I»

e

m

ftiarnM ot MUIMY itan Mnms « k iwo-AiTOHutm mMcnoi UM

i

D D C

U k B

Behavior and Syttemi Research Laboratory Office of the Chief of Research and Development

U. S. Army

February 1970

Th.t documant hit b««n .pB,ov.d »or pubhc r.l.... .nd ..I.: 1» dLUibonon !■ unhm.t.d.

dip

·•·

THIS DOCUMENT IS BEST QUALITY AVAILABLE. THE COPY

FURNISHED TO DTIC CONTAINED

·A SIGNIFICANT NUMBER OF

PAGES WHICH DO NOT

REPRODUCE LEGIBLY,

[

ftKifflOi of «UTUY mm nmmti m A TWO-AinilATIVi KOKTIOi fASI

• ■■»*► «

KHAVKM ANO SYSTEMS «ttAACN LAMAATORV

OH>c» C•»■•< ef I—i and 1—i^—1 OcpsrtMsnt o* th# Army

Roon 239. Th« Co..i«nonwe»lth Building 1320 Mfilaon Boulevard. Arlington. Virginia 22209

February 1970

Army Project Number 20024701A723

Tactical Operations Systems c-12

This document has been approved for public release and sale; Its distribution is unlimited.

I

»otiwoto

otnmnn mt

<* 9m TACTlCM CMRAf KM mTB« TOt) »of« u«.i •• » pio- •c« b* liicn #KiftiOn aikiNt ««d Mfomatioii ••••«maiKin »to»

r to ttciiitatad Tto prM«nt »»uis aHU «HU c*t«in aapactt of twrcacxion 10 KW ••cogn.l.on of MOHSNCO pail*«n« of «iliMTv OCtivMiO« ThO ObiKliv«

Ol 4010 utofwl •" oHon« io MNprovo lt»r rocognilion of Mdi KM •«»•cipoiion of

Behavior and Systems Research Laboratory

Picrtoi 9 mit«* ivnit pAttwm m A raoMfiMAtivi Mü

•till

• %mmh —p— <* cm» t a«» nwpinw of |WW «^ ••'

•f ■Hill. Mr«fMl IPNMfetHUr «f «nvrvMl Mi tmn»mtf\i tt bt ••«• «BiHM «f «p»t«M flw mjKiiMi» kM Mi •■»» ma p«|i

^«•dü»»

of »v»«!» rapTMOTliiig tiflfil pMleni« ifom). digit «ni low pomm iirangtti conimuou« vt iiMrtto ooctfrwco. ani oipononco fram on« to ton 10D-triol parwdt «oro prooontod on o CftT acrotn in tyotomtlic dMign to 48 «tlittod «on. Tho «nlitloi mon indicatod which of two WWHIV octivitioo (1110011 or rott) wot likoly to folio« ooch two-ovonl duo. Rotults in tho form of prodiction of the moot froquontly occurring third event in 0 toquonco and confidence (high or low) npretsed in tho docition wore »ubioctod to onalytit of varionco.

Findings

Given a pattern of events occurring with sufficient frequency (80% of the time) and sufficient experience with the patterns on the part of the decision maker, the man learned to predict the third event in a sequence as often as it occurred. However, whan the pattern occurred less frequently (68% of the time), the men recognized only one of four critical patterns.

Subjects' confidence in their predictions increased as their experience with the task increased. Confidence was affected by pattern formjn that aubjeeta' eenfidonee in their predictiona for one of the eight fenwa (reat, reat, attaeh) was signifieently loaa than far the elhore. Confidence was not affected by pattern strength nor continuity.

