Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 418
SHARING DIVE q~SXS }BETWEEN PEO=E AND ~ ~ SPACE SYSTEMS
Willian H. Starbuc3:
DOT ARE 1~ CAPTIVE ADAGES OF PEOPLE AD A?
M~r~i~'s capabilities Charge ~ y slowly, Whereas Juicers'
capabilities here An fast~r~g~. The cost of a Dry eminent
has dry forty percent per argon for cover thirty yeas;, arm Dry
sizes have ~ even Are rapidly than that (Al~us, 1981; Toong and
Greta, 1982). Mutation is have been accelerating nearly 25
percent yearly, the cost of logic haggle has been draping Dally
rapidly, and the Elation work dorm Lath each unit of ~v has
An rising thirty percent per am.
Ash Are reliable and very ~ Weller.
-
~ — ,.
her a has b
Use' interfaces and
pricing larynges have Repaved considerably, es—:ially aver Me
last decade. If hogan Beirut had evolved as rapidly as Darters sin e
the Ad 1950s, the best r = Acid now finish a 26-mile marathon In
2.3 secorx~s, a bright student Ward Deplete all s~hoolir~ fag
kinden3artQ to a AD. ~ a bit aver two days, r~rmal eater';
would consume ore calorie per month, and half of African families
Ward be ~ Are than $141~000, 000 anally.
The i~prover~nts In ~tiy Am;, sizes, and sums have
generally ~ the At ~cimistic forests of yesteryear, as has
the prolife~tic~n of is. ~f;1 l", however, have been the
forecasts predictir~ that is would Shortly be able to imitate
human beings. For example, In 1960 Since cptimisti~1ly ~a+ -
that "Duplicating the problem solving and informati~-har~ling
capabilities of the brain ~ not far off; it Scald be surprising if it
were not Vised within the nest decade" (Sloan, 1960:32).
is have not, in fact, develcged an ability to reason very
much like people, and ~ ter s; ~ cation of h w ~ thought has had
little success (Albus, 19813. When computers look most effective
solving problems, the computers ,,-~^ quite different techniques than
people apply (Weizenbaum, 1965; Wine grad and Flares, 1986~. For
example, Newell et al. (1957) studied students' efforts to pace
theorems in mathematical logic, and inferred that the students search
for proofs, using heuristics that generally lead toward proofs kut do
not Guarantee them. Challenged by such Park, Wang (1963) devised a
computer program that efficiently proved all 200 theorems in the first
five chapters of Principia M~thematica. Job-shqp scheduling affords
418
OCR for page 419
419
anther exile: &ientifi~manag~nt studies of human prc~uction
schedulers led to the development of Gar~tt charts to portray
graphically the activities of various machines, arm thus to help human
s~h~ulers visualize the cascading implications of alternative
assignments. Muters generate j~dh~p sc~h~ules by solving
~nteger-progra~r~ problems that no human card solve correctly
without machine assistance.
me differ; be~reer~ people and caters have an illusory
quality, insofar as people ted to take prevalerlt human abilities for
granted ark to notice rare or inhuman abilities. If computers did
operate exactly like people do-~working at the same speeds, making the
same mistakes, showing the same fatigue, complaining about unpleasant
tasks, and so on--people would regard computers merely as inhuman
labor. Computers most impress people when they augment human abilities
significantly - by working silently and tirelessly, by calculating with
dazzling spear, or by displaying toted consistency.
But the quite real difference= between people and computers are
persistent and profound. Rather than regard computers as potential
Imitators of human beings, it makes better sense to look upon them as a
distinct species a species that prefers different languages, reasons
with somewhat different logic, finds comfort in different habitats, and
consumes different foods.
Computers are much better symbol manipulators an] much stricter
logicians than pecgle; and computers are much more decisive, literal,
precise, obedient, reliable, consistent, and transparent. Computers
can act both much more quickly and much more slowly than people. If so
instructed, computers will carry out utterly absurd instructions or
they will remain completely calm in the face of impending disaster.
Computers easily simulate what-if conditions; and they can extrapolate
even the most farfetched implications of theories or conjec Ares.
People, on the other' hand, possess brains that are so much mare
complex than the largest computers that comparisons make no sense.
These brains carry on numerous simultanecus and interacting processes'
some of which operate entirely automatically. Without even trying,
people prom=== visual and auditory data of great complexity. People
can shift levels of abstraction freon detail to generality and back,
they separate foreground images form background images, they
distinguish patterns while remail aware of conks, ark they attend
to important or unusual stay i while ignoring un~rtant or routine
stimuli. Peccable halve quite extensive pries that possess ~anir~ful
structures; ark if they have relevant information In their Brie=,
r~1 = Bestial 1~, Urn ;+ wry hem ma" bestial lair Iffy ; I
~_~ ~~` ~~— a_ A_ ~~ ~. ---a. _` _~_ __. People can
Berate with imprecise ark somewhat in~lete plans, and they can
extrapolate Bed pat experiences to navel situations while
Agonizing that they are indeed Operating outside ache limits of their
direct experience (A11en, 1982; Dreyfus and l~reyfus, 1986; Mbray, 1986;
Reason, 1986; Winograd and Flores, 1986~.
Perhaps most importantly, people are more playful than Muters and
better at making m~sta~s. Whereas computers obey instructions
literally, people often ignore or forget instructions, or interpret
them loosely. Not only do people tend to deviate from plans ark to
OCR for page 420
OCR for page 422
OCR for page 423
OCR for page 424
OCR for page 425
OCR for page 426
OCR for page 427
OCR for page 428
OCR for page 433
OCR for page 434
OCR for page 435
OCR for page 436
OCR for page 437
OCR for page 438
OCR for page 439
OCR for page 440
OCR for page 441
OCR for page 442
OCR for page 443
Representative terms from entire chapter:
decision support
420
test the limits of assumptions, but many human perooph~al skills and
response yes Repel on Ring deviation fray e~ations or
goals that may be evolvir~. Sometimes, people begin to doubt even
their most basic beliefs. main, people generally expect to male
vistas; and to learn f, =~ them, arm Dative people may be very good
at learning freon mistakes. If they have sufficient time, people can
learn to correct then stakes arm they can reprogram themselves to
take advantage of hi situations. Although computer; also
conserve arm react to deviations, ~ have not yet exhibit huh
capability to reprise goals for themselves, to reprogram themselves, or
to question their As tic premises Valiant, 1984). Raters must
be toil to leant from they experiences, and efforts to enable aches to
Hearst have, so far, been restrict ~ very narrwr domains of
activity. Also, Muters are good at not making mistakes in the fist
place, so they have less need to learn fee., mistakes.
