one: the grill area itself.
economies
as compared
The environment of being
as an external
see Reitman,
with the world
to be productive?
of the environment. memory
the following
and arrive
all of this structure
has considerable
as an external
overwhelming.
if ever does the cook appear
for the cook to use is an external environment
simultaneously.
perception-action
contact
are uncertain
such as
the
appear
in the cook's
of interaction
but nearly
orders
and I believe
tight perceptual
Rarely
mode
multiple
and the task demands
interaction, degree
devices
at the grill, to coordinate
while preparing
to a number
representations
serves
activity
Task demands
structured,
selection
care of its own "truth
for rational
intensive
representation"
consistent,
each order,
must be sensitive
"problem
updates
cooks
of a cup of coffee.
manual
sort of action.
cognitive
is highly
problem
Such laboratory
the past two years
and also because
The cook maintains
this perceptually
upon
settings,
over
the cook uses an assortment
is an intensive
environment
efficient
situation
human-environment
with the environment.
and ongoing
time so that
are made available.
day in both complex
of short-order
with substantial
Observe
detached,
the necessary
practice
with the cook, one even has an (albeit
I have studied,
this task in any formalism
seems
However,
over task demands.
Describe
always
every
for the price
conversation
of the many
interaction
There
and enough
As an example,
operational
secured
combined
will be apparent.
conditions
of the same properties
in complex
food order
fryers
this issue.
task solutions
are performed
fairly extensive
setting
behavior.
conducted.
"experiments,"
systems
of skilled
to clarify
and perception-action
are rarely
These
is needed
rich enough
both cognitively-intensive experiments
research
aspect
the
perceptually
memory
and Wilcox,
Action
to action considered
as a
accessed,
store, is internally 1978).
The world
takes
15 When
uncertainties
information through
do occur
underspecifies
using such an external
constraints
on activity),
perception-action
rather
of a steak?
Flip it and see.
underside
the environment, solutions
and thereby
of meats
external
memory
information
in order
of orders
"displays"
variable
to covary
flipping
meats
doneness
And
reduce
cognitive
of doneness,
with a visible
so that the doneness
of the underside.
not only structure
How
well cooked
uses the structure
is the
already
present
in
the cook may organize
or plates 1988),
to serve
environment
the cook may adopt
of the top side can always
generate causing
the strategy
new
a hidden
of continually
be used as a reliable
the cook is both performer
the
as a temporary
and may even
in the controlled
In a very real sense,
can often be resolved
For example,
(also see Beach,
For example,
available
in order to make perception-action
burdens.
constraint
one.
knowledge.
may lay out dishes
to be prepared
by introducing
stored
create
(i.e., perceptually
these uncertainties
the cook
he or she can dynamically
available
placement
than accessing
representation
indicator
and on-line
of the
interface
designer. Skilled
human-environment
structure
as well as a source
environmental similar
interaction
structure
cognitive
created
efficiencies.
organization
enablement displays
used in toasters self-produced
and ovens
processing
structured
demands
the acquisition
information.
in promoting
A model
itself,
contributed economies
Because
through
resist
mechanisms
of skill acquisition
and thus increasingly
and
of the possibility
internal
how the actor's
the
mechanisms
skills will always
efficient
and
at one
memories
role to the timing
of such situated of more
in the static
the external
to the wor/d.
also have to describe
by activity
in cognitive
cognitive
a role
and is thus short-lived,
whether
For example,
plays
in dynamic
external
informative
for
environment to the actor,
of skill development.
of the responsibility
action
mechanisms
skilled,
dynamic
perception-action
to the task.
of the development
would
itself
"on-line"
result
The
strategies
process
of structure,
by the cook,
memory
structure,
interaction
increasingly
over the course Much
to offload
set of environmental
human-environment becomes
But both forms
solutions
solely in terms
a fixed
happens
by the cook play a similar
environmental
description
mereIy
exploited.
and routinized design
to environmental
of the grill area and is thus reflected
or continually
created
own "tricks"
structuring
of the design.
to be subsequently
the environmental
the design
of perception-action
dynamically
faithful
through
during
point in time by the designer
structure
by the cook's
The former
while the latter happened permanent
of environmental
created
to the structure
of this type is thus both a response
for dynamic
at the interface interaction solutions
relies
between
human-environment the performer
upon abilities
to tasks,
and where
to exploit
interaction
lies in the perception-
and the world.
The development
environmental
structure
none naturally
exist,
to create
to obtain
additional
of
of
16 environmentalstructurein sucha wayasto enableperception-action solutions.If suchstructureis not providedby thedesigner,theperformerwiU seekto createit throughactivity thatintroduces newformsof structure.Theproductivityof thismodeof behaviorrequirestheavailabilityof sourcesof informationto specifytheenvironmentalconstraintsto which behaviormustbe sensitivein ordertobeeffective,andtheavailabilityof actionscapableof bothchangingthe environmentandof creatingadditionalsourcesof informationto furtherenabletheperceptual guidanceof activity. Theseare,I believe,featurescommonto nearlyall environmentsin which the acquisitionof fluent,dynamicinteractionis observed.They arealsofeatureslackingin themany technologicalenvironmentsof interestto cognitiveengineering,duelargelyto interfacesthathighly restrictperception-action accessto thecontrolledsystem.Theabsenceof suchfeaturesis one majorcauseof thedifficulty of acquiringskills in suchsystems,andthereasonthattheendstateof learningis oftenoneof only partiallyeffectiveadaptation. TOWARD The previous psychological
discussion models
methodological above.
Many
identified
are suggestive
by Brunswik's
particular,
Brunswik's
and modeling, Hammond, specification cognitive which
1955),
on taking
and Gibson's
engineering
might offer
familiar
with Gibson's
for cognitive views
to cognitive
engineering
environment
interaction
The ecological
(1966,
theories
1979)
as purely
of skilled
(Brunswik,
nicely
interaction
of In
as the unit of analysis 1952;
can be described
blend
was
explanations.
system
identifying
to support
design.
it is important First,
perceptually might
a
1956;
as perceptual
with the claims
with the world,
that
and models
as the unit of analysis.
modeling
especially,
approach
of how knowledge
framework
(1979),
outlining
that have been
on psychological
interaction
initial steps toward
does not require.
by Gibson,
engineering.
on activity
for
to human-environment
the human-environment
system
I will take some
models
approach
in the Lens model
constraints
conditions
of the gaps in our knowledge
for psychological
constraints
is most in need of models
approach
the some
focus on how fluent
take the human-environment
as discussed
important
best represented
of environmental
In the following
action,
can provide
of necessary
It is time now to turn toward
that an ecological
and Gibson's
emphasis
perhaps
features
tools for cognitive
(1952)
a number
design.
to address
of the possibility
pioneered
structure
environmental
of the necessary
may yield fruitful
PERSPECTIVE
on identifying
with the potential
interaction
environmental
ECOLOGICAL
has centered
to support
strategy
discussed
AN
surely
opportunities However,
to first discuss
it does not require guided
activity.
be possible
and for readers
what an ecological
that we conceive
Direct
although
the ecological
perceptual
it is likely
already approach
of all human-
guidance
of
that it is specific
to
17 thoseinformation-richenvironmentsin whichperceptionevolvedor to artifactualenvironments designedto mimic suchenvironments.Thereis noreasonto expectthatevolutionanticipatedthe modernaircraftcockpitor the wordprocessor.In suchenvironmentstheneedfor post-perceptual processes suchasproblemsolvinganddecisionmakingis quite likely. The ecological and information
processing
both contribute Second, studying
fluent
human-machine
arguing
for using
These
in which looks
environments,
much
(1952)
be described
Hammond,
The relations validity. judgment
the other
side.
the organism
of them,
the use of abstract often
found
engineer.
are the places
before
in him, was
that are representative
And for better
fashion,
tO
as a description using
or worse,
terrestrial
where
of
a cockpit
or
environment.
For this reason,
these
the ecological
represents
1975).