Application of Findings:

Activity patterns of an enemy may be the most tangible cues to his plans, and recognition of these patterns may be critical to successful military operations. The present experiment has set tentative limits on the event patterns men can learn to recognize to a useful extent. A computer may be a useful aid in detecting weak but significant patterns in an apparently random set of enemy actions.

o» ttuu» iviNt nnmm m A naMttvativi MSCMIION »MI

CONflNfS

OUfCTiVCt Of TNI

yrTHOo

» mio»

Subjects Apparatui and It laut I Procadur« Independent Variabl«« Dependent Variables Ovtign

RESULTS

Prediction Data Confidence Data

CONCLUSIONS

IMPLICATIONS OF FINDINGS

LITERATURE CITED

DISTRIBUTION

DO Form 1473 (Document Control Data - R&D)

1

4 4

J € 1

I 11

15

If

17

19

HOl*fS

1. fr«fMwy Mi ««•ililMMl prvtetttlty «irvcinr« for

f. tap^risMUl 4—ttß 7

)• kmlfi» «f ««rlMK« tmtmry cMlt M ^rMlctio« 4«c« 6

4. An«lT«l» of »«f 1«IK» MiHMry l«blt M cotif idvnc« d«l • 14

% Casparl»a« of •«•r««* prediction «ad confldMK« for ••<h p«ti*rn fora 14

1. Pr«dtctloat of «trongor p«tt«rn» tcroa« poriods •■ • function of pattern strongth 9

2. Proportion of predictions of stronger second-order patterns scross period« 10

5* Proportion of predictions of stronger second-order patterns as a function of continuity 12

4. Proportion of predictions of stronger second-order patterns for continuous vs discrete presentation and for each pattern strength 13

^

^CRCC^flON 0» «UfAUT CVfNT HUimi IH A MK) AUIANATiVl HKOlCrKM TA»

fitlwrt •WI<MMI«4 4»f ^roc«t*in| tytmm» proal»« la^rovad accuracy, coa^lttMtM, «ad tp««d of tnforMtlon procaislnf. Th«M ifimm wmy • l»o kt cipablc of llaliod data tvaluocioo for doclaion Mbing. Such • •yacoa hat boon propoaad by Muardt ^1). In tha auloaatad tyttmm, aach now ti«o of lnfor«acion la avaluatad and antarad into a probabilistic infotaatloa procoaalng ayacao «bich updataa tha currant dlagnoaia on tha baaia of tha now tnforaatlon. Sine» ayaco« input* ori|lnata fro« «any ■on, tourctt. and locatlona. aach ita« la likely to ba tvaluatad indo- pondontly aa It ontara tho ayatao. Frequently, however, tha Intorrala» tlonahlpt of itaaa of inforaatlon aay add dlagnoatlc valua beyond that obtained by aoparata and auccasslva avaluatlona of Individual icana. for example, ton« »der two item«, an intelligence report of a naaalva tncraata in tha number of enemy troops participating In military exer- claaa, and do«oaclc nawa of tha assassination of a president.

Each of thosa events may have on« meaning if thay occur aaparataly. Their Joint occurrence may b« a strong Indicator of enemy aggraaalon. Sine« the two event» hava been reported from d Iff «rant sources, thay may ba evaluated separately by an automated Information processing system. The comnander, however, may well Interpret the Joint occurrence of theaa events as a pattern leading toward aggraaalon. He may "perceive" a threat and conclude that the enemy is about to attack.

Needless to say, this ability requires experience in observing the activities of the particular enemy under consideration. As a commander or decision maker gains this experience, he renders more accurate Judg- ments concerning present and future enemy activities. In a more analyt- ical manner of speaking, he is learning which aspects of the enemy's activities are relevant cues in predicting future enemy events and which are not. By learning which cues are associated with which outcomes, the comnander is learning the enemy activity patterns.

A well-known probability learning paradigm, which has served as a useful vehicle for investigating many facets of human learning, requires the subject to predict in a series of trials which of two lights will come on. While his first prediction can be little more than a guess, succeeding choices can benefit from knowledge acquired through observa- tion of earlier outcomes. Thus, his sec md prediction is made in light of what happened on the first, his third in light of the first two, and so on. The second and subsequent predictions, then, may reflect his perception of one or more cues which are present within the experimental environment. A cue may be defined as any stimulus or set of stimuli which is correlated with the occurrence of an event (e.g., one of the lights). Thus, any cue affords the subject the opportunity to improve his prediction accuracy.