People are, hammer, pretty diverse and flexible. Sane people can
learn skills and perform thy; that other people find impossible; arm
since MA can chase freon a large pool of applicants, the extreme
capabilities of e~ccepti~al Maple Me more important In space systems
than the average capabilities of Opine people. me people Do
Berate space trysts first receive thorough training, so their
deficits of ir~rierx~ Chid be small; but this training itself may
i ~ e serious liabilities, such as a tendency to rely on
well-practiced habits in Rachel situations.
Ire people are flexible and complex, they often surprise
scientists and systems designers: People may change their behaviors
significantly in response to 06bensibly small environmental changes, or
people may change their behaviors hardly at all in response to
apparently large environment=] changes. How people react to a
situation may depend quite strongly on the sequence of events leading
up to that situation, including the degree to which the people Ace
themselves as having helped to create the situation. Accurate
statements about ~crosoopic details of human behavior rarely prone
accurate as statements about general, ~acrosccpic by rival patterns,
or vice versa. For exa=le, experimental studies of people who Ale
being paid law haIrly wages for making repel ted choice between two
cleanly defined, attract synods that have no implications for later
events probably say lithe abed he an behavior In Pro-life settings
where actions may have per~;istent arm Dismally significant
I= arm Me actors may no even perceive ~chenselves as
having choices. Tersely, broad generalizations about the by riors
of May people ~ diver~;e si - lotion probably say little about the
behaviors of carefully Ale people who are performing Unusual tasks
in Dish they have great experience.
The research issues that are important for designing human-ccmputer
systems son to be
421
root advocating any research aimed at describing human capabilities
general The designers of space systems should not Kept on general
theories, but Should test fairly realistic ~c-pps of interface=,
her ~ re, art sof ~ are, with people who are as well train ~ arxt as able
= real astronauts and controllers. The designers should also
investigate the sensitivity of performance measures to small variations
in theft designs (Gruenenfelde' and Whitten, 1985): Do small design
changes produce large changes ~ performance? Both to improve the
quality of designs and to improve users' acceptance of designs,
experienced astronauts and controllers should participate ~ the
designing of interfaces an] systems; and because early decisions often
constrain later mcdifications, astronauts anS controllers should
participate from the beginning of any new project (Grudin, 1986~.
PEOPLE INTERACTING WITH CCMEUTERS
Tbday's computers cannot imitate people very closely, but the
differences between people and ccmput~'s imply that combinations of the
two can achieve results beyond the capabilities of each alone. For
that reason, NASA should devotee research effort to improving the
interactions and synergies between pecgle and computers.
Five research topics seem especially interesting and important
because (a) ~ can see how to pursue them and (b) ~ can foresee some
research findings that would translate directly into improved
performances by space systems.
1. Fostering Trust Between People and Expert Systems
2. Creating Useful Workloads
3. Anticipating Human Errors
4. Developing Effective Interface Languages
5. Using fitful Interface }$e=Eihors
Fostering Trust Between People and E ~ Systems
Decision-support systems are co Touter programs and data bases that are
~ tended to help people solve problems. Some decision-support systems
merely afford then r users easy arouse to data; other decision-support
systems actually pro pose solutions, possibly basing these proposals on
data supplied by their users (Woods, 1986b).
Expert systems are decision-support systems that attempt to embody
the specialized knowledge of human experts. m ear proponents argue
that expert systems can, in principle, make specialists' knowledge
available to nonspecial lets: every CPA might be able to draw upon the
combined expertise of several tax specialists; every genera]
practitioner might be able to make subtle diagnoses that reflect
advanced training in many specialties. Expert systems might perform
even better than human experts: Cc~put~s may be able to Obtain data
~t would be unavailable to people (Burke and Not, 19871.
Muters' huge marries and high Speeds might enable then to
422
investigate Ire alternatives or to take a~mt of more contingencies
than people consider. Computers may also adroit sew of the apical
errors to with people typically fall prey, arm thus may draw same
infer that pec$'le wed miss (Bcibrocr et al., 1986~. Advocates of
statistical Elision theory value is' ability to adhere Elite
Tricky to such foveae. Some perusals wick have stern;
fonalllating Cations arxt people then sc~cq these
rations and deciding whether to adept =en (alike ark Normals'
1987; Dreyfus and ~yf=, 1986; Woods, 1986a, 198Eib).
Not everyone holds an optimistic view of exert systems' Gentian.
Star~f;11 ark Waltz (1986:1216) remarks: '~e-bas~ expert systems ...
tow to fail - Ply for problems even slightly outside their area of
expertise ark in unforeseen situations." ~yf~= arxt Dreyf~
(1986:108) have argue that human experts do not follow decision rules
but instead they ~ "ache actual ~ of tens of chooses of
s;i~tions", ark that "If one asks the experts for Naples one will,
effect, force the escort to regress ~ the level of a beginner ark
state the nines he still r~ but no Ever uses. " Consequently,
Snafus and ~yfus (1986:109) predicted "that ~ any do n Rich
people e~i):'it holistic understar~ir~, rho systems based upon heuristics
will consistently do as w~1 as experienced eats, ever if those
experts were the informants who provided He heuristic rules. "
Dreyfus and ~us' critique may be valid. Dutton and I (1971)
Spent six years studying an expert production Ruler nab Charlie,
includli ~ cue full y-:.r investigating his pr ~ are for estimating how
much p m auction time any schedule represented. Charlie estimated time
by using the relation:
Production Time = Schedule Length / Speed
'We gradually were disabuse] of the idea that Charlie has a ccmpuLation
prcac5ure for sF-C~ and were ~~nvinoed that he obtains his speed
estimate= by a table lock-up. A hat is, Charlie has memorized the
associations between speed and schedule characteristics, and he looks
up sews in his merry in somewhat the way one looks up t-]eghane
n~s ~ a diary. In Air interviews;, Charlie ta~ as if the
existence of a Incitation procedure was a Noel idea, intriguing to
conflate but diffi—fit to conceive of. He thinks of He s~=
his table as discrete Embers distal leaf Hen a law series of he
experiences. Mt~ he can interpolate ark extrapolate th-~
Cars implyir~ that He stored speeds Bust be Specific exiles from
a systematic family of rompers—he dints the interpolated values
and speaks of them as hypotheses to be testy in application. the
story vales - are so One Are reliable Hat they It be a different
1.,: ~ ~¢ : am ~ ; ~~ =1 - ~~ - he_
V. ~~V=~V~1 me' ala ~ fact/ Belie ~ at, for a
large proportion of his table entries, specific r~ situations
in which the costar was Bunter ark the sky ~nred. the
Only is that he does not so Cement, apparently, are those
appropriate to situations arising almost Lily" (Dayton err! Startup,
1971: 230) .