The organism
in which forms
concepts
This framework and environment
As shown
has available
criterion
approach
and vary in cue utilization. in that each concept
has a number
1 about
here
mutually
(xi's) which
bear
diagnosis).
and vary in ecological
use of the cues
(rs, i's) to arrive
at a
The framework
is an expression
of
properties
terms.
is a
structure
(e.g., a medical
on one side of the model
of attractive
Insert Figure
a set of cues
forms
(e.g., see
1, the Lens model
and cognitive
to be judged
makes
and environment
concepts
in Figure
may take various
the organism's
in compatible
of parallel
how both environmental
an environmental
the ways
of how the human
the principle
the the cues and the criterion
of parallel
study
to
out.
model
performance.
may take various
the principle
occurs.
a commitment
like Brunswik
than it does the natural
and Steinmann,
which
(re,i'S)
between
Similarly,
be carried
the Lens
Brehmer,
to judgment relations
Gibson,
to the cognitive
abstractions
in an integrated
framework
contribute
against
the types of displays
sense,
of interest
imply
Modeling
offered
Steward,
symmetrical
made
rallied
as the basis of scientific
of interest
should
for Cognitive
Brunswik
specified
environments
or carefully
Gibson
perception;
more like a laboratory
interaction.
does not necessarily
But in a larger
behavior
to be at odds, but may instead
of human-environment
environment.
conditions
a target
engineering
Resources
could
systems.
environmental
are the target
to cognitive
in the natural
be considered
approach
for the study of visual
existing
room
understanding
of an ecological
behavior
displays
the conditions
need not always
to a more complete
the adoption
information
control
approaches
that result
has a counterpart from representing
on
18 Perhapsmostimportantly,theLensmodelframeworkallowsthe modelerto measurethe degree to which theenvironmentalstructurewhichrelatesthecuesto the criterionis reflectedin the mannerin which thecuesarecognitivelystructuredto produceajudgment. High levelsof achievementareanindicationof a highly adaptive cognitive organization; i.e., a cognitive strategy that mirrors previous
comments
activity well.
the environmental
is likely
in this chapter
to be required
In addition,
behavior
of the cue structure
systems.
Did a particular
We often
depending
upon
the answers
the Lens model
representing
skilled
also need models more complex deficiencies.
cognitive
interpretation we go about
abilities
information
and actions
perceptually
guided
about
confronted
can (and should)
my previous system
with result
(facts
information
Note that these environmental as being
properties
Like formalisms
not to represent selection
However,
Second,
would
be
this in how
model
as suggested
(p. 272).
(facts by
Judgmental specifies
the
to gain more diagnostic are exactly
impediments
for rational
for
is a epistemological
only probabilistically
variables
the major
demands
we
has two important
the environmental
of last resort"
for
suggest
and we must be careful
the performer).
environmental
described
to identify
to action
to be judged.
difficulties,
activity
the available
comments
the state of the world,
the Lens model
about
an approach
as the unit of analysis,
in order
in order to keep distinct
is a "cognitive
interaction.
of displayed
The Lens model
as the criterion
of manipulating
I have previously
of human-machine
for developing
behavior
represented.
of conceptual
model
upon when
are not available.
task environments of fluent,
judgment
capable
action
validity
of the reasons
operator
the Lens model
applying
itself to serve
this interpretation
will only be called
criterion,
skilled
judgments
A first step toward
and the cognitive
et al. (1975),
point
However,
this perspective,
can give rise to a number
the world),
Hammond
fluent,
itself is not explicitly
opportunity
formulating
a good starting
interaction.
of modeling
forcasters,
such as these.
offers
From
Weather
usage
adapted
of remedial
as
of both the ecological
incorrect
perfectly
interaction
or both.
and analysis
making
types
view of cognitive
an understanding
that take the human-environment
activity.
are selected.
an action
different
of representing
action
in the design
of the
of less than fully productive
structure,
can yield
an operator
to questions
models
for the purpose
how actions
Quite
the causes
but only an analysis
of a potentially
framework
capable
First,
framework
about
from
As many
human-environment
the cognitive
interests
human-environment
that we not only require
to allow
error result
information?
skilled
for cue utilization
have similar
must be sensitive.
sort of adaptivity-oriented
one to localize
structure,
or was the error the result
not fully diagnostic
Thus,
allows
and their policies
errors.
information,
some
err in their predictions,
for these
behavior
in order to understand
the environmental
frequently
to which
suggest,
the Lens model
to either
for example,
structure
action,
those
features
of
to the development then, it may be quite
19 possibleto interpretthe skilledselectionof actionwithin theLensmodelframework,however, suchamodelis not likely to capturethosespecialfeaturesof taskenvironmentsthatallow for the acquistionof fluent human-environment interaction. The need to explicitly promote
it, suggest
performer
be described.
proposed
perception
results
behavior
the performer.
decision
space descriptions
resulting
from problem
environment, change
solving
particular
position
most concerned affordances,
concerning
with those
capable
perception-action
affordances
in order
of the world
environmental
structure
For our purposes,
in terms
space
the emphasis
of an affordance
for the two deficiencies
the possibility
judgment
that perceptual
perception)
suggests
However,
structure
Regardless
behavior
is still a valuable
Gibson
was
to specify of
in other
situations
or information
must be taken into account
or to combine
actions.
in more elaborate
information
specifying
of the type of either
to an affordance
structure,
tool in understanding
a
how
and behavior. space
is especially
of the Lens model
to the identification information
will
one to any
detection
or impoverished,
can in some cases play the role of the criterion
from passive
(i.e., direct
in cognition
of
of the
for action
is available
may have to engage
structure,
to orient
of affordances
is reflected
the concept
required
description
activity.
as the perceptual
affordance
In such cases, performers
activity
to guide
opportunities.
to select
from theories
does not commit
information
may be restricted
in order
akin to the
descriptions
resources
description
can be described
present
space
it
will result.
perceptual
to action
space,
will be a dynamic
may be detected
to detect the affordance
to compensate
affordance
space
the immediately
or post-perceptual
description
Second,
behavior
with other information
perceptual
resources
interaction
to be productive.
activity
in which
for action
resulting
and the performer's
space environmental
situations
the problem
space
structure
to the environment
other than that specifying for behavior
affordance
of orienting
access
structure
how affordances
and in such cases
information
cognitive
an affordance
as both the environmental
an affordance
an affordance
or the cue space descriptions
In most cases,
Note that creating
theory,
how
An action-oriented
of the opportunities
can be called
from decision
theory,
over time and a dynamic
in terms
of affordances
to understand
for action.
that
of the skilled
and theory
in order
opportunities
description
resulting
physics
description
of the environment
of the world
of how the environment
with his ecological
environmental
The resulting
judgment.
as well as those features
the problem
to environmental
in a description
presents
probabilistic
interaction,
Gibson,
an action-oriented
may orient
approach
f/uent
that we must consider
should
(1979),
represent
that we must relax
identified
because above.
in the Lens model,
and selection
is capable
important
the a priori
assumption
First,
an
thus shifting
of opportunities
of fully specifying
it provides
for action.
the affordance underlying
the
20 Lensmodelthattheavailableinformationis onlyprobabilisticallyrelatedto thecriterion. The questionof determiningtherelationshipbetweentheavailableinformationandtheenvironmental affordancestructureis anempiricalone. The ecologicaltaskanalysisframeworkpresentedbelow is anattemptto integrateBrunswik'sLensmodelandGibson'saffordancetheoryinto a unified frameworkfor modelingskilledhuman-environment interaction. The Framework Integrating
concepts
the ecological framework system.
for Ecological
Task Analysis
from Brunswik's
task analysis
framework
is a symmetrical
symmetries
uses a principle
between
of parallel
cognitive concepts
is evident
perceptual
and the lower
which
represents exploit
to suggest
surface
task analysis
model
structure
between
begins
of environmental
structure
the world,
and the environmental
depth
structure,
so to speak. are models
of environmental
of surface
description
in the Lens model. available
perceptual
structure
controlled
system.
structure surface action
perceptual
structure
structure
the Lens
structure
of surface
model
however,
ecological
of the environment.
as the actions
made
available
of environmental
between
in the middle
shown
and environmental
structure
systems,
which
a similar describes
captures
about
controls.
structure.
The
role to the cue the
the state of the
only the surface
also requires
In a human-machine
of the
one can think of surface
displays
task analysis
and
the surface
action
plays
a distinction
the performer
structure,
perceptual
of the
we rely upon
behind
task analysis
by interface
and action.
descriptions
framework,
from interface
framework
also
here
In human-machine available
model
to capture
framework
exists remotely,
structure
in ecological
The model
as the information
of the environment, action
perceptual
information.
Unlike
of surface
concepts
structure.
at the interface
which
in
the proposed
perception
of two complementary
exists
structure
The two descriptions
description
environmentally
which
results
human-environment
between
action
2 about
model,
the proposed
the upper
As in the Lens model
surface
diagram,
symmetries
theory
of parallel
structure,
between
with the creation
of the environment.
environmental
the integrated
the principle
certain
in the relationship
affordance
2. Like the Lens
and environmental
Insert Figure
Ecological
and Gibson's
in Figure
task analysis
This latter symmetry structure,
model
depicted
arrangement
Not only does ecological
certain
Lens
system,
perceptual
a description
of the
one can think of surface
21 In theLens structure
model
framework,
is the relation
person
is attempting
action
structure
between
the readily
to judge.
and depth
environmental
the relation
change
Similarly,
structure
the person
between
available
is the relation is attempting
the difference
between
given
distinction
reflects
the difference
between
readily
of environmental
potentially
covert
is a description
depth
structure
relationships,
actions
depth
of cognitive
structure
actions
actions.