Tb« pr«s*nc invtatiiatlon conccmt th« Itarnlng of «vAtit »cructur«, that It, th« orderly occurr«nc« of «vtnts within th« «nvlrorawnt. and th« cu«s th« structurt affords. A «Miralnal cut la «n «v«nt In A t«rltt of «v«nta which occura «or« fr«qu«ntly than any o'.htr «v«nt In that ttriat. A firtt-ordtr cut la an «v«nt whoa« occurronc« algnalt what tht ntxt tvtnt will b«. For txanpl«. th« tight of t lightning fltth it t cut which aignals tha aound of thundtr. A atcond-ordtr cu« it a pair of tvantt whoat occurrtnet aignala tht tvtnt which la to follow. In baaa- hall, If th« firtt two pitchtt to t batttr trt atrikaa, th« next pitch ia utually a ball.

Of coura«, th« cuta obatrved In nature and man ar« not alwaya rail* abl«. The aound of thunder from distant lightning flaahea la not always heard. A ball doea not alwaya follow two strikes. However, aa long as aound follows lightning more often than silence follows, and aa long aa a ball follrws two strikes more often than a third strike, the cues are valid.

Previous probability learning experiments have Investigated man's ability to respond to various classes of cues (marginal, first-order, jecond-order) of varying strength. The results of these studies show that subjects are capable of learning marginal cues, as evidenced by the fact that their response frequencies generally match the relative fre- quency of the respective events over trials. For a review, see Luce and Suppes (2). Subjects also match or exceed matching (overshoot) events signaled by first-order cues (5) (4) (5) (6). There has been little research concerning second-order cue learning and the findings are not straightforward. Data from Bennett, Pitts, and Noble (7) suggest that In a five-event environment (the occurrence of one of five lights), subject:, are unable to learn second-order cues. In a two-event environ- ment (the occurrence of a left or right light, L or R), Strub and Erickson (8) found that subjects did learn certain highly structured second-order cues. Subjects tended to match their response frequencies to all second-order events when conditional probability was set at .92 (a value indicating the probability of the occurrence of an event, given the occurrence of the two preceding events). When it was set at only .72, however, subjects seemed to take the specific form of each second- order event pattern into account. The four more frequent patterns were obtained by combining all possible second-order cues (L-R pairs) with an L or R third event as follows: LLL, RLR, LRR, and RRL. At the .72 level, subjects overshot LLL, matched RLR and LRR, but did not learn RRL at al1.

In a frontless war such as that being fought in Vietnam, the activity patterns of a local enemy may be the most important clues to his position and intention. Therefore, the timely recognition of these patterns can be critical to successful military operations.

■Ma

OBJECTIVES OF THE PRESENT STUDY

Sine« vary littl« la known concerning military pattern perception, ruMrch at a relatively bailc level Is needed to answer some fundamental queatlons. The present atudy addreaaad Itself to the following questions: First, does the form of the pattern affect the subject's perception of It? For example, If an enemy pursues a course of repeated attacks after rest- ing two days (where eacli day's activity, rest or attack, constitutes an event), a second-order pattern of the form reat-rest-attack would be ex- hibited, la this form any more or leas difficult to recognize than other forms such as attack-attack-attack, attack-rest-rest, or rest-attack-rest? A second queatlon concerned the range of pattern atrength vlthln which patterns are recognized. Assuming subjects can recognize s cond-order patterns of >?f strength (&), can they also recognize them at 80^, are any patterns recognizable at 68^7 Third, does continuity affect second- order pattern learning? Is It more difficult to discover second-order patterns in a continuous flow of events than in discrete second-order pattern unite? Fourth, to what extent does experience, the repeated ex- posure to the patterns, enable the subject to perceive the relevant second-order patterns? Finally, does the answer to any one of the above questions depend on the answer to another, that is, are there significant interactions among the independent variables under investigation in the present study?

In summary, the objectives of the present study were as follows:

1. To determine the effect of different forms of second-order patterns on pattern recognition.

2. To determine the effect of second-order pattern strength on second-order pattern recognition.

J). To determine if continuity affects second-order pattern recognition.

4. To determine the effect of experience, i.e., the extent to which second-order pattern learning occurs over periods within the experiment.

5. To assess any interactions among two or more of the above vari- ables (form, strength, continuity, and experience).

METHOD

Subjects

Forty-eight enlisted men from Ft. Belvoir, Virginia served as sub- jects. All had GT scores of 110 or above. Many had served one year in Vietnam. Average length of service was estimated to be 13 months.