423
We ~~1cula~ bat Margie he Prized approx~t=1y 5000
prc~uction Is courting ~ various si—lotions. BE we also
discover" bat C~arlie's pr~uction-~ estinm-P~ cauld be predicts
quite accurately by a simple Ian-=' Nation that had a ~ni~ful arm
generalizable interpretation fir. tense of the physics of the production
press. Rather than thousands of machine Is, this linear elation
Firm only a few hundred parameters. mu=, we could state a
procure that was simpler than the one Margie used; and because this
artificial pa ~ ure had a Eibysi~1 interpretation, a user could mare
confidently extrapolate it to novel production situations.
Cane of the.best-~n e~pert-system projects not only precut - a
heuristic program, DENDRAL, but also led to the development of an
efficient algorithm for generating molecular structures (Bennett et
al., 39831 . Evidently, the heuristic program has received little
practiced use whereas the algorithm has had much (Dreyfus arm rlreyfus,
1986) .
One Ivies question is: ply must expert systems closely resale
human ems? The proponents of eat systems typically equate
expertise with human beings, so they see imitating hen expertise as
essential to creating expert System;; and Fife critics focus on Me
differences between emoters arrt people. Yet, computers possess
different abilities than people. Outer pr~ra~i~ efforts that
have begun by imitating human behavior have often ended up unwire
techniques that made no pretense of imitatir~ human behaviors; arx!
erasers and scientists have Reprised, without imitating human
expertise, many techniques that enable computers to exceed ~ best of
human capable ities.
Other questions arise con ~ people's willingness to depend upon
computer-based expertise. Collins (1986) mtervie wed actual and
potential users of several widely known expert systems for accounting,
chemical analysis, mathematics, medico diagnosis, and
c~ter~nents ordering. She fat only one of these expert
system that has active users: the one for ordering ~
cx~or~nts PRO. It he straight-forward logical presses and it
draws rx' sibtle inferences; it vainly helps sales persormel forget no
details when they fill ~ orders, and the -Cares personnel said they
appreciated not having to waste Ached time worrying about details or
waiting for an ~ a human exam. It may be relevant that the
.
users of this system sold computing equipment. Concerning the other
expert systems, potential users expressed considerable distrust, of
other human e ~ as well as computers; and the potential users may
view these systems as threaten m g their own expertise. However, the
people who actually participated in creating these systems said they do
trust them and Lucid, but do not, use them. Collins inferred that
trust in an expert system comes either from participating in the design
purr=== or from being able to change the system to reflect one's own
expertise. This inference meshes with the general pattern of
psychological research, but neither of these options was available to
the cc~utingff3pipment sales personnel, JO were the users voicer Me
grit trust ~ an expert system.
424
duplex issues surrour~ ache idea Mat a user Should screen an Bert
sys~n's rations and decide wrencher to adept then. If an
expert syst~n draws the sad inferences that its user would draw and if
it r~s the sad actions that the user wed choose, that user
will racily learn ~ trust ache ~m. Such Moms to be ache ~ with
the en Dyson for ~r~ents ordering. Such a sy~i~cem may
relieve people freon having to perform boring or easy work, kilt it alas
very little to a user's mt~ll~ual capabilities, whereas In
principle, computers' precise logic are extensive computation
capabilities and the incorporation of exceptionally high-quality
expertise sight enable expert systems to draw substantially better
inferences than their users and to choose distinctly better actions.
Yet a user is quite likely to distrust an expert system that draws
significantly different inferences and that chooses significantly
different actions than the user wend do. If the expert system also
us-= a ca~putatic~nal pr~e that diverges quite dramatically from
human reasoning, the system may be unable to explain, in a way that
satisfies users;, shy it draws certain conclusions and not others.
Dish users; may never discover whether an expert system Is making
good r ~ ations or bad ones.
this calls to mind the experience of a manufacturing firm that
installed one of the first computer-based systems for job-shcp
scheduling. The system's creators promised that computer-generated
schedules would produce considerable savings ~ comparison to
human-generated schedules. The factory's managers, however) were not
entirely sure of the goodness of ccrputer-generated schedules, and they
wanted to minimize the poplin insult to then hen production
schemers, so the managers told the schemers to follow the
c3~uter's Sedations as long as they agreed with them, but to
substitute their clown judgement Ben they thought the cc~uter had made
bad r~ations. An evaluation conducted after ore year Eyed
-that the combated system had yields no imprc~vemen~ whatever.
But r~ may be able to suggest same answers to then issues, at
least in part; and god design may be able to resolve Alien: Exit
systems, even the canes that Cannot r~i2~1ly explain the ring
that 1~ them to mace Main rations, Child be able to
explain By they believe their rations to be good. Pepple So
cannot formulate a good = rustic may be able to recognize a good
Vatican or a bad one, and people do satires ~ze their
awn l;,nitations. At least sane of the people To manage factories have
lean ~ to trust ~ ter pr~rmns for pa ~ action scheduling or
inventory control even though the=" people could nck themselves
generate the computers' solutions.
the foregoing observations highlight the practical significance of
research about the factors that influence people's trust in Hers'
expertise. In what way should a decision-support sysbem's know' edge
and logical rules fit each user individually? Given cpportNnities to
tailor int~fa~= to their p~;onal preferences, inexperier~ users
may design illterfa~= poorly (I=nais arx] [Knauer, 1982): Do users
trust system; more or less when tailoring is Ed until the users
gain considerable experience? How do task characteristics affect a
425
userls wills to trust a decision~ort system? In mat
cir~nstances disc a user decide to Angst a spur system that
captures me Repledge of ems Woo the user does not knew
E~;ona~y? bat kiss of experiences lead a user to trust a
decisiq~rt system bat the user regards, at l=~t partly, as a
blamed? That kids of friends encourage a user ~ see a
decision~port systems I;~n;~cations and to ove~icle bad
r~=n~ati~s?