2 is a description
is reflected
and the
side, the surface/depth
and intended
structure
surface
side, the surface/depth
On the action
on the left side of Figure
and the model
of how environmental
available
state a
between
available
On the perceptual
and inferred.
and depth
the relation
the readily
to effect.
structure
and the environmental
framework
between
reflects
perceptual
information
in the proposed
distinction
model
surface
The
of these
on the right side of the figure
in cognition,
and ultimately,
behavior.
The Principle
of P_II¢I
The description
Concepts:
of environmental
necessary
to capture
perceptual
guidance
specifying
environmental
of action covert
or remote
form,
of activity
depth
process
identify
the congruence
perceptual
action
between
is suggestive
describing
structure
description
environmental structure.
the analysis
a performer structure
model
namely
Various
in detail
of perceptual
perceptually
forms
type of necessary
guided
for cognitive of mismatch cognitive
structure
task to
between
Mismatches
to overcome
are possible: activity,
is to
another
Matches
activity.
activity
some
the ecological
surface
description.
of fully
of action
to specify below,
for
guidance
to infer
guidance
and action
of environmental
of demands
information
perceptual,
As will be described
capable
perceptual
the perceptual
is
The possibility
information
framework,
of surface
structure,
for fluent,
is possible.
of perceptual
Rather,
and action
the
each form of
and a different
type of
solution. process,
must be discussed.
using
envirornmental
of action.
of a different
design
of activity
detection
these two forms
are indicative
of both perception
of the proposed
the models
opporunities
non-specification
interface
structure.
by examining
these two models
Before
In terms
of the environment.
action
indicate
in terms
guidance
as the perceptual
property
proceeds
these two models
remedial
structure
Structure
relies upon the availability
affordances.
the surface
analysis
mismatch
surface
as the use of one form of surface
namely,
between
and Action
how the skilled perceptual
is not to be described
be described
Perceptual
reflects
It is easy to overlook
The model
s perceptual
is described.
the symmetrical of surface
capacities
This description
both the perceptual
of the models
perceptual
structure
as a frame
of reference
is relational
capacities
that such descriptions
nature
are actually
relational
and
is an environmental in which
in the sense
of a performer
of perceptual
the
that the resulting
and the environmental in nature,
as we often
22 speakasif a perceptually-generated differentiationof theenvironmentis purelya functionof the environmentalstructureandnot afunctionof perceptualcapacities.However,the factthatsome formsof environmentalstructureareseenasobjectswhile othersarenot,or thefact thatsome aspectsof theenvironmentareseenasbeingbluewhile othersarered,areasmuchfactsaboutthe perceptualsystemastheyareof theenvironment.A perceptually-oriented environmentalmodelis a relationalconstructcreatedby usingperceptualcapacitiesasaframeof referencein which environmentalstructureis measuredanddescribed. The secondmodelof environmentalsurfacestructurerequiredfor ecologicaltaskanalysisis an action-orienteddescriptionof theenvironment.Themodelof surfaceactionstructureis an environmentaldescriptionusinga performersactioncapacitiesasa frameof referencein whichthe environmentalstructureis described.This descriptionis relationalin the sensethattheresulting environmentalmodelreflectsboththe actioncapacitiesof aperformerandtheenvironmental structure.This action-orientedenvironmentalmodelrepresentstheworld in termsof its opportunitiesfor action. This environmentalmodelthusgenerates a differentiationof theworld in termsof thedegreeto whichvariousspatiotemporal environmentalregionsareconsistentin variousdegreeswith thetakingof variousactions.An action-oriented environmentalmodelis a relationalconstructcreatedby usingactioncapacitiesasa frameof referencein which environmentalstructureis measuredanddescribed. The symmetricalnatureof theperceptually-oriented andaction-oriented environmental descriptionsrequiredfor ecologicaltaskanalysisshouldbe apparent.Thesetwo environmental modelsdiffer only in thatoneis relativizedto thefunctionalityof the performersinputmechanisms while theotheris relativizedto thefunctionalityof theperformers outputmechanisms.Thetwo modelsthusreflecttwo differentprimitivedifferentiationsof the environment,onegeneratedby usingperceptualcapacitiesto understand how theworld is carvedup with respectto perception, andtheothergeneratedby usingactioncapacitiestounderstandhowthe world is carvedup with respectto action. As BarwiseandPerry (1983,p. 11)havesuggested, Theemphasisis onhow theorganismdifferentiatesits environment,on the sortsof uniformitiesit recognizesacrosssituations.Differentorganismscanrip the same realityapartin differentways,waysthatareappropriateto theirown needs,their own perceptual
abilities
and their own capacities
the structure
the environment
with respect
to the environment
independent
of the organism
organism
reflects
properties
displays
for action.
This interdependence
to the organism
is extremely
and the structure
important.
For while
s individuative
activity,
the structure
of the organism
itself.
(My emphasis)
reality
between
of the organism is there,
it displays
to an
23 Neithertheperceptually-oriented environmentaldescriptionnortheaction-oriented environmentaldescriptionresultsin a moreprimitive,privileged,or objectiveontologicalpictureof theworld. However,theclaimis sometimesmadethatanaction-orienteddescriptionof theworld in termsof affordancesis scientificallyillegitimatebecauseit is relativizedto theactor,andis thus in somesensesubjective.Notethatthecorrectresponseto this possiblecriticismis not to adoptthe heroicpositionthataffordancesarein somesenseindependent of theperformer: theyarenot,they arisefrom usingtheperformersactioncapacitiesasaframeof referencefor environmental description.Rather,to counterthis argumentonemustmerelyemphasizethatthe supposedly scientificallylegitimateperceptually-oriented environmentaldescriptionswhich portraya world of objectsandpropertiesarejust asrelativizedto thecapacitiesof theperformerasareaction-oriented environmentaldescriptions.Onemerelytakesperceptualfunctionalityasthe frameof referencefor environmentaldescription,while theothertakesactionfunctionalityastheframeof reference. The Process
of Ecological
I shall call a process are created models
whereby
and mismatches
are identified
What I believe now be stated
quite simply.
human
and environment
(1990)
rational
cognitive ecological
which
Before
task analysis
specify
ecological
attempts
outlining
to define
the analysis
task analysis
is to define
processes
that may be required
Anderson
s rational
analysis,
then,
necessary
cognitive
activities
might be carried
common
perspective.
be its primary cognitive
concern,
activities
to support
the present
However,
ecological
whereas
the current
such as memory,
effective
functionality
a few comments
the
and by for effective
computational-level and Anderson of a variety
comparing
s
of
the goals
of
may be valuable. functionality
leaves
of any cognitive
interaction.
aside the question
In this much
formulation
categorization,
s (1982)
human-environment
task analysis
can
between
interaction,
of vision models,
the necessary
approach out.
or artificial) fluent
in spirit to Marr
in particular
system.
modeling
will be necessary
the necessary
s approach
to cognitive
(natural
processes
process,
in the two
of a human-environment
underlying
functionality
structure
of a human-environment
task analysis
what cognitive
and action
represented
approach
an interface
is thus similar
and Anderson
The goal of ecological
task analysis
to which
perceptual
differentiations
of an ecological
to def'me the necessary
analysis
of surface
with the principles
task analysis
attempts
models.
the degree
is consistent
An ecological
which
an ecological
contribution
helps
models
the environmental
A preliminary
to identify
so such an analysis
theory
between
to be the central
is required
behavior.
environmental
and described,
system
doing
Task Analysis
the two approaches
takes human-environment of rational
and problem
analysis solving.
Like of how any share
a
interaction
is concerned
with
But as Anderson
to
24 himselfhasnoted,all thesecognitive of action
selection,
a problem
solving
words
the nature
and for Anderson, or decision
Anderson
abilities action
making
uses to describe
of the environmental
are useless selection
exercise
is to always
(p. 192).
the processes models
if they do not in some
that result
to be understood
Perhaps
underlying
way serve
action
as the result
it doesnt
much
selection,
but what does
from such a choice
(decision
the goal
matter
of
what matter
is
trees and problem
state spaces). But as our previous intensive
processes
comments
for explaining
focus not on the sufficiency Ecological
in any particular mechanisms
manner,
task analysis
allow
for a description
when
an examination
environmental
The Principle
above,
surface
below.