- 5

Apparatus and Stimuli

Six cathode ray tubes (CRT's^ were used In the present study. Each CRT consisted of a vertically tnourted televisvon-like screen on which stimulus material was displayed and a horizontally mounted typewriter keyboard at the base of the screen on which the subject typed his re- sponses. Each CRT screen was positioned at eye level. The CRT was linked to a computer which was programmed to generate separate event schedules for each subject prior to each session and to coordinate the simultaneous use by six subjects.

Event sequences were prepared as follows. The computer selected events without replacement from a 100-event population in order to insure that the exact probability specifications were realized after every 100 events. The specific order of the stimulus events for each subject was determined randomly at the start of each experimental session. Thus, while event sequences for subjects in each group contained the svne con- ditional probabilities, the exact ordering of events from alternative to alternative was different for each subject.

Procedure

Six subjects worked individually during each txperinental •••sion. Prior to the first session, they were briefed concerning the general nature of the experiment decision «eking*. The/ were also told not to disci.sa the experiment among the«eelvea until its conclusion. Each «an was then assigned a Oil and each recelwod the following Instruct Ions on his scroon:

Till» ts an experiment in basic ■Hilary dtclsto« mailing. Your cask will bs to decide which of two sctiviilss will occur. ihSi Is. lo predict whether enemy X is going to (I) sttsck or j rssi. You tailcsis your decision by pressing s 1 or ? on your koybonrd sni then pressing tbn "send" but can. flsst yon «111 Ind tests your conlidencn by pressing sni "sondtag** If ynu ikisk you Hsvo bot tor Ibnn • ^0*%> ebenes of being correct, or ? If your dnct- sinn uns. gsits frsnbly, s gunss. Hsat. you will bo In*

M to ubtcb nf tbn cwn sctivitiss did, in fnet. Tbs asst pugh of tbn "snni" kutton mil start n

now docisien eiinsilon. fnllnw tbls sens procsduro for Ihn oncontf sad subaannani dnciatons.

in two ospanmnsMal sasetana. «sming an« after- nsistad of fiaa ! »--cbmica partada witb a lo<nineta nslnly tn ret town aye strata prodncad by tha «an- display. tacb anofact uarbad si bis mm pnen gal

kraak iimo. TW «nmiag taaatmn ostemaod ffgna spprosinataly «• la Pb tba ffawlnaina nf iba

nf IW

tba sfioma«« aaaaisn fmn l*or m l^rv>. At imam, anb|ncca uava dabriafad ««nsaming tba

In «hiah tnny nad garticipaca4.

• «

mmmrmmmmmm

Independent Variables

Form. The first Independent variable was the form of the second- order pattern. Table 1 Illustrates the eight possible second-order pattern forms. Event sequences were constructed such that the forms AM, RAR, ARR, and RRA occurred more frequently than AAR, RAA, ARA, and RRR fA » attack; R - rest).

Strength. The second independent variable was the strength ( fre- quency of occurrence) of the second-order pattern. Two strengths were employed, .68 and »dot These percentages represent the second-order con- ditional probability of occurrence of the second-order patterns. ("Under high strength, £iven that the last two events were A, a third A would follow 2^ of 25 times {B0$)t while R would follow J of ?•) times (20%). Thus, the second-order conditional probability of AAA and AAR was .60 and .?'. respectively.] Table 1 indicates the structure of the second- order patterns in terms of both frequency of occurrence and conditional probability for high and low strength.

i

Table 1

FtEOTCMCY AMD CONDITIONAL PtONABILlTY STRUCTVU FO« EACH I* -TRIAL CVOTT SCQUENCI

for« Preqweocy

.*-0eys Yesierdey Todey

llllMMl Prokebiltly

tlrMcih mg»»

1^

if

•

Continuity. The third Independent variable was the amount of day-to- day continuity. For the continuous case, events occurred consecutively over 100 days; thus, the event which was displayed as having occurred yesterday was the correct event on the preceding trial. In the discrete case, reports of the previous day's activities did not follow a temporal order beyond three days; the subject was simply shown the reports for the past two days and requested to choose the next event. A fresh set of reports for the past two days was then presented.