Cnea~cing Useful Workloads
A=oma~cic~n Is to malce chum; responsible for Rhine, -say Dams
arm ~ leave the Tonne, difficult tasks for people. Cone ran
for this may be me Option that r~onr~tine harks are interesting
ark ballerina, arm Is worthy of hogan attention, w~P~as routme
tacks appear -my awl ~int~ting, arxi so ante to people. Bit a
Ore important reason may be the practical id bat designers can figure
cut how to animate r~u~cinized activities whereas they carom
effectively automate activities that vary.
mis division of labor pros the Constance that, as automation
pry, people's work Ames more arm more diver~;e and
unpredictable arm it ta~s on rave am Ore of an ~~
fire-fighting character. At the say Tim, cutting people out of
rout he tasks isolates them freon ongoing information a ~ ut ~ at Is
happening and forces them to acquire this information while they are
trying to perform nonroutine, difficult tasks. The human controllers
in a system may not even be warned of gradually developing problems
until the system exceeds critical limits and alarms go off OWeiner,
1985). Thus, people's w ark grows less do-able and more stressful
(Senders, . 1980) ; and eXcrene stress and extreme time pressure may cause
people to do poorer work and 1~== of it.
In many tasks, automation also inner= the dhort-term stability of
me variables used to monitor perforce; as Weiner (1985:83) put it,
llautomation tunes At mull errors and Realms opportunities for large
canes." De Geyser (1986) has suggest that this dhort-term
stabilizatic~n causes the human operators to Shift face an anticipation
logic to a recovery logic: instead of keeping track of events ark
trying TO manage there, the Orators wait for significant urxiesirable
events to ~~ a. Furthermore, "At the highest automation stage, the
production Astor he only very Sketchy cooperating images of process
arm installatic~n.... He will not make a huge investment ~
c~vaticm, i, judging, est~hli~hing relationships, gathering
of data without being certain of its usefuines=. The operator does not
invest psychologically in a role With escapes him" (De Geyser,
1986:234-235~. Hence, De geyser et al., (1986:135) have advocated that
"the person still play an active part in the or~oirg activity, not
because this presence is rehire, but because it autanatical ly keeps
he person up to date on the current shale; of the system, the better
to rid if an evens y situation develops." This seems a plausible
hypothesis, but an equally plausible hypothesis would be that operators
426
r
tend to work Isis ly when they are performing ~e kits of
activities Mat card be automated.
De Geyser also t however, potted cut that serious eminencies call
for as not automation as possible because they produce extreme time
pries, ~r~ly complex problems, and extreme dangers all of
which greatly Engram the capabilities of human Repairs. Of carte,
people are Utterly unable to r~ as quickly as scam eminencies
I. -this poses a Ca~22. As long as the desigr~rs of a system
have sufficing urxi=5~i~ to be able to prescribe has the system
dhauld respond to a serials emergency, they ~ha~lcl irx~orporate this
underset ~ the sys~n's automatic responses. But such complete
und~star~i~ Uphold imply that the aromatic system works so well that
a plar~-for serials emergency rover occurs. Consequently, when a
Trials eminency does arise, is not design error one prominent
hypothesis abaft its cause, and dens that hyp ~ heals not render suspect
the diagnostic information being produced by the system? Any
system-design process establishes a frame of reference that identifies
some events as relevant and important, and other events as irrelevant
or unimportant; and a cost-effective system monitors the relevant and
important events and ignores the irrelevant and unimportant ones. But
this is likely to mean that the system lacks information about sane of
the events that produce a serials emergency, and Me incomplete
informaticm that the System does have available may well lead human
diagnosticians astray. Treater, hen operators who participate
contirma~sly in a system might grow so famniliar with the system and its
current sta== Tat Hey overlook an~rnalies AL lack the cibjeccivi~ to
Newport effectively to a serious er~enc~y.
I~yir~ to diagnose the causes of an we emergency AL to
"relcp remedies, human Orators must ur~erstar~ Is are ather
~nadhines Icily why, which implies that they are quite comfortable
Off Inters are with Me cat newels Hey in orporate; but on He
Other hart, human Orators must distrust they computers; arxt
IBM Hers sufficiency to be able to sift
c~q#generated information with skeptics eyes. sim; 1 arly,
canfider~e in Fir tsainir~ can help people remain cello in an
emergency, but canfia in Heir training also blinds people to its
~hort~ai~. It Anus seems likely Hat the people who do He rest
good in ~nPrgencies have an ability to discard there preconceptions arm
to loo]: at sits from resew points of vi~r (urchins arm Aphids,
1959; Wat21awi~k et al., 1974). NASA Should investigate the degrees to
which such an ability varies among people and can be predicted or
taught.
WorKL=~-c vary in duration as well as irr~nsi~r. People can cope
with very it workloads for Art periods, yet Hey experience
stress frma federate workloads that persist for leg periods (burner
and Mask, 1984~. Sane pysiologic~1 reactions to stress, such as
ulcer; arid ~3n~bility to infection, tale time to develop. thus, the
short - duration shoe flights do not afford a good basis for
for~sti~ the workloads ~ be experienced con lord—urination tars an a
space station. NA5A Child contirn~e to investigate He workload
427
experiences Id fun long stays in confined spans such as
Arlta~ca, Sealab, and nuclear shrines (Blush, 1984).
Anticipating Than Errors
Overloading causes people to mad errors, but so do boredom,
inatter~cion, and indifference. Than errors are botch prevalent and
inevitable (Senders, 1980), and Mary human firs are desirable despite
. . . _ . . _ .. . . . . .
. :nelr Is.
People experiment, and she of their expert - nts turn salt
badly. People deviate fern their ~~cructions, and some of these
deviations have bad consequences.
. ~ . . ~ . ~
Norman (1983, 1986) and Reason (1979, 1986) have initiated research
into Off causes of errors and ways to prevent or correct them. Norman,
for inset, distinguished errors in intention, Rich he called
misword, frcm errors ~ carrying out intentions, With he railed
slips. He classified slips according ~ their saucers, and then sought
to prescribe reties for variants slips.