The result structure,
important structure
task analysis
and action
begin
after creating
problem
state space, decision
structure
to cognitive
structure
process
however, model
tree, linear
processes rules).
of perception
of the environment
guide
possesses
cue-criterion
(e.g., heuristic Rather,
requires
search,
a certain function), comparative
we let the examination
the selection
the bridge
of the model
in
between
and process.
structure.
in detail
It is very depth
That is, we do not depth
structure
(e.g.,
and thereby,
a corresponding
evaluation
of alternatives,
of the surface of depth
with
of the environmental
the environmental
intrinsic
the
the congruence
will be discussed
structure
of
the environment
of a model
and action
selection
of the models
congruence
that modeling
Only
Structure
they differentiate
cognitive
that
activity.
by then examining
is the specification
task analysis the models
in which
models
of activity.
intensive
provide
is governed
in perception-action
with the creation which
of their
interaction.
of action
and Comaitive
must
sorts of governing
mismatches
proceeds
of internal
however,
selection
guidance
in a construal
of examining
that the task environment
linear cue combination
begins structure
The process
and a complementary
result
Depth
in terms of the manner
of this exercise,
by assuming
surface
The analysis
and action.
action
or any other cognitively
ecological
to note that ecological comes
solving,
design,
the use of environmental
indicates
task analysis
into one of
the conditions
what
of the perceptual
models
Environmental
these two models to perception
in terms
Concepts:
structure
respect
depth
problem
and the environment.
between
selection
from cognitively
of human-environment
to meet this goal through
will ecological
of Parallel
on limiting
to define
any instance
of these environmental
making,
As mentioned
human
of action
to support
not by assuming
has as one if its goals
for modeling
attempts
structure
terms of decision
perceptual
but rather
starts
resulting
can be rationalized
Modeling
but rather
therefore,
whether
and action,
behavior.
of these frameworks
task analysis,
selection,
perception
productive
will be necessary
Ecological
all action
or from more efficient
these frameworks
necessity.
suggest,
perceptual
structure,
and action
and thereby
the
25 correspondingstructurefor cognitiveprocesses.Hereinlies themajordifferencebetweenthe proposedapproachandmanycurrentmodelingapproaches in cognitivescience:considerationsof perception-action functionalitydefinethenecessaryfunctionalityfor cognitiveprocesses, rather thandefiningperception-action functionalityby an a priori cognitive model of action selection. I have tried to make the case for sequencing points
in this chapter.
We are obviously
corresponds
to, and helps make
serve
selection.
action
however,
Which
is available
cognitive
the performer,
processes
interaction
best describes
the processes
environmental
models
environmental
structure
is reflected
environmental
and cognitive
limitations
without
Yet I wonder doing
realistically action
selection
environmental
behavioral
purpose
processes
format.
with perceptual models
applied
and action unlimited
the world capacities,
a variety selection
in terms
might
on the other hand,
requires
capable
of the world
of preserving s perceptual
however,
and a between
behavior. necessary
for ecological
the content and action
structure models
of any
that rationalize
low-dimensional are advantageous
be organized
of how the
in a context-free,
general
environmental
structure
of raw
has the potential
to create
environmental
my own view is that any method
many
in a setting
of the fine details
and action
structure,
task
of these descriptions
and are thus suggestive
interaction
the
any ways in which
representations
of the interaction
different
Note,
the congruence
is that relatively
of contexts
skilled human
complexity
The richness
solving
Nevertheless,
underlying
structure.
of the perceptual
dimension.
the mechanisms
about
Such low dimensional
action
identifying
models
of the models
that
ways in which
structure
One great allure of environmental
or problem
across
underlying
Describing
of almost
situation.
depth
of the
model
As a result,
to identify
upon productive
to get more precise
can be used.
in that they can be easily
in order
they make
the cognitive
in the two cases.
are selected,
description
to the richness
making
descriptions
psychological
schematic
injustice
as decision
processes
designs
the best description
the different
of environmental
can be examined
if it is possible
a potential
complex
cognitive
place constraints
a highly
and behavior
to
perceptually
demands
interface.
to describe
that
behavior,
(control-display)
that provides
with another
once a model
models
or biases
This is obviously analysis.
of internal
interface
may be quite unlike
in the two cases in cognition
to which
structure
be engaged
for productive
of the cognitive
model
interaction
framework,
model
Different
of previous
depth
that will actually
of the degree
in terms
with one interface
underlying
corresponding
action.
that the cognitive
will be needed
that like the Lens model
activities
analysis
at a variety
of environmental
will be necessary
can differ radically
with the result
underlying
cognitive
productive
in this fashion
for a model
activities
by a detailed
to specify
for the same task environment upon
looking
sense of, the cognitive
can only be determined
information
the analysis
capable
of
of any reasonable
of the environmental
seemingly
necessary
for the
26 fluentoperationof theperception-action system,severelyovertaxesour highly limited cognitivelinguisticresourcesfor environmentaldescription.I doubtthatatthecurrenttime a taskanalysis techniquecapableof guidinghuman-environment interactionmodelingcanpossiblybemuchmore thana chargeto themodelerto undertakethelong andarduousprocessof identifyingand describingthepotentiallyoverwhelminglyrich interfacebetweentheperformerandtheworld. For thisreason,thefollowing descriptionof the analysisprocesswill consistof an abstract discussionof the possibleresultsof anecologicaltaskanalysis,alongwith a concreteexampleof how suchananalysiscanbeperformedfor a particularbehavioralsituation.Figure3 depictsfour possibleresultsof anecologicaltaskanalysisin termsof thecongruenceof theresulting perceptually-oriented andaction-oriented environmentalmodels.The grid linesin eachof the schematicenvironmentalmodelsindicatethemannerin whichtheenvironmentis spatiotemporally differentiatedwith respectto eitherperceptualcapacitiesor actioncapacities.The four caseswill be describedseparately. InsertFigure3 abouthere
Case I: Perceptual
overspecification
In the first case shown respect
in Figure
to the environmental
different Object
perceptually
situations
displays
of action.
perceptual
so it becomes
perform
this function
Object
used,
salient
to the differentiation
however,
Perceptual
likely that a significant identify
consistencies
relational
in the action
spaces,
naturally amount
to reduce
emerge space.
some
although
to enable
supported
nor supported
of categorical
ability
for coping
space.
a display
to overcome
naturally
through
or instance-based to action.
action.
the dimensionality
perceptual
When
learning Note,
of the
These
displays
information
in
in a manner
based
solution
is not
perceptual
occurring
result
display
for pattern design,
may be required however,
in an
relational
the possibility
configural
Many
with perceptual
that could potentially
this solution.
from perception
for productive
action
When
some
action.
that are differentiated
is one process
available
in the mapping
available
over-differentiated
features
and action
must be perceptually is neither
the originally
recognition
properties
solution
with
for productive
opportunity
with the lower dimensional
pattern
of the perception
is over-differentiated
of opportunities
are an attempt
will have to develop
alignment
recognition
displays
reflected
the performer
overspecification.
in terms
all point to a single
aligned
by organizing
such a way that perceptually identical
environment
are one type of design
overspecification space
3, the perceived
differentiation
distinct
or configural
of action
it is
in order
that there is still
to
27 thepossibilityof fully productiveperformancein all thesecasessinceperceptionmerelyoverspecifiesaction,it doesnot misspecifyaction,asin CaseIV below. Case II: Perceptual
underspecification
Here the perceived differentiation
environment
in terms
information
in order
perceptual
to uniquely
in memory,
information
in order
the displayed such cases. rather
specify
about
specify
effects
of an action,
simulations
are two approaches
performance,
however,
Internalization
Simon
effects
physically
trying
state without
possible.
capacities
of the cognitive
of an action out
solutions
the use of an internal
subtle)
In this situation demands
model,
differentiation Perceptual
to which
specification
of action
fluent
or conceptual
performance learning.
learning
can be expected
The performer
state,
of action
or fast-time Unaided
may be necessary
solving
models
into account
the dimensionality
information
the
or about
may not always
Here,
be
the perceptual
of the perceptual
into alignment
to develop
when
and
initial perceptual 1969).
can eventually
in
of Newell
behavior,
this form of activity
s pre-established
future
to the performer.
of this type, however,
perception
in
displays
Exploratory
to overcome
of the environment
performance
of the downstream
model
(e.g., E.J. Gibson,
in order to increase
into
that may be necessary
for taking
although
this
skill can be acquired.
The problem
that is available
of stimulation
Perceptual
before
keep
memory
about
a number
Predictor
available
activities
available
differentiation
change
differentiation.
dimensions
Case III:
learning
is perceptual
as to bring the perceptual
apparent.
must be taken into consideration. is a method
for aiding
knowledge
in the form of an internal
as descriptions
of the performer
environmental
dynamics
Building
identifies
requires
learning
state information.
A form of perceptual
underspecification
considerable
in the environment;
is that information
this information
to past or future
can be viewed
downstream
action
of
with the perceptual
alternative.