Experience. The fourth independent variable was experience, which consisted of ten 100-trlal periods.

Dependent Variable«

1. Predict ion-»The proportion of tines the subject chose the «ore frequentIv occurring patterns. If a and r are choices, the choice score equals AAa v RAr -t- ARr ♦ RRa/100.

2. Confidence--The proportion of cisNs the subject indicated high as opposed to low confidence. If 'V is high confidence, the confidence score equals AAh ♦ RAh ♦ ARn • Rth/lQO.

Tnkl« ? HUsiretes the •«^•rievntel design, tatcnrn eirengtb •«^ pattern condnuiiy, the two vnrinfclee edninietered et c*n leenle, uere kniu««»>a«k)ect eerteblee «ilk 12 nnkjncie wllMn MCh of «lit Um »•• •niltag eapnrineniel gfnnf«. ftutnrn Core end «ngnriMK« «nre ulilil»- Mb|««« venebie». eecb •«%)•<( mcniving nil fnvr M<<«C« t- enrving in nil tan »nrtete.

im «Mlyni« nf «triMM «I Ik* frniKtinn «nie «on pnctomatf Ml e •t IM« tmlfl* m«in in tetln «. «Ml« •i^itlcfta« «em

•tffM«t •• low», •«mngifc. MMi«ei«v. mä mtttimm ptwi4o infMai r«l«MM «o IW tire« tow e*|Mg««M nf «fet mmtf, Ik» intlMMie«* nf •Mk 9i IkMt »ftMl» tn ll««l«4 > «Mk iMofM«n4 eignitueMly W««k M IMM mm o«Wf ettM«. fke«. «k* tlt«k ok}Mll«« «f «Moeeing iMnf««ii«M «Mat •»• iMlikl— «MI W ■MRHirti legiriw« la V«M at ika «lMltl«Mi laiatatiiaM la «ao aaalfait HMM la tkkla 9*

8

5 1

SI

M •

W ■

i

ill

3 3

o o

i

| . . ,

I

S5 5 ■ I I I...

SI ! i ....

l 1 * - 4

1 l . . .., 2 • • •« S * ' •«

i i

i i

»I1 i I

2 : i

'I1

• 7

T«blt 3

ANALYSIS Of VARIANCE SUN1ARY TAAU ON PREDICTION DATA

Source M MS r P

S«tw«*n Subjects (#?1

Continuity (C) t B C fau/SC

I 1 I

44

8044.?ie 146.406 470.0V 2^.6T2

M.15 4.oe

.001

.09

WMhln S«*)*€tB tl&t*) for» rr) r B • r BC f B 1 B c r B |M/tc

9

• HB.i^O

TtVÖTO 900.081

10.09 i •* ..v ».49

.001

.01

.10

■B^fflMC« (|) Cat 1 a C 1 a • a C I a |M/K

* »

las.ao«

14.99»

It.w • >

1.19

.001

.10

fat f Blat f a 1 aC f■1aS a C fata |a»/fC

n 91

4».19T 10.941 19.9W

9^4 1.48

l.lt

.10

To fMllltain iMnrpcntmina nf tka ■ apart—a» a airantta tntut* •ell« (p< J0ö\), «MB KBiKtia* «coma for all pttimrm form acroaa foriol» «ara »Imtoi M a fowtiao «f pattor« at rangt ii figora I), Mt* •itMmia prnfortlMM rnfftoct • «Inn k«c atnnif incroaaa ( p»rin4a a«! nan to • Inool «Anra praiictinn of tW atrontar patt« •ro nMdiiat tWir orofortin« nf WWW («ID). TW Ian atcnntia pottoma. an ckn otnor knntf. «aliikit littla iniicntion nf aacnni-nrinr raiapiltian inplnnt ikM tan pattama am unlaamakla naon thn con* 4iii«aal prakakility nf tknir natnrtancn la aM M .C8 nr tola». Ulla inpl teal inn nnat to l«npnm4. tonnaar to I to fnm a nnpnrianra intnr- aclinn | f * .'»I) Ntonn in lakla * Intnrnmtntinn »f tkia affnet «ay to win to laapatllnn nf ftmn I« It la annarant frnn fifara ? ttot all tmnpa adiianai a lovnl of onntary af tto atta<b*attn<a«attack pattam ttot naa ant raitfcto far any nttor pa*tarn. In aMIllnn. thla lavnl fnr MA «na attainto aftot < • tnala. aftar ufcica pnifonMnca laualto off.