Norman's categories and prescriptions.
Table ~ lists same of
P~nizir~ errors' importance, NA~'s Than Factors R - earn
Division is currently conducting some well-thought-cut research on
error-~etection and on error-tolerant systems. Error-detection systems
could warn people when they appear to have omitted actions, to have
acted cut-of-order, or to have taken harmful actions. Error-tolerant
systems would first detect human errors through unobtrusive monitoring
and then try to Icy them.
This research has much to ~ it.
But scone errors are very
costly to tolerate, and scene errors are very costly or impossible to
correct. So human~tcr systems Should also try to predict human
errors ~ order to make serious errors unlikely in advance (Schneider
et al., 1980; 5hneid~man, 19861.
and more errecclve Bean cure, and rcs~rch on error pr~entlon But
usefully cc~ler~nt the current projects.
.
~ ~ _ _ _
That is, prevention -may be cheaper
Of ~r:;e, all hour systems egress scale assertions about
~e'' hen participants.
~ . . ~
These assumptions have nearly always been
1 - loch; arm cney nave nearly always been static, insofar as the
assertions have not Charged ~ response to pepple's actual by riors
(Rouse, 1981; Turner ark bused, 1984~. For many -racks, it would be
feasible to explicate fairly accurate Gels of people. In fact,
newels red red be very accurate in order to make useful predictions or
to suggest There adaptability to people's actual behaviors ~ght pay
off. is might, for example, predict that people who respond to
stimuli q~iblcly are more alert than people who respond slowly; or they
might predict that experience people would resporx] more quickly than
~ienc~ Ones; or they might predict that people would be Ire
likely to behave in habitual ways than in un~1 ways; or they might
predict that people would be 1-cc concerned about small discrepancies
den huh activity is Burring. n~ on a review of human-factors
reseat, Simes arrt SirsJ;y (1985) hypothesized that:
428
ABLE ~ Scam Error Categories arrt Prescriptions
Forming the Wrong Intentions
Made errors:
misclassifications of systems' Awes
Ascription errors:
a~bigua~s statements of
intentions
Misdiagncees:
Eliminate ~es.
Give better indications
of medic.
Use different Unarms
In different rides.
Arrange controls
~ly-
Give contr~ls
distinctive Bihar.
Mike it diffirtllt or
infusible to take
actions that have
serials, irreversible
hem;.
Suggest alternative
explanations.
Point At discrepancies
that might be
c~rer100k".
Activating the Wrong Behaviors or Iriggering Behaviors at the throng
Times
Onissi~s
.
Capture errors:
very familiar behaviors replace
1-C~ farnil;~r behaviors
Metier actual behaviors where
similar be rior so diverge.
Redid pale of
un~leted actions.
Minimize a~rerlaE~pir~
Warrior;.
Same:: Nomlan (1983, 1986)
experience or frequent ,,-~ of a Aster System ~#S
pooled need for insatiate f~dk (closure) ,
experience or frequent BEEP d~#s the importance of hen
limitations in information ~xssir~,
experience or frequent ~~=- Eases the in pact of sensory
overstimulation,
433
The characteristics on the left were incorporate into the Star user's
concepts] Gel. The characteristics on He right we attend to
avoid.
"me following main goals were ~r';ued In designing the Star user
interface:
firm; 1 far user ~ s conceptual Yodel
arming are pointing versus ruing ark
typing
Hat you see is what you get
universal cuds
consistent y
simplicity
model- interaction
user tai loran 1 its
"...We decided ~ create electronic count~rts ~ the physical
Objects in an office: paper, folders, file cabinets, mail boxes, ark
so on an electronic metaphor for the office. We hoped this Acid make
the electronic 'world' seen more familiar, less alien, and ~ less
crainir~.... We further decided ~ make He ele~z~n~c analogies be
concrete objects. Its wed be more than file naps on a disk;
they weld be represented by pictures on the display semen. Hey
wed be select by Pointing to then.... To file a dent, YOU
_ , , , _
_ _ · · a · a _ _ e _ _ ~ a a he
would mere it to a picture of a file drawer, just as ye take a
EShysi~1 pine-. of paper to a physical file cabinet."
N~A's Virtual E3nriromnen~c Workstation illustrates= a ~~ Are
avant-garde metaphor (Fisher et al., 1986~. This project wed give a
rdbot's operator the sensations and E~r~bive of the robot: Screens
~ the c~pera~r's helmet wed Than vie; taken by cameras on He
rat; sensors would pick ~ the ~erator's arm ark finger cements
ark translate then into retirements of He Let's arms; ark He
~rator's gloves Hula let the cooperator feel pressures Cat the
r~ot's fingers foul. (she Operator wend have the sensation of being
inside the rat, and the robot wed become an extension of ~
c~erator's arm ark hard Cements, ever though He robot might be maTly
miles from ~ ppera~r.
Although metaphors consti~ a fairly new frame of reference for
the designers of ~n~cerfaces, a designer or user can look upon every
interface as a metaphor of scmet}ling, ark thus the design issue is not
wrencher ~ at a metaphor but Cat metaphor to ad—. Each metaphor
has bud advantages ark disadvantages. As She's designers not, an
effective metaphor can botch r~c:e the Hunt of learning Cat
inexperience users ~ do ark accelerate Cat learning. Art effective
metaphor can also tap ink users' weal—evened habits and Sherry
reduce errors and ~ rinses; and experienced use as wok as
inexperienced users Than such i:~pmv~nts. For instar~, 7—gard et
al. (1980) slightly reedified a text - Star so that its Is
riled Short English sentences: me original, notation
RS:/~/,/OK/;* became CHANGE: ALL "JO" 10 "Ok', ark the notatior3al
carmnand FUlD:/l=IlI/ became FO~PRI:) TO "INCH". As Table 3 straws, such
434
changes improved the performances of fa Ply experienced users as well
as Inexperience - users.