Perception
cause
of the environment,
is one strategy
are not perceptually
for making
will require
of environmental
order to gain access
available.
of the appropriate
and these effects
action
underspecification
than past state, is not perceptually but the selection
history
available
One likely
information
information
displays
perceptually
alternative.
the previous
the appropriate
of perceptual
to the environmental
not enough
action
this memorial
using trend or historical cause
is simply
is that there is hidden-state
and then integrate
interface
There
with respect
the appropriate
something
to uniquely
Another
candidates,
future
action.
of action
must know
information
is underdifferentiated
of productive
underspecification
i.e., the performer
of action
space
so
with the action-oriented requires
become
without perceptual
(perhaps
initially
sensitive.
significant
cognitive
competencies
provide
28 theability for uniquespecificationof productiveaction. No rule-basedinformationintegrationis required. Neitheris informationaboutthehistoryor futureof theenvironmentnecessary. Therefore,no internalizationof environmentaldynamicsis requiredtoprovidetheseformsof information. If it werethecasethatthedesignercould alwaysbecertainthatthe perceptuallyorientedandaction-orientedenvironmentalmodelsof thehuman-environment systemwerecorrect, interfacesthatsupporttheperceptualspecificationof actionwouldbeour undeniabledesigntarget. However,if unanticipatedchangesoccurin eitherthediagnosticityof the displayedinformation, thefunctionalityof theinterfacecontrols,or theenvironmentaldynamics,a fluent,informationally encapsulated perception-action modeof controlmaycarryonwithouttheperformerpayingheedto theseenvironmentaldisturbances. (_Se
IV:
Perceptual
misspecification
In the final case the mapping action
is unruly.
perceived endow
Behavior
situation
misspecification, for finding
might well be productive
invariance
missspecifies
in aggregate
models.
action
do think it is important
misspecification,
that such cases
that the performer
of reasoning
decision
making
(Kahneman
solving
(Kotovsky,
Hayes
attention
to the concrete
models.
There
environment
that a given
productive
knows
(e.g., Wason and Tversky,
presentation
is always
really defined
model
the modeler
1985)
structure
some
1979; Tversky
The long history
no environmental with respect
fails to represent,
situation
that people than do many
stuff that is context to a model;
it is simply
and as a result,
those
I
upon perceptual does not.
I suspect
of findings
abstract
on the 1975),
and problem
pay considerably
more
cognitive
as opposed
aspects
I
to be a case of
1981),
those
with
However,
Johnson-Laird,
and Kahneman,
all demonstrate
of a problem
based
1972;
to be a
these variables.
of a catch-all.
things that the modeler
and Johnson-Laird,
cognitive
can be described
among
selection
opt
that perception
finds what appears
action
and instead
can be considered
under the assumptions
than we may care to admit.
and Simon,
is, of course,
Context
that when
perceptual selection
judgement
it is something
typically
of probabilistic
the covert relationships frankly,
but yet observes
are more frequent
context-sensitivity
structure.
this case because,
of action
depth
the currently
in such situations
the construction
of human
framework
indicating
to emphasize
it is likely
through
and productive
but not because
to overcome
account
and that the environmental
correlations
information
of behavior
of knowledge
performance
task analysis
(judgment),
will have little to say about
information,
Models
amount
available
in this situation,
In fact, the Lens model
and cue-criterion
perceptual
informative.
with a considerable
case of the ecological
cue-cue
the perceptually
or else give up hope for a deterministic
and environmental special
between
is particularly
the performer
of _lction
aspects
to relevant of the
that are rendered
29 incapableof producingbehavioralvariance.Findingswhich demonstrate the intensivecontextsensitivityof cognitionandbehaviorcanbeseento be,in part,a reflectionof thefact thatmany currentenvironmentalmodelsareeitheroverlyabstract,or perhapsevencut acrossthe grainof the perceptually-oriented andaction-oriented environmentaldifferentiationsthatarethebasis of ecological
task analysis.
to complex attention
or probabilistic
to advance
Jagacinski, displayed
modeling
(Sheridan,
provide
and a mixture typical
subjects
control
implementing automation
activities in many
mappings
out as much
cognitive
that give rise
by increased
modehng.
to engage
for interface
systems
apparatus
modern
consisted
operator
and the need to design
Miller
controls,
systems.
world.
was motivated
similar
The selection and
by the introduction
to support
this form of
interaction. The experimental use manual
apparatus
(joystick)
over four additional designed upon
to which and enemy
activity craft)
for this purpose. i.e., to engage base.
Subjects
missiles,
or automatic
cargo
piloted
was confined.
The scout s major
Subjects
the world.
The scout
used the additional
both stationary
and mobile
capacity)
to successfully
mode
levels
of performance
of dynamic
interaction
activity
a text editor
the partially
was to discover equipped
management
objects
shown
mile area (cargo radar
objects,
and unload
it at a home
constraints
(e.g., fuel,
of practice
were required was quite
of much
world
the discovered
each 30 minute
characteristic
of control
specifically
100-square hidden
to act upon
of action
mode
could
with a 1.5 mile radius
complete
the selection
subjects
forested
the entire
craft and to load cargo
of resource
hours
using
showing
was therefore
enemy
to a number
and many
display
over which
used a supervisory
commands
craft primarily
also had to attend in order
Crews
the scout within
map or situation
at skilled
often seamless
control.
color graphical
within
of a scout vehicle,
strings of action
Subjects
The task was quite complex However,
the cockpit
(autopilot)
craft by entering
for the experiment.
a dynamic,
simulated
skilled
experimental to achieve
session. mastery.
rapid and a fluent behavior
to
control
for planning
interfaces
and
At an abstract
and supervisory
is responsible
in
of rich, graphically
in a simulated
systems,
micro-world
(Kirlik,
and discrete
control
craft operating
automated
design
human-machine
in both manual
a system
for (often remote)
in a dynamic
of both continuous
of many
task, in which
existing
interaction
resources
1987) of a set of semi-autonomous
of a supervisory
elaborate
level, the experimental
technology
level, the task required
as by increasingly
that could
of information
the kind of interface
unruly
can be straightened
of human-environment
At a concrete
sources
the apparently
Task Analysis
approaches
1991).
cases
of cognition
modeling
of Ecological
We have performed
that in some
accounts
to environmental
An Example
order
I suspect
and
was observed.
of
30 Theapparent relying
economy
heavily
with some
of control
advantages
of the short-order was possible
identifiable
uncertainties
interface
which
the interface
action
selection.
model
capable
through
design
may have required
example
the simulated
environmental
concerns world.
and cognitive
can be found
in Kirlik,
configuration displayed
of search
in light brown.
modeling
(the unfilled
circle)
The lightly
action:
scout
performed objects,
locomotion
in open rather on the other hand,
were considered influenced refuel
for cargo
search
at home
forested
to resolve
task analysis where
as well as those
of
the control-
situations
in
intensive
the design
mode
of
of a process
description discrete
search
of this model action
paths
as well as a description
selection
in the laboratory
Figure 4 is a depiction
and dark green,
of
task
of a world
here indicated
forested
for the scout
regions,
in white,
here shown
respectively.
Only
were as light
home
base
of the scout are shown in the figure.
Figure
and enemy
regions
with scout
was more efficiently located
4 about
craft requires
than forested
here
consideration radar.
of two capacities
Locomotion
due to the need to navigate performed
in forests.
since fuel expenditure during
of continuous
and heavily
objects
base at some point
were not
behavior.
The open regions,
and sighting
path selection
situations
were used to motivate
as light green
to be more densely
available
to mode
action
to use a more cognitively
(1991).
to subjects.
and the initial location
the world
mode,
for dynamic
Insert
Searching
productive
An ecological
processing
the selection
and Jagacinski
gray and dark gray, were displayed
in relation
behavior
modeling
as it was displayed
representation,
above
were not always
to identify
subject
A more complete
Miller,
upon
information.
subjects
problem
were
that a perception-action
in order
of the task analysis
t_sk analysis
The present
was performed
that subjects
as discussed
unlikely
and actions
perceptual
mimicking
mode
it seemed
information,
a perception-action
The results
as an external
some of the constraints
supported
of successfully
An ecolo#cal
since
with non-specific system
information
However,
from the displayed
associated
led to the hypothesis
of this processing
cook.
in all cases,
this human-environment display
in this environment
upon the rich set of graphical
of the attendant
our example
easily
of behavior
the middle
in lightly
In addition,
for
was most efficiently around
forested
trees.