•

90

80

i TO

•0

90

40

.*'*

.*''

^ ^^y

^_ i i i i I

- T..«i

%~**\ .«•% of %ltd«9»> p«llt»n« jctm* pMitii M « I^Ktion dl (MII«*II %lta«iq|h

.90

eoL c

s s

i I I Ml

.*''

^X

V> /* ßr - xx—^

h / V /

A" *" x V. ,'

40h

t. A AAA ■ ***

• AR» M MA

■L i

f.^*? tapomon oi pradKiioM

4 «

100 - T"0' P««>«tf»

•0

:

The form x continuity interaction of Table 5 fp < .01) is plotted in Figure 3-' Note that the means for groups presented the continuous sequences were above those of the other groups for all patterns except rest-rest-attack whose prediction was less frequent in the continuous groups. A further isolation of the rest-rest-attack pattern is shown in Figure 4 which traces the patterns across levels of continuity as a function of pattern strength. Here, it is apparent that the decrement in predicting the rest-rest-attack pattern occurred only at low strength, although the form x strength x continuity interaction was not significant (p < .08).

In summary, while form, strength, continuity, and experience all had significant effects on the proportion of predictions of the stronger pat- tern, each of these effects was involved in one or more interactions. The form x experience interaction is interpreted as a tendency for attack« attack-attack to be recognized better than any other pattern across the ten 100-trial periods of the experiment. The strength x experience inter- action is interpreted as a tendency for all patterns to be better learned in the high than in the low strength conditions across periods. Finally, the continuity K form interaction is Interpreted as due to the tendency for three of the four second-order patterns to be better recognized In the continuous than In the discrete sequence, while the fourth pattern, rest-rest-attack, was better recognized In the discrete sequence.

Confidence Osla

The dependent variable used to estimate confidence was the propor- tion of tlae the subject Indicated a high rather than a low confidence. Table 4 is a sumary of the results of an analysis of variance performed on these data. The analysis did not yield as many significant effects •s the analysis of prediction data, but the significant effects which resulted complement those of the analysis of predictions.

Neither the main effect of atrengih nor that of continuity was significant, indicating that confidence In the correctness of choice was not affected by pattern strength or continuity. As indicated by the significant effect of experience p- . '\ confidence increased across periods. This finding is not surprising, since subjects displsyed in- creased skill in pattern prediction scross periods.

The main effect of form p < ,001' agrees with the prediction analys s in that degree of confidence expreesed by subjects reflected the saae ordinal relationship as pattern predictions. This relstlonshlp «ay be observed in Table which presents data combined across periods, •trentths and continuity levels The conclusion drawn from these data Is that subjects experienced varying aawunts of difficulty In mastering esch pattern (of«.

::

f o Q.

•» c o w 4- v>

I

I

.90

SOW

.70

.60

.50

40

i

A AAA ■ "AR

• ARR X RRA

/N

x

Continuity

Piqure 3 Ptuportion ol pradiclion» ... suong^ »erond ordw pailern» M ■ luncnon ol connnoiiy

I? •

»W 11

■s

I ♦- u

I §

I

.90

.80

|. .70

.60

.50

40

t

/ >--

D'»c'ft» Contmuout

Continuity

▲ AAA ■ RAR

• ARR X RRA

— High Strength

Low Strength

f II|.II.- 1 P'upurlion o( prinliclions of stfiKM|»'r second ofdn pMftnM fw cnnlmutMJs vs discrelc pr»S(>nl.iliuii .imi l(»t »>.u h pattern slrm|lh

l •

Table 4

ANALYSIS OF VARIANCE SUMMARY TABLE ON CONFIDENCE DATA

Source df MS F P

Between Subjects (47)