TABLE 3 Text Editing With Different Command Languages
English-like
Cams
Users with less than 6 hours of experience:
Percentage of tanks completed correctly
Percentage of erroneous commands
Users with more than 100 hcNrs of experience:
42
11
Percentage of balks completed correctly 84
Percentage of erroneous commands 5.6
Notational
stands
28
19
74
9.9
.
souRcE: I-dgard et al. (1980)
But every interface metaphor breaks down at some point, both became
a metaphor differs from the situation it s;~1atPc and becalm=- an
inter face differs from the computer it represents. People in real
office= can take actions that users cannot simulate ~ Star's
electronic office, and Starts electronic office allows actions that
would be impossible ~ a real office.
.
Similarly, a robot might be
unable to reproduce some of its operator's instinctive finger
movements, and an operator in a shuttle or space station would lack the
mobility of an unconfined robot.
. '. . ~ ~ .
Yet, users are likely to draw strong
inferences about a computers capabilities from the human-cc=puter
interface. Ledgard et al. (1980:561) noticed that "the users made no
distinction between syntax and semantics.... TO them, the actual
commands embodied the Cantor to such an extent that many were surprised
when told after the experiment that the two editors were functionally
identify . "
Cone implication is bat an interface metaphor, Fib an interface
language, Id magentas ntentional artificiality in order ~
warn us ~ of itch li~nitatim s. Are sure of the ~nt~;ti~re exp ~ stir s
that users bring to metaphors especially important to fulfill? For
example, in design m g the Virtual Environment Workstation, might it be
essential to use cameras that closely approximate the spacing and
movements of human eyes in order to avoid having to retrain the
erator's s~ic vision? Under stress, people tend to rearers
fit - it specific, learned, complex models bac3` to generic, cc~mnon sense,
simple models: Which of the equations that users have unleaded
through tray d~= stress reawaken? Does stress, for instance,
435
Cream use' responsiveness to Crete, visible stimuli and
dream their r~nsiveness to abstract, ir~risible stimuli?
A so implication is that designers Chid carefully explore the
limitations of an interface metaphor before they adopt it, and Hey
. . . . . . —
snare 1003; upon a metaphor as ~~ choice flea a set of alternatives,
each of Rich has advantages arm disadvantages. However, the Citing
i~erfa~ metaphors have been developed separately, with considerable
emphasis being given to them union; arm the proses that
Merely ffmm have been poorly docents. So, interface ~-signer';
need to be able to Grate alternative metaphors, they row ~eptual
forks that highlight the significant properties of different
metaphors, and Fey row systematic research to dormant these
Plies.
* * *
All of the foregoing topics imply that a ~ should adapt both its
amoral and the Naples ~ pr~rmns to its use take amount, for
example, of its user's topical mortise, experience, fr~y of
Am, or mortal dexter)=. His ados for de~rel~t of
~ticated interface software (a Focally Use' Interface Management
System) that will Agonize He nerds of different users, allow
different users to express their personal prefers s, and print
use' individ~,a1it~r. Whys, He cater row to be able to identify
a user quirkily and Reciprocally, and if Foible, withal infusing an
identification posture that would irritate People or delay their
arts in an arty.
BOOKIE ADD CHIN END POEM
Efforts to justify apace systems in erratic tangs will keep pressing
for higher and high levels of Sable productivity, and so
plan; will tend to program the Orators' activities ~ detail. But
rarely heavy workloads Rise the probabilities of human error, and
Cuter; will always be better than people at working tiny and
obediently adhering to plans. PI pie contribute to ~ ace - Its their
able ity to deal ~ th the unexpected, and in fact, to create the
ure,4x~sbed by experimenting and innovating. -they can make these
contributions better if they are allowed some slack.
Space systems' tasks are not all located in space. Spa~. systems
inevitably make educational contributions that transcend any of they
immediate chelations g - 1s. C>ne of the major contributions of the
apace pr~3rmn to date has been a pat~ra~h - a photograph of a
clald-bede~ed h=1 ~ of water and dint isolated in a black void. Before
they saw that ~hotograp, people's undcrsta~i~ that Unkind shares a
con fate had to be attract and i~lec~l; the ~ih~rap ho
made this under~ar~i~ Ore partible and Iris.
People play central roles in Locational activities because they
serve as identifiable portals of referee In setting; that wood
otherwise seem mechanistic, rate, and alien. Another of the apace
436
p~rmn's major contributions, ~11~ it put Space exploration into
words schist caught the hen imagination, was Neil A. Art's
unforgettable Gestation: "mat's one small step for a man, ace giant
leap for mankind" (July 20, 1969~.
SPRY OF ENTICES AND LIONS EAR RESELL
Fostering ]~ Between People are Expert Systems
~ Pat ways dhalld a decision~support ~tem's kr~ledge arm logical
nines fit each user individually? Do users trust systems more or less
tailoring ~ ~=tpon~ u~il the users gain oonsid~ble
ex~ri~e?
How do tack characteristics affect a user's I ~ trust a
clecision~support system?
In what cir~nstan~s doe= a user decide to trust a Later system
that captures He knowledge of exams ^~ the user does not kr~w
personally?
Pat kinds of ex~rier~s lead a user to trust a deceit
Stem that the user r~s, at let partly, as a black-box?
Pat kits of experi~s Frye a user to ~ a dec~sion~port
pys~n's l;,n;~cations and to Override bad r = Nations?
Creating Useful Workloads
Does performing activities that calld be alienated actually keep human
ppera~rs up to date on the stales of a system, or do creators tend deco
work ~anistir~lly when they are performing routine activities? Do
human Aerators who perform activities. Hat card be automated remend
Are effectively to a Trials end beadle theta participation
update= then on the current status of the system, or does continues
participation ~ carats so familiar with He in and its
current status that. Fey mrericok ar~nalies and lack the Objectivity to
porx] effectively to a Curious emergency?
MESA should investigate the degrees to which an ability to discard
Ereconceptians varies among people and can be predicted or taught.
What have been the workload of experiences curing long stays ~
confined spare= such as Sealab, Antarctica, and nuclea' submarines?
Anticipating Human Errors
Research on error prevention might tunefully complement the current
projects on error detection and error tolerance. For many tasks, it
would be feasible to explicate fairly accurate models Of people that
would enable human-cecputer systems to predict and adapt to human
errors. An fact, models need nck be very accurate in order to make
437
useful predictions or to suggest Me adaptability to pepple's actual
Saviors it pay off.