Sighting
areas because
the fuel range
objects
constraints
rates were designed
so the scout
third of the experimental
session.
had to
31
InsertFigure5 abouthere
Figure5 showssearchpathscreatedby two differentsubjects.Both beginby travelingnorth alongthe boundaryof theinnerforest,turn eastto follow thetop boundaryof this forest,loop backwestalongtheboundaryof theupperforest,turn southalongtheborderof theleft forest, visit homebasefor refueling,thendepartto theeastandthenthe north,at which point thesession terminated.This boundaryhuggingbehaviorresultedfrom theinteractionof the two majorcriteria for searchpathselection:searchingforestswith the 1.5mile radarto discoverobjects,and locomotingthroughopenterrainto coverasmuchareaaspossible. Whatis themostfaithful descriptionof thecognitiveactivitiesunderlyingsearchbehavior? Whatis an appropriatemodelfor thedepthstructureof this environment?Weof coursewould like tofind a modelof depthstructurethatcapturesthoseenvironmentalfeaturesto which cognitionand behaviorwassensitive.Onecouldof courseformulatethisprocessasa constrainedoptimization problemandusea generate-and-test proceduretocreatealternativepathsandthenevaluatethem with respectto anobjectivefunction. However,thereareaninfinite numberof possiblepathsand thecomputationaldemandsappearoverwhelming.Onecouldalsoattemptto describethisprocess in rule-basedterms,usinganenvironmentaldescriptionin termsof perceptuallysalientobjects suchasforestsandtheir borders.With this model,thecognitiveprocesses underlyingsearch behaviorwould bedescribedin termsof themanipulationof symbolsstandingfor discreteaspects of environmentalstructure. Ecologicaltaskanalysis,on theotherhand,suggests thatwe delayassumptions aboutmodeling environmentaldepthstructureandcognitiveprocesses until afterthemodelsof environmental perceptualstructureandactionstructurehavebeenconstructed.An initial cut atconstructingthe perceptually-oriented environmentalmodelrequiredfor ecologicaltaskanalysiswould beto describetheobjectsandpropertiesperceptuallyapparenton this mapdisplay. However,aswill be seenbelow,ecologicaltaskanalysissuggeststhatwe iterateandrefinethis initial perceptuallyorientedmodelafterconstructingtheaction-oriented modelby usingthelatterto helpidentify any initially overlookedperceptualinformationcapableof specifyingtheaction-orientedstructureof thisworld. The informationusedasthebasisfor actionselectionmaybeconsiderablymoresubtle andrich thantheinformationpreservedwhenusingperceptionin a purelydescriptivecapacity (e.g.,Neisser,1988;Bridgeman,1991;Shebilske,1991). Constructingtheaction-oriented environmentalmodelrequiresthe useof actioncapacitiesasa
32 frameof referencefor environmentaldescription.Sincetwo actioncapacitiesunderliesearch behavior(locomotionandsightingobjectswith radar),we mustdescribethe action-oriented structure,or affordances,of this environmentfor bothlocomotionandsightingobjects. We will considersearchaffordancesto bea simplecombinationof thelocomotionandsighting affordances.Figure6 showsthedistributionof locomoting,sighting,andsearchingaffordances asmapsof theworld in which thepathsshowin Figure5 weregenerated.Figure6ashowsthe world asit appearedon themapdisplay. Figure6b showsthe locomotingaffordance,calculated by assigninga valueof zerofor openregions,a valueof -1.5 to lightly forestedregions,anda valueof -2.0to heavilyforestedregions.Thesevalueswereassignedtoreflect thedifficulty of rapidly flying thescoutthroughtheseregionsdiffering in treedensity. Darkerregionson themaps indicatehigheraffordancevalues. InsertFigure6 abouthere
Theaffordancevalueswereselectedby attemptingto constructanobjectivemeasureof the degreeto which relevantactionscouldbeperformedasa functionof environmentalstructure.For example,Figure6c showstheworld sightingaffordancestructure.To constructthis map,a four dimensionalvectorwasassociated with eachworld locationto indicatethepercentageof areathat wouldbecoveredby scoutradarcenteredatthatlocationthatwasopenregion,lightly forested region,heavilyforestedregion,andareabeyondtheworld boundaries.For eachpoint,the inner productof this vectoranda sightingaffordancevectorwastakentodeterminethesighting affordanceof a particularworld location. The sightingaffordancevectorwasthe samefor each world locationandindicatedthedensityof cargoandenemycraft within eachof the four typesof regions.The sightingaffordancevectorhada valueof zerofor openregionsandareabeyond world boundaries,andavalueof 1.0for lightly andheavilyforestedregions. A maximal sighting affordancewouldexist,therefore,in caseswheretheentirescoutradarrangecovereda forested region,andaminimal sightingaffordancewouldexistwhentheentirescoutradarrangecovered eitheranopenregionor areabeyondtheworld boundary.Thegradedstructureof the sighing affordancedistributionresultsfrom thecomplexinteractionbetweenthecircularradarcapabilities of the scoutandtheirregularlyshapedopenandforestedregionsthatdeterminedobjectdensity,or moregenerally,theinteractionbetweenthesubjectsactioncapacitiesandtheenvironmental structure. Figure6d showsthe searchaffordancestructure,createdby simplysummingthevaluesof the
33 locomotingandsightingaffordancesateachworld location(i.e.,a locationaffordssearchif it affordsbothlocomotingandsightingobjects).This maphasbeenrescaledto clearlyindicatelocal optimain thesearchaffordancestructure.Consideredthreedimensionally,this mapindicates peaksandridgesof high searchaffordanceandvalleysandholesof low searchingaffordance.The peak areas
indicate
open regions In order define
the best compromise
and sighing to define
models
objects
congruence
between
our original
perceptually-oriented
oriented
model.
however,
capacities
structure
be necessary
First,
the search
available
provided
to fully specify
perception manner
would in which
locomotion
also have sighting
perceptual
information
perception
would
have to supply
defined
by the 1.5 mile radar radius
needed
to assess
here I will simply forms
however,
affordances
was necessary
structure.
to first identify
did have the perceptual
ability
in a manner
that
information
search
affordance
foveating
(and could
of the world
the search
world
location,
experiments
although
area are surely
conducted),
we may expect would
be an
the information
available
environmental
experiments
of such experiments affordance
the
a circular
of such experiments
of psychophysical
however,
Given
be straightforwardly
the action-oriented
a
of the necessary
area within
that indicates
that the results
affordances,
the nature
are possible,
The result
whether
structure.
psychophysical
manner
is perceptually
at a particular
of forested
Although
what kinds
can identify
we can define
judgments
to specify
we attempt
in an isomorphic
then information
the sighting
Constructing
for the sake of this exercise
the world
To fully specify
this is possible
model
structure,
world
If perception
forest,
when
errors.
of human
What perceptual
differentiates
of the amount
that such perceptual
to specify
structure?
or heavy
of the scout.
is available
environmental
possible.
affordances.
Specifically,
measurement
Using
the displayed
were constructed
perceptually-oriented
the search
Let us assume people
assume
to which
of systematic
empirically-based to specify
the degree
extent
model
a measure
We now ask the question,
knowledge
measure
Note that
than does the action-
in this the world?
to be able to measure
in this case.
differently
the
models.
model
affordances.
affordances
through
we now examine
environmental
the world
affordance
light forest,
to this task, and also to
process,
of the displayed
by locomotion
is open region,
solutions
the perceptually-oriented
possibly
note that the perceptually-oriented
location
certain
within
could
for locomotion
out of alignment.
to the most faithful
to specify
to the differentiation given
differentiates
with knowledge
perception
demands
and action-oriented
or affordances,
together
to find a way in which
would
and cognitive
are apparently
contained
structure,
the depth
perception-actions
structure
model
The two models
the action-oriented
specifies
for fluent depth
the conflicting
regions.
the perceptually-oriented
what information
perceptual
in forested
opportunities
of environmental
between
structure.
model,
to conduct. suggested
(If results
that indicated
34 the subjectscouldnot reliablyestimatethesearchaffordancestructurewe wouldthenbeableto definethekindsof cognitiveactivitiesnecessaryto do sofor behaviorto be productive,i.e., to be in alignmentwith searchaffordances).Whatwould searchbehaviorlook like if subjectswere simplyallowing searchbehaviorto begovernedby theperceptuallydetectedsearchaffordances? We wouldperhapsexpectin thiscasethatthe scoutwouldbeflown up thesteepestgradientin the searchaffordancestructurefrom its presentposition. But while thelocal organizationin search pathsmay bedescribablein thisfashion,searchpathsalsohaveaglobalorganizationthatis not well capturedby this simplesearchmodel. This mismatchbetweenobservedbehaviorandthebehaviorthatwould resultfrom a simple perception-action solutionto this taskis suggestiveof whatkindsof additionalcognitivedemands this taskmakesupontheperformer.Weconstruelong-rangeor globalpathplanningasthe selectionof a sequence of waypointsto bevisited,whereeachwaypointis a peakor ridge in the searchaffordancestructure.We still can,however,construeshort-rangeor local navigationin termsof the simpleperception-action modeldiscussedabovethatresultsin searchaffordance gradientascent.The selectionof anappropriatesequence of waypointsis constrainedtheneedto avoidbacktrackingthroughpreviouslysearchedregions,andalsothe needto returnto homebase atsomepointin themiddlethird of the mission.Thereis no readilyavailableperceptual informationcapableof specifyingtheseconstraintsuponproductiveactivity. Thus,we havea case of perceptualunderspecification of action,wherethe subjectmustapparentlytry out a numberof alternativesolutionsto assessthedownstreameffects(i.e.,consistencywith backtrackingandfuel constraints)of each,priorto selectinga globalsearchpath. InsertFigure7 abouthere
Theresultingmodelis shownin Figure7. At the startof a session,themodelidentifiesthe peakareasin the searchaffordancemapascandidatewaypointsto visit during amission. These peakareasweresubmittedto a generate-and-test mechanismthatattemptedto orderthe waypoints to acceptablymeetbacktrackingandfuelconstraints.Notethatalthoughacognitivelyintensive processis neededfor this purpose,arelativelysmallnumberof waypointsareconsidered,since the searchaffordancestructurecanbeusedtoobtainarelativelylow-dimensionalrepresentation of the world for long-rangepathplanning(i.e.,the setof local optima). Many of the fine detailsof local searchaffordancestructurecanbeignoredduringthisprocess.The outputof this processis anorderedsequence of waypoints.The first waypointis thenselectedasa destinationandwas
35 thusconsideredto possess an affordancefor visiting. The scoutdid not fly in a linearpathto the waypoint,however,sincescoutmotionwasdeterminednotonly by the visiting affordancebut alsoby thelocal searchaffordancestructurein thevicinity of the scout.Detailedmotion commands
for the scout
structure
to operate
force exerted
by second
affordance
basis.