Strength (S) Continuity (C) S x C Ssw/SC

1 1 1

44

117-515 555.775 22.750

1116.791

Within Subjects (1872)

Form (F) F x S F x C F x S x C F x Ssw/SC

5 5 5 5

152

814.124 231.178 258.904 68.955

105.401

7.72 2.19 2.46

.001

.10

.10

Experience (E) E x S E x C E x S >: C E x Ssw/SC

9 9 9 9

596

185.964 12.919 19.509 26.205 29.070

6.40 .001

F x E F x E x S F x E x C F x E x S x C F x E x Ssw/SC

27 27 27 27

1188

9.165 10.596 9.458 7.807 9.156

1.00 1.16 1.05

Table 5

COMPARISON OF AVERAGE PREDICTION AND CONFIDENCE FOR EACH PATTERN FORM

Pattern Fo cm AAA ARR RAR RRA

Prediction

High Confidence

79*

75^ .

52*

61*

58*

64*

69*

68*

14 -

.J

CONCLUSIONS

The present study provides some insights into the ability of deci- sion makers to perceive patterns in a series of events. A very simple pattern such as repetition of the same event is easily recognized at high and low frequencies of occurrence. More complex patterns require a fre- quency of occurrence greater than 68$ before they will be perceived. Finally, pattern recognition depends a great deal on the amount of con- tinuity in the event sequence. Patterns are better recognized when they occur continuously over time than when they occur in discrete units.

IMPLICATIONS OF FINDINGS

Two implications derive from the present study: Man is capable of learning second-order military event patterns when they occur with a fre- quency of 60$ or higher. Subjects exposed to the high strength patterns initially predicted their occurrence only 60$ of the time (Figure 1), but by the conclusion of the experiment were predicting the second-order pat- terns as frequently as they were occurring, 80$. Thus, if the event pat- terns are of this strength, complex automated systems are likely to add little to performance in terms of pattern recognition. This implication was also drawn by Howell (9). On the basis of the results of several cornmand and control system simulation experiments, Howell stated that systems which habitually handle predictive data of high strength have little to gain from automating the decision process.

In situations in which the patterns are of low strength, however, computer aids would be quite helpful. The present findings indicate that persons are unable to recognize three of four patterns occurring at 68$ strength. Only one pattern, attack-attack-attack, was recognized at this strength level. Thus, subjects were not able to take advantage of a considerable amount of predictability present in the .•ent sequence. It may well be that many enemy activity patterns would occur at a low strength level to capitalize on man's difficulty in recognizing weak patterns. Computers could be programned to perform many complicated statistical analyses on enemy events and to detect weak but significant patterns in an apparently random set of enemy actions.

1^

U1IIATUII ClflO

f. UM«. I. D. Mi f. ifpii. rf«f«rMK«. «cility. «ai •u»)«c«tv« prptabilny. tn t. 0. Uc«. I. t. IMII. Mi I. OftlMUff 'M«.). HM«4kvok of HaiH«Mtlc«l Nycheloty Vol III), «M tort: tUlty.

5. AndorM«. H. N. effect« of first-orior coailtloMl proteklllty I« • two-chotc« iMrain« tltiMtloo. Jtmrifl fii t«Mrt«wfl r%rtho\nu. lOiD. ^ ,*,5»

4. OiopaMi. J. f. Th« «poclng of oo^uontlolly dopondont irtcla la probability loarntng. Journal of taporfamal faycht loay. l>6l.

*. Englar J. Marginal «nd conditional stlaulua raapon»« probabilttta« in verbal conditioning. Journal of Cupari—ntal Paycholpay. l**^*, 52, 50Vn7.

Haka, H. W. and R. Hy*an. Parcaptlon of th« atatlatlcal atructura of a random aarlas of binary symbol a. Journal of Eupcrimantal Ptycholoay. \*\ g^ 64-7^.

7. Bonnctt. U. P.. P. M. Pitts, and M. Noble. Th« l««rntng of ««qu«n> ti«l dcp«nd«ncl««. Journal of Exporlwanf 1 Paychology. 1»V . At.

■*. Strub. M. H. «tvl J. R. Erlckton. S«cond-ord«r dependency In prob- ability learning. Journal of Exp«rim«nt«l Paycliology. HbS, Jß, ?€l-26r.