Devel~pi~ Effective Interface Indulges
visually all s - flies of ~terfacela~ges have involved indivi~
people working on tasks Cat they could perform alone. Because Space
systems create strong social contexts, interface larynges that
aE~prox~te nag larynges may turn out to be such Are value In
apace systems.
Using Manful Interface MbtaEShors
Are scan of the ~ ;tive expectations that users bring ~ men hors
especially Important to fulfill?
Urger stress, people teal to revert from specific, learned, complex
models back to generic, ~n-sense, simple models: With of the
exhumations that users have unlearned through tray does stress
reawaken?
Interface c9P=igner'; need ~ be able to generate alternative
metaphors, they need cone ~ dual fee ~ rks that highlight the
significant properties of different metaphors, and they need systematic
research to document these properties.
General
NASA should develop a sophisticated User interface Management System
that will recognize the needs of different users, allow different users
to express the ~ personal preferences, and protect users'
individuality.
Is there a way for a computer to identify its user quickly and
unequivocally, without imposing an identification procure that id
irritate people or delay they ages In an emergency?
Since NA5A can choose fawn. a large pool of applicants, the extreme
capabilities of exceptional people are more important Man ache average
capabilities of typical people.
The people ^o Ate Space systems first receive thoralgh
training, so gear deficits of ~e~ri~ce sand be small. Nearly
all of the reseat ~ cat human ~ uter interaction has foals ~ on people
who lacked thorough training and who had little experience with
computers, so most of these findings may not extrapolate to the
well-trained and experienced operators of space systems. A here is need
for studies of well-trained and experienced users.
Avoid research armed at describing human capable ities in general.
Instead, test fairly realistic mock-ups of interfaces and systems, with
people go are as well trained and as able as real astronauts and
controllers.
438
Investigate Me sensitivity of perforate ~ to small
variations ~ designs: Do srnal1 design charges produce large Charges
in performance?
Both to improve Me quality of designs and to improve use'
acceptance of designs, experienced as~crona~s and controllers Chard
participate in the designing of interfaces and system;. Becalms= early
decision off constrz`m later m~ifica~cions, astronauts and
controllers should participate freon the Baird of any rear project.
AC~=I~1~
mis report has beer improved by constructive suggestions Fran Michael
Burlm, Janet l~kerich, Kenneth Inudon, Henry Incas, yes Milliken,
Jon Larry, Carte Webster, faith Weigelt, and Hazily W~lbers.
Or;
At, J. S.
1981 Pp. 97-99 In Brains. Behavior, and R~ics. New York:
M=raw~Hil1.
?.
Allen, R. B.
1982 Cognitive factors In human ~n~ction way cc~utert;. Pp.
1-26 In A. N. Badre and B. Shneid~man, ~~.,,~Directions In
H=nar`,/~ter Interaction. Norwood, N. J.: xylem.
Badre, A. N.
1982 feigns Shades for sequentially displayed information.
Pp. 9-193 tuna. N. Badre and B. ~neiderman, eds.,
Directions In ~nan/~tter Interaction. Norwood, N.
Ablex.
· —
Bennett' J.J ~' B., ~,-P., I, F., ~ ~
1981 ]hel=11=Lpr~ra~E;. Pp. 106-123 tuna. Backhand
E. A. Feigeriba~, ens., me Handbook of Artificial
Ir~ellig~x:e, Volume II. T - : Ante, Q=1 if.: Wi ~ liar
}~u~.
South, B. J.
1984 me pathology arm safety of weightlessness. Pp. 63-79 in
G. W. }math, ea., Space Safety and Porsche, 1982-1983. San
Diego, Waif.: African Astranautical Society.
B - raw, D. G., Mittal, S., arm Stefik, M. J.
1986 Eat i: perils arm praise. ~nicati~s of the
Association for Sting Machinery 29~9) :880-894.
439
Slyly, M.
1987
J., arx] Norman, J.
Sized pa hologi=~1 testing:
critique.
18(1) .
an ave~iew ~
P~h and Practice
Canfield Smith, D., Any, C., Shell, R., V=plar~c, B., arm Harslem,
B.
1982 Minnie the star user interface. I: 7(4) :242-282.
Collins, J. S.
1986 me Usefulness ofEs~tSyste=: A Mirage? ~us~ipt,
School of nosiness Administration, University, of
Wis~nsan-Milwaukee.
De Bacht~n, O.
1985 It is Sat it's used for~jcib perception and Stem
evaluation. Pp. 689-692 LAB. Sashay, ea., H~an~r~uter
Interuction~teract '84. I: Elsmrier.
De X~, V.
1986 Ironical assistant e to the orator in ~- of incident:
scan lines of thought. Pp. 229-253 in E. Holinagel, G.
. . _ _ _ . . . .
Martini, and D. D. Woods, eds., Intelligent Decision Support
in Process Ernrtro~ts. Berlin: Springer.
Bayous, H. L., and Bayous, S. E.
1986 Pp. 110-118 in Mire Over Mythic.
New York: Free Sass.
Lanais, S. T., arm I~u~r, T. K.
1982 P';y~logi~1 iTnrestigatic~ns of natural terminology for
Snare & query lar~uages. Pp. 95-109 ~ A. N. Badre arm
B. Shneiderman, Is., Directions in ~ranJ/O~ter
Thteraction. Norway, N. J.: Ablex.
Dutton, J. M., end Starbu~k, W. H.
1971 Finding C~arlie's n~n-time estimator. lip. 218-242 in
J. M. Rattan am W. H. Stazbuc~k, ens., Inter Si~rn~ation
of Ian BE rior. New York: Wiley.
Fisher, S. S., Ivy, M., Henries, J., arm Rdbinett, W.
1986 Virtual E~ironff~nt Display System. M~s~ipt, Aerospace
Human Faders Reseat Di~risic~n, NP5A Is R - eared Center,
Mbffett Field, Waif.
Green, T. R. G.
1982 Picture; of patrons am other processes, or he to do
Dims with lir=;. B~haviour and Information Technology
1:3-36.
440
Green, T. R. G., Sime, M. E., and Fitter, M. J.
1980 me pebble; the pry faces. Er~onc~mics
23(9) :893-907.
Gamin, J
1986 Designing In the dank: logics that compete with the user.
Pp. 281-284 In M. Man~i arm P. Or~ton, ads., Roman Factors
In Grating Systems CHI '86 Conference P~:r~. New
York: Association for Outing Mushily.