on the visiting
that was very sensitive
visiting
affordance which
one component laboratory Kirlik,
relative largely
Miller,
the perceptua/ly-measurable
weights
to search
An evaluation
by the current
affordance
model
waypoint
values
in a direct structure.
of the process
with the a of motion
relative
to the
in meandering
a large weight
on the
path to the current
This search
model
interaction
in the
human-environment
adequacy
affordance
the direction
resulting
In contrast,
affordance
search
combined
affordance
resulted
of skilled
when
determined
structure.
affordances
of the empirical
and Jagacinski
waypoint,
the local search
process
which
on the local search
provided
disregarded
force field,
of the current
to the local search
in a complete
task.
Large
affordance
motion
waypoint
by considering
upon the scout as an attractive
by the visiting
on a second weighting
were created
model
was
can be found
in
(1991).
CONCLUSION I have tried to make the case that modeling integrated
human-environment
a framework
in which
can be carried details
behavior,
but rather,
In addition, concept, match
it assists
can then specify
task analysis
provides
in perception-action
the design
of modeling
and cognitive
in the identification
and mismatch
to be a theory
the process
the perception-action ecological
framework
environmental
of displays
and controls
human-environment
interaction
global
to flesh
for design. through
structure.
phenomenon of cognitive
instance
is offered
of of humanprocesses
have available.
For an already the process The results
existing
design
of identifying of such an analysis
and action
structure
into
better alignment. Modeling environmental
skilled
structure
promises
to be a messy
that relies upon
business.
the intensive
as
out the
of interaction,
may already
that bring perception
take the
interaction
is in no way a new theory
the scientist
demands
which
task analysis
to be done in order
any particular
resources
theories
human-environment
any new models
models
of cognitive
Ecological
of the more
As such, it does not contribute
it guides
requires
underlying
work remains
The proposed
it may be considered
interaction.
upon
for the mechanisms
task analysis.
although
environment
the search
design
as the unit of analysis.
It is quite clear that much
of ecological
cognition,
relying
out.
system
to support
exploitation
We are led to this conclusion
once
of we
or
36 acceptthepossibilitythattheability to skillfully performwithin sucha widerangeof environments is atestimonyto our adaptiveabilities,ratherthantestimonyto anyabstract,context-free,generalpurposemethodswe mayhavefor the selectionof action. Perhapswe only resortto generalpurposemethodsfor actionselectionwhenwefail to find (or adjusttheenvironmentto allow) moreefficientperception-action solutionswhich areenabledby exploitingmanyof theconcrete detailsof the perceptionandactionstructuresof our world. The searchfor modelsof skilled interactionshouldnot bemisleadby thefact thatwhatpeoplecando mostefficientlymay be hardestto describe,andwhatwe do mostinefficientlymaybeeasiestto describe. ThroughoutthischapterI haveadmittedlytakena cavalierattitudewith respectto a numberof detailedpsychologicalissues.This is not becauseI do not think suchissuesareimportant,but ratherbecauseit is all tooeasy(for me,atleast)to get soenamoredwith psychologicalminutiae thattheforestis lost for the trees.Thus,theemphasishasbeenon theissuehow we will know a cognitivemodelcapableof guidingenvironmentaldesignwhenwe seeone. In orderto takesome positivestepsin this direction,I havehadto pointout thedeficiencesof varietyof existing modelingapproaches in cognitivescience.I havenot donethis out of lackof respectfor such research.It is simplythecase that the constraints that guide modeling in normal scientific practice are quite
different
than the constraints
give rise to problem these solutions support
decompositions,
I quite expect
for decomposing
the cognitive
operative
factors
as many
impediment Those
in human
to design,
environmental
machine
design
against
is in many models
aiding
strategies
the symptoms statement design models
welcome
I have portrayed.
unless
However,
theoretical
than the cause.
must police
that is the explicit
basis.
linked
of the kinds of psychological
community
process:
Constraints
integration,
and criticize
the social
my own proposals and physiological
from
the cognitive
the role of such factors
of our theories
problem
much
design
activity
Obviously
I believe
that treating
models
necessary
to support
for no one is likely
process
of behavior.
solving
of the chaos
is a major
as well.
that I have done for the design over-rationalization
and
as we must to
structures
cleaved
One must be on guard
to current
its own research, goal.
along
to describe
these criticisms
design.
give rise to sets of solutions,
be so easily
a far cry from the explicit
that are too tightly rather
cannot
may claim
in the modeling
be due to the lack of a sound
Perhaps
If the failure
experience
cases
else will come
systems
to support
we aim for theoretical
problem.
assume.
we should
with concrete
practice
modeling
modeling
decompositions
when
that someone
of us normally
have tried to guard design
problem
give rise to new problems
design.
stressors
that must guide
search
Current
over
of the design against
what I
process
proposing
may
design
lest one run the risk of treating the cause cognitive to produce
requires
an explicit
engineering. readily
The
applicable
Acknowledgments
I am grateful was supported Shively,
to Kim Vicente by NASA
technical
monitor.
for helpful
Ames
comments
grant NAG2-656
on a previous to the Georgia
draft of this paper. Institute
This research
of Technology,
Robert
J.
References
Anderson, Barwise, Press.
John R. (1990) Jon, and Perry,
The Adaptive John.
(1983)
Character Situations
of Thought.
Hillsdale,
and Attitudes.
NJ:
Cambridge,
Erlbaum.
MA:
The MIT
Beach, K. D. (1988) The role of external mnemonic symbols in acquiring an occupation. In M. M. Gruneberg, P. E. Morris, and R. N. Sykes, (Eds.), Practical Aspects of Memory.: Current R_search and Issue_, Vol. I. New York: Wiley. Birmingham, H. P. and Taylor, F. V. (1954) A design philosophy for man-machine systems. Proceedings of the I.R.E., Vol. XLI], No. 12, (pp. 1748-1758).
control
Bohlman, L. (1979). Aircraft accidents and the theory of the situation. Resource management the flight deck. Proceedings of a NASA/industry workshop. NASA Conference Proceedings 2120. Braitenberg, Valentino. The MIT Press.
(1984)
Vehicles:
Brehmer, B., and Joyce, North Holland.
C. R. B., Eds.,
Experiments
(1988)
in Synthetic
Human
Psychology.
Judgment:
The
Cambridge,
SJT View.
New
MA:
York:
Bridgeman, Bruce (1991) Separate visual representations for perception and for visually guided behavior. In Pictorial Comml,mication in Virtual and Real Environments. Stephen R. Ellis, Mary K. Kaiser, and Arthur J. Grunwald, (Eds.). New York: Taylor & Francis, (pp. 317-327). Brunswik, Egon. (1952) Encyclopedia of Unified
The conceptual framework of psychology. In, International Science (Vol. 1, No. 10). Chicago, IL: University of Chicago
Brunswik, Egon. (1956) g..xperiments. University
Perception and the Representative of California Press.
Design
Press.
of Psychological
Card, Stuart K., and Newell, Allen. (1989) Cognitive architectures. In Jerome I. Elkind, Stuart K. Card, Julian Hochberg, and Beverly Messick Huey, (Eds.), H0man Performance Models for Computer-Aided Encneering. Washington, D.C.: National Academy Press, (pp. 173-179). Carroll, J. M., and Campbell, human-computer interaction.
R.L. (1989) Artifacts as psychological Behaviour and Information Technology,
Carroll, John M. (1991) Introduction: John M. Carroll, fEd.). New York: Claxton, Claxton, Donald, Press.
Guy. (1988) fEd.), Growth Merlin.
(1991)
The Kittle House manifesto. In Designing Cambridge University Press, (pp. 1-16).
