General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

Users may download and print one copy of any publication from the public portal for the purpose of private study or research.

You may not further distribute the material or use it for any profit-making activity or commercial gain

You may freely distribute the URL identifying the publication in the public portal If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from orbit.dtu.dk on: Sep 14, 2020

The human data processor as a system component. Bits and pieces of a model

Rasmussen, Jens

Publication date:1974

Document VersionPublisher's PDF, also known as Version of record

Link back to DTU Orbit

Citation (APA):Rasmussen, J. (1974). The human data processor as a system component. Bits and pieces of a model. RisøNational Laboratory. Risø-M, No. 1722

A. E. K. Risø Risø-M~Title and author(s)

The Human Data P r o c e s s o r an a System Component

B i t s a ri d P i e c en o f a Mode 1

11 e r i s Ra s m 11 s s e n

5 1 pages

Date June

Department or group

El ectronics

Group's own registrationnumber(s)

Abstract Copies to

Du ring recent ye a r s t h o i > y i r; t e m s e n p; i .1 e e rdesigning control room s f o r i ri d m >1 r i a 1 proc e t * J ;plants has been faced with complex human factorsproblems. In conventional control rooms the ar-rangement of meters, recorders,and control keyson the control desk, and the choice of properlydesigned knobs and dials are the chief objectivesfor human factors con;-; i dera t Ion • The advent ofcomputer-control led display systems with the var-iety of mod e s f o r d; \ t a c o n d i t i o ni r i /'; a n d p r e s e n -t a t i o n a v a i i a b 1 e to s u p p o r t, t \ i e o p e r a t- o r s c a. 1 1 sfor an int e g r a t e d systerns d e s i g n . The format lingand e n c o d i n g of di .r i p I ca y s s h o i x I (i b e p r o p e r 1 y m a t c h e dto the characteristics of human mental data pro-

cessing, and close interaction ir, needed between

sy s terns eng i ne e ring ex pert i se a nd p sy cho1o gy. For

information and results from another professional

a rea to be i d e n 11 f i e d and d i g e s t e (i a ri i n t e r -

disciplinary working hypothesis or reference

fr a m e i s n e e d e d, a nd t he p r e se n t rep ort de s c r i b e s

such a preliminary mode] aiming at the support of

futures reading, discussions, and experiments.

Using as reference the tasks of process

operators, the report d i sour.ses the role of the

c e p t i v o ge n e r a 1 i z; * t i o n r,, d y n; i m i c m o d e 11 ing a n dm o t o r coord i na t i ve fun c t i o n s of s u b c o n s c i o u s datap r o c e s s ing,a nd of t he dif fere nt t y p e s o f me n t a1modeli; and procedures used in conscious data pro-

cessing.

Revised edition of internal memo, N- JO, June '/'$.

Available on request from the Library of the DanishAtomic Energy Commission (AtomenergikommissionensBibliotek), Risø, DK-4000 Roskilde, DenmarkTelephone: (03) 35 51 01 , ext . 334, t e l e x : 43116

XAVIER DE MAISTRE,

"Ce chapitre n'est absolument que pour les métaphysiciens. 11 va jeter le plusgrand jour sur la nature de l'hommes c'est le prisme avec lequel on pourra analyseret decomposer les facultés de l'homme, en séparant la puissance animale des rayonspurs de 1'intelligence.

Il me seroit impossible d'expliquer comment et pourquoi je me brülai les doigts auxpremiers pas que je fis en commengant mon voyage, sans expliquer, dans le plus granddetail, au lecteur mon Systeme de l'ameet de la bete. - Cette découverte ménta-physique influe d'ailleurs telleraent sur mes idées et sur mes actions, qu'il seroittrés-difficile de comprendre ce livre, si je n'en donnois la clef au commencement.

Je me suis appercu, par diverses observations, que l'homme est compose d'unearne et d'une bete. -Ces deux Stres sont absolument d?stinets, mais tellement emboitésl'un dans l'autre, ou l'un sur l'autre, qu'il faut que l'ame ait une certaine supér-iorité sur la bete, pour etre en etat d'en faire la distinction.

Je tiens d'un vieux» professeur (c'est du plus loin qu'il me souvienne) que Platonappeloit la matiére l'autre. C'est fort bien; mais j'aimerois mieux donner ce nompar excellence å 1-a bete qui est jointe å notre ame. C'est réellement cette substancequi est 1'autre, et qui nous lutine d'une maniére si étrange. On s'appergoit bienen gros que l'homme est double; mais c'est, dit-on, parce qu'il est compose d'uneame et d'un corps; et l'on accuse ce corps de je ne sais combien de choses, bienmal-å-propos assurément, puisqu'il est aussi incapable de sent ir que de penser. C'esta la bete qu'il faut s'en prendre, a eet étre sensible, parfaitement distinct del'ame veritable individuT qui a son existence séparée, ses gouts, ses inclinations,sa volonté, et qui n'est au-dessus des autres animaux, que parce qu'il est mieuxélevé et pourvu d'organes plus parfaits.