Gru~felder, T. M., armful whiten, W. B., II
1985 Augmenting generic reseat with prototype evaluation:
experience in applying generic research to Specific
pros. Pp. 913-917 In B. S6ha~, ad., ~nan~uter
Inte~ction~Interact ' 84 . ~sterd~n: Elsmrier.
Jarke, M., Turner, J. A., Stahl, E. A., Vassi ~ Ian, Y., mite, N. A.,
and Mi~hiels~, K.
198S A field Aquatic of natural language for data retrieval.
~:~-~; ~ansacti~s an Software Engineering SE-11(1):97-114.
~d, H. F., Whiteside, J. A.J Singer, A., and Spur, W.
1980 me nay language of interactive ~ems. ~nications
of the Association for ~T~utin~ Machinery 23~10) :556-563.
Iubbins, A. S., arm urchins, E. H.
1959 Rigidity of Behavior. Eugene: Un~versi~ of In Books.
Mbray, N.
1986 yelling cognitive activities: human limitations in
relation to ~ter aids. Pp. 273-291 in E. lIollnagel,
G. M - Mini, and D. D. Woods, ems., Intelligent Decision
Support in ~ss Enviro~ts. Berlin: Springer.
Newel, A., Gbaw, J. C., arxt Simon, X. A.
1957 Empiric explorations of the logic theory machine: a case
sly In heuristic. Joint ~ter Conference Western
inks 11:218-230.
N = ~ D. A.
1983 Design rules bask ~ analyses of human error.
C~nications of the Association for Grouting M~c~hin~y
26(4) :254-258.
1986 New views of information pressing: implications for
intelligent deposit support systems. Pp. 123-136 In
E. Hol~gel, G. M~ncini, arxID. D. Ways, ~c., Intelligent
Decision Suborn In Pr~ess E~vir~nts. Berlin:
Springer.
441
Payne, S. J., Sine, M. E., and Green, T. R. G.
1984 Perceptual structure cueing In a simple caner language.
International Journal of Mdn-M~dhine Studies 21:19-29.
Reason, J.
1979 Actions ret as pearled: the price of autc~natizatioa,. Pp.
67-89 In G. Underwood are R. Stevens, - a. ~ Aspeccs of
Consciousness, Volume I. Unwon: Academic E=ss.
1986 Recurrent errors in process environments: some implications
for the design of ~nt=1ligent decision support systems. Pp.
255-270 in E. Hollnagel, G. Mancini, and D. D. Woods, edge.,
Intelligent Decision Support in Process Environments.
Marlin: Springer.
Rouse, W. B.
1981 Human-computer interaction in the control of Dynamic
systems. Computing Surveys 13(1):71-99.
Schneider, M. L., Wexelblat, R. L., and Jende, M. S.
1980 Desi ~ control languages from the user's perspective.
Pp. 181-198 in D. Beech, Ed., Command Language Directions.
New York: North-HollanS.
SenSe m, J. W.
1980 Is there a cure for human error? Psychology Today
14~4~:52-62.
Sheppard, S. B., Bailey, J. W., and Bat ey, [Kruesi3, E.
1984 An e`~rrical evaluation of software documentation formats.
Pp. 135-164 in J. C. mamas and M. L. Schneider, ads., Human
Factors in Cb=puber Systems. Norwood, N. J.: Ablex.
Sheppard, S. B., Kruesi [Bailey], E., and Curt is, B.
1980 The Effects of Symboloqy and Spatial Arrangement on the
Comprehension of Software Specifications. Arlington, Va
General Electric CcE pan i, IR-80-388200-2.
,. .
Shneiderman, Ben
3986 Seven plus or minus two: central ~ sues in human-ccmputer
interaction. Ep. 343-349 in M. M~ntei and P. Orbeton, ed=.,
Human Factors in Computing Systems. CHI '86 Conference
Proceedings. New York: Association for Cc mputing
Madhinery.
Simon, D. K., ark Sit, P. A.
1985 Human factors: an exploration of the psychology of
human~ter dialogues. Pp. 49-103 in H. R. Hartson, ea.,
Advances In Human~r~uter Interaction, Volume I. Norm,
N. J.: Ablex.
442
Simon, H. A.
1960 the corporation: will it be managed by machines? Pp. 17-55
in M. Anshen arm G. L. Bach, "3s., Management and
Colorations 1985. Evanston, Ill. : Harper & P=w.
Stanfill, C., arm Waltz, D.
1986 Tears ~ry-bas" reasoning. C~nications of the
Association for Garroting machinery 29(12) :1213-1228.
[bang, H. D., arm Gupta, A.
1982 E=s~al Inters. Scientific American 247(6) :86-107.
I=ner, J. A., arm Harass, R. A.
1984 Software ergonomics: effects of Inter application design
parameters on Orator task perforate arm heals.
E~onami~ 27 (6): 663-690.
Valiant, L. G.
1984 A theory of the lean - le. ~nications of me
Association for Outing finery 27(11) :1134-1142.
Warm, H.
1963 TO mechanical mathematics. Pp. 91-120 in K. M. Sayre
arm F. J. Crimson, Ells., me Modeling of Mire. Natre 1~3,
I.: University of NOtre name mess.
Watzlawi=, P., Wealdar~, J., arm Fisch, R.
1974 Carte: Principles of Prciblen Formation arm Emblem
Pesol~ion. New York: Nortan.
l
Weiner, E. Le
1985 Band the sterile Spite Collar l~;tacto~ 27(1) :75~90.
Weizenbaum, J.
1965 F~.TZA—a Or pr~rara for the stem of natural lar~uage
~ nication between man arx! marine. ~nications of
the Association for Outing Mythic 9(1) :36-45.
1976 ~ Peer arm Iran Reason. San Fears isco: F= .
Winograd, T., arm ~ores, F.
1986 U~star~i~ is arm Ignition. Norway, N. J.:
Ab1ex.
Woods, D. D.
1986a cognitive technologies: the Sign of joint humn-madhine
Lenitive Me AI Magazine 6 (4) : 86-92.
443
1986b Paradigms for intelligent decision sort. Pp. 153-173 in
E. Holl$,agel, G. Mdncini, and D. D. Woods, "c., Intelligent
Decision Support In less Er~viroTments. Poplin
Springer.