How do you tell a good cognitive Points in Cognition. New York: Origins
theories: The case of 8, (pp. 247-256).
of the Modem
Mind.
theory when Routledge.
Cambridge,
Edwards, E. (1988). Introductory overview. In E.L. Wiener Factors in Aviation. New York: Academic Press.
Interaction,
you see one?
MA:
Harvard
and D.C. Nagel
In G.
University
fEds.)
Human
Fischoff,B., Slovic,P., andLichtenstein,S. (1978) Fault trees:Sensitivityof estimatedfailure probabilitiesto problemrepresentation.Journal of Experimental Psychology: Human Perception and Performance,
4, (pp. 330-344).
Flach, John M. psychophysics. Flach, _,
(1990a) Control with an eye for perception: Precursors Ecological Psychology, Vol. 2, No. 2, (pp. 83-112).
John M. (1990b) 2(3), (pp.
The ecology 191-205).
Gibson, James J. (1979) Houghton-Mifflin.
The Ecolo_cal
Gibson, James J. Houghton-Mifflin.
The Senses
Gibson, Reasons
(1966)
James J. (1967/1982) for Realism. Hillsdale,
of human-machine
Approach
systems
to Visual
Considered
to an active
I: Introduction.
Perception.
as Perceptual
James J. Gibson autobiography. NJ: Erlbaum, (pp. 7-22).
Systems.
Ecological
Boston,
MA:
Boston,
MA:
In E. Reed
and R. Jones,
Hammond, Kenneth R., Hamm, Robert M., and Grassia, Janet. (1986) Generalizing over conditions by combining the multitrait-multimethod matrix and the representitive design of experiments. Psychological Bulletin, Vol. 100, No. 2, (pp. 257-269). Hammond, Kenneth R., Stewart, Thomas R., Brehmer, Berndt, and Steinmann, Derick O. (1975) Social judgment theory. Center for Research on Judgment and Policy Report No. 176, University of Colorado Institute for Behavioral Science. Hammond, Kenneth R., Hamm, Robert M., Grassia, Janet, and Pearson, T. (1987) Direct comparison of analytical and intuitive cognition in expert judgment. IEEE Transactions on Systems. Man. and Cybernetics, Vol. SMC-17, No. 5, (pp. 753-770). Hammond, Kenneth R. (1955) R_vicw, 62, (pp. 255-262).
Probabilistic
Hutchins, Edwin. (1991) How a cockpit NASA Ames Aviation Safety/Automation 1991.
functioning
and the clinical
remembers its speeds. Program Investigators
P. N. (1975) Models of deduction. and Process in Children and Adult_.
Kahneman, D. and Tversky, A. (1979) Econometrica, 47, (pp. 263-291).
Prospect
Kirlik, Alex, Miller, R. A., and Jagacinski, uncertain environment II: A process model submitted for publication. Klein, Gary A. (1989) Recognition-primed Machine Systems Research. Greenwich,
An analysis
at the Second August 14-16,
for Cognitive
In R. J. Falmagne, (Ed.), Hillsdale, NJ: Erlbaum.
theory:
Psychological
Paper presented Meeting eeting.
Hutchins, Edwin. (1988) The technology of team navigation. Institute Report 8804, University of California, San Diego, La Jolla, CA. Johnson-Laird, _Representation
method.
Science
Reasoning:
of decisions
under
risk.
Richard J. (1991) Supervisory control in a dynamic of skilled human-environment interaction. Manuscript
CT:
decisions. In W. B. Rouse, JAI Press, (pp. 47-92).
(Ed.),
Advances
in Man-
Kotovsky,K., Hayes,J. R., andSimon,H. A. (1985) Why aresomeproblemshard? Evidence from thetower of Hanoi. Cognitive Psychology. 17. Larkin, words.
a diagram
is (sometimes)
McRuer, D. T. and Jex, H. R. (1967) A review of quasi-linear 0n Homan Factors in Electronics, HFE-8, (3), (pp. 231-249).
pilot models.
Marr,
Jill H., and Simon, Cognitive Science, David.
(1982)
Herbert A. (1987) Why Vol. 11, (pp. 65-99).
Vision.
New
Meister, Hopkins
David. (1989) Con¢¢ptual University Press.
Meister,
David.
(1984)
Letter.
York:
of Human
Factors
Factors.
Society
Ulric.
Newell, Allen, Prentice-Hall.
(1976)
Cognition
and Simon,
Herbert
and Reality. A. (1972)
New
Bulletin,
York:
Human
27(10),
W.H.
Problem
Donald
Rasmussen, to Cognitive
A.
(in preparation)
Jens. (1986) En_neering.
Things
Information New York:
That Make
MD:
Freeman Solvin¢.
Us Smart.
The Johns
systems.
Address
and Company. Englewood
Cliffs,
and S. W. Draper,
Reading,
Processing and Human-Machine North-Holland.
Transactions
(p. 2).
perceptual
Norman, Donald A. (1986) Cognitive engineering. In D. A. Norman User Centered System Design. Hillsdale, NJ: Erlbaum. Norman, Wesley.
IEEE
Baltimore,
Neisser, Ulric. (1989) Direct perception and recognition as distinct presented to the Cognitive Science Society, August, 1989. Neisser,
a thousand
Freeman.
Aspects
Human
worth
MA:
NJ:
(Eds.),
Addison-
Interaction:
An Approach
Reitman, Walter, Nado, Robert, and Wilcox, Bruce. (1978) Machine perception: What makes it so hard for computers to see? In C. Wade Savage (Ed.), Perception and Co maition Issues in the Foundations of Psychology. Minnesota Studies in the Philosophy of Science. Volume IX. Minneapolis, MN: University of Minnesota Press, (pp. 65-88). Rouse, (pp.
William
B. (1987)
Much
ado about
data.
Human
Factors
Society
Bulletin,
30(9),
1-3).
Sarter, Nadine phenomenon.
B. and Woods, David D. Th_ Intcmat;ional Journal
(1991) Situation awareness: A critical but ill-defined of Aviation Psychology. 1, (1), (pp. 45-57).
Shebilske, Wayne, L. (1991) Visuomotor modularity, ontogeny and training high-performance skills with spatial instruments. In Pictorial Communication in Virtual and Real Environments, Stephen R. Ellis, Mary K. Kaiser, and Arthur J. Grunwald, (Eds.). New York: Taylor & Francis, (pp. 305-315). Sheridan, Factors.
Thomas B. (1987) Supervisory control. In G. Salvendy, New York: John Wiley and Sons, (pp. 1243-1268).
Suchman, Lucy. Communication.
(1987) Plans and Situated Actions: New York: Cambridge University
The Problem Press.
(Ed.),
Handbook
of Human-Machine
of Human
Tversky,A. andKahneman,D. (1981) Theframingof decisionsandthepsychologyof choice. Science.
211,
(pp. 453-458).
Wason, P. C. and Johnson-Laird, P. N. (1972) Content. Cambridge, MA: Harvard University
Psychology Press.
of Reasoning:
Structure
and
Whiteside, J., Bennett, J., and Holtzblatt, K. evolution. In M. Helander, fed.), Handbook Holland, (pp. 791-817).
(1988) Usability engineering: Our experience of Human-Computer Interaction. New York:
Wise, James A. (1985) Decisions in design: Behavioral Decision Making, George Wright
Analyzing and aiding the art of synthesis. fEd). New York: Plenum, (pp. 283-308).
In
Woods, David D. and Roth, E. M. (1988) Cognitive systems engineering. In M. Helander fed.), Handbook of Human-Computer Interaction. New York: North Holland, (pp. 3-43).
and North
f-
J S
/
Figure
Figure
1. The Lens
model
Captions
of Brunswik.
Figure 2. The Ecological task analysis framework. Like the Lens model framework for human judgment, the ecological task analysis framework represents the integrated human-environment system in a symmetrical arrangement. Matches between the perceptually-oriented and actionoriented environmental differentiations indicate opportunities for the fluent perceptual guidance of activity. Structural mismatches between the models of surface perceptual structure and surface action structure are indicative of demands for cognitive activity, and corresponding models of the environmental depth structure. Figure 3. Four possible results of an ecological task analysis. Each case reflects a different type of match or mismatch in environmental perception and action strucuture, and is thus indicative of a different form of cognitive demands, and a different type of remedial interface design solution. Figure
4. Map display
showing
Figure
5. Search
produced
paths
home
base, the scout,
by two different
and forested
subjects
areas.
in the same world
Figure 6. Four representations of the same world. (a) Displayed affordance representation. (c) Sighting affordance representation. representation scaled to emphasize local optima in the affordance
configuration.
representation. (b) Locomotion (d) Search affordance structure.
Figure 7. Process model of search path generation. Local optima in the search affordance structure are organized into a series of waypoints via a generate-and-test scheduler. Detailed scout motion is produced by a combination of attractive forces from both the current waypoint and the local search affordance structure in the vicinity of the scout.