• . Hovel I. U. C. Some principles for the design of decision systems: A review of six years of research on a conmand-control system simuls- tion. AMRL-TR-t'-l^e, Aerospace Medical Research Laboratories, Wright-Patterson Air Force Base. Ohio, September I967. (AD 665 469)

- 16 -

mmamm m «ttttAtt tvwn mntmm ta A IWI HIB—tirt ftmaim um

«.•-.. « tir«»

rrrrrr «• *•<•. 3 •• •••«• 3 ■STTT^T

iiltf^' '^

* Tactic«! OpcraiiMM •ftt««*

« c-lf

T«CMIIC«I 1«—rill MM« VI

• »••»*«•«••••••«•••

T>>lt 4oc«aw Ulllt«ll«4«

»?• (or pabllc r*lMM Ml«; its 4iMrik«ci«i !•

TT-C»,rT=iTT777^77r

Offlc*. cM»r e( toMArcli «ad «••h. D. C.

At part of !»• c niinuin« «(fort to provido roaoorrh i«(or«stlo« «tolch will facillt«!* ■•xiaiMi oporatloo* offacti«•«»•• withiii tmmmmk infonMiio« proiasslag •jr«* tMM. th« TACTICAL OfOlATIOKS SYSTEM 'TOS) Work Unit un4»rtook a ttudy of parcoptloo of ■ilitary rv»ni pcttomt. Tho focut of th« »tuiy wa« tha rocosaltlo« of coafla« coo« to en*«y action. Objoctlvtt of tho invostt (•<><>*>, r«pirt«d in tho proaonl publ teat loo, wart to dottrain« th* rol* cf 1) difforoat fonw of Mcomd^ordor patttnw. 2) socood^ordor pattern •trangth. Z) continuity, and 4) «Kporlooc« •• factors In tho recognition of aocond-ordar pattema. In tha exparlaant. 4^* onltttod nan vlawad aa^uoocoa of ollttary event» rapraaanting eight pat tarnt (for»}, high and low pattarn ttrangth. rootlnuowa vi dltcrata occurrence, and cxparionc« (fro« on* to tan 100-trlal parioda) praaontod In ayttoMitlc datlgn on CRT tcraana. Tha aubjacta partlclpatod In two axporlaantal aaaalon^ each contlttlng of flva lOO-tnal pariodt *iih lO-ainufa Intarval braaka. lach aan workad individually and waa talf-pacad. Tha «an tndicatad which of two ana«y act ivitiaa, A(ttack) or R(eat), waa likely to follow aach two provioua event«. Thay alao roapondad to indicate confidence in thair dacitiont. Findingt «howad that givan event pattama occur- ring with high fraquancy (80^ of tha tit»), tha daciaion maker learned to pradict th* third taquantial event tt oftan at it occurred. However, under condition! of l*aa fre- quent pattern occurrenc* ( CM or lets of tha time), pattarn recognition did not occur: Only ona aacond-ordar pattarn waa pradictad aa oftan at it appaarad. Confidence in pre- dicted dacitiont wat tffactad by incraate in axparianca and pattarn font but not by pat- tern ttrangth nor continuity. It appear« from tha praaant findingt that tentative 1 imitt are tat on tecond-ordar military pattarn recognition. Man can perceive strong aacond-ordar pat tarnt. Computer aidt would be quit« helpful, hqwevar, in situation« where nd 1 it ary event

BS IM*VS«l4/«# *' M*k**M •* eWM «•»•. • it» M. «WCN ■•

' ITS e*« «MM* uet. . i > . Unclaaaifled iM«fily CU*«lflc«lU«

•MU«*ltl*i

art roMto«« tey Twto)

»yi»»»iluy lMnii«t

•Kit»?« fM^ii«M«) p*«il<tlM

MllllMTf »«ircWIaty

Military MMvlty »•««•*••

t(«il*ttc«l «Mlyal»

laforMiio» 4iMUy«

T«ctlc«l 0»*r«tioM Sy«c«M (

». • •• .-•♦ •

I »*•

- 20 - Unclassified

J