Messieurs et mesdames, soyez fiers de votre intelligence tant qu'il vous piaira;mais défiez-vous beaucoup de 1' au tre, sur-tout quand vous etes ensemble.

J'ai fait je ne sais combien d'experiences sur 1'union de ces deux creaturesheterogenes. Par exemple, j'ai reconnu clairement que l'ame peut se faire obéir parla bete, et que, par un fåcheux retour, celle-ci oblige trés-souvent l'ame d'agircontre son gré. Dans les regies, l'une a le pouvoir législatif, et l'autre le pouvoirexécutif; mais ces deux pouvoirs se contrarient souvent. - Le grand art d'un hommede génie est de savoir bien.elever sa bete. afin qu'elle puisse aller seule, tandisque 1'ame,délivrée de cette penible accointance, peut s'élever jusqu'au ciel.

Mais il faut éclaircir ceci par un exemple.Lorsque vous lisez un livre, monsieur, et qu'une idée plus agréable entre tout-å-

-coup dans votre imagination, votre ame s'y attache tout de suite et oublie le livre,tandis que vos yeux suivent machinalement les mots et les lignes; vous achevez lapage sans la comprendre et sans vous souvenir de ce que vous avez lu: - c-ela vientde ce que votre ame, ayant ordonné a sa compagne de lui faire la» lecture, ne l'apoint avertie de la petite absence qu'elle alloit faire; en.sorte que 1 ' autre con-tinuoit la lecture que votre ame n'écoutoit plus."

"S'il est utile et agréable d'avoir une ame dég.xgée de k matiére au point de lafaire voyager toute seule lorsqu'on le juge å propos, cette faculté a aussi ses incon-véniens. C'est å eile, par exemple, que je dois la brulure dont j'ai par 1é dans leschapitres précédens. -Je donne ordinairement a ma bete le so in des apprets de mondéjeuné; c'est eile qui fait griller mon pain et le coupe en tranches. Elle faita merveille le café, et le prend rneme tressouvent sans que mon ame s'en mele, åmoins que celle-ci ne s'amuse å la voir travailler; mais cela est rare et trés--difficile å exécuter; car il est aisé, lorsqu'on fait quelqu'operation mécanique,de penser å toute autre chose; mais il est exti•emement difficile de se regarder agir,pour ainsi dire, -ou, pour m1expliquer suivant mon Systeme, d'employer son ame åexaminer la marche de sa bete, et de la voir travailler sarus y prendre part. -Voilale plus étonnant tour de force métaphysique que l'homme puisse exécuter.

J'avois couché mes pincettes sar la braise pour faire griller mon pain; et quelque--tems apres, tandis que mon ame voyageoit, voila qu'une öouche enflammée roule surle foyer; -ma pauvre bete porta la main aux pincettes, tt je me brülai les doigt&."

VOZAGE AUTOUR DE MÅ CHAMBHJE, chapter 6 & 8, Paris, 1?94,

CONTENTS

1. INTRODUCTION 1

2. A MAN IS QUITE SIMPLE - BUT IN A COMPLEX WAY 1

3. GENERAL OUTLINE OF A SIMPLE MODEL 5

k. THE SUBCONSCIOUS MAIN PROCESSOR 7

k.l Generalization and feature extraction 7

k.Z Generalization of input signals from man-made systems 9

4.3 The internal dynamic model 11

k.k Output system, motor function ~\_k

4.5 Overall system function lk

5. THE CONSCIOUS DATA PROCESSOR l6

5.1 Input information l6

5.2 Processing models and procedures 17

5.3 Memory functions 20

6. COOPERATION BETWEEN CONSCIOUS AND SUBCONSCIOUS PROCESSING 22

7. LIMITATIONS OF THE MENTAL DATA PROCESSING 2.K

8. THE SEQUENCE OF MENTAL ACTIVITIES IN OPERATOR TASKS 26

8.1 Initiation of the operator's activity 28

8.2 Observation and selection of information .. 29

8.3 Identification of system state : 31

8.k Interpretation of the situation and the performance

criteria 33

8.5 Evaluation of the appropriate task and procedure J>k

8.6 Preferred mental sequence in familiar tasks 35

8.7 Shunting effects and mental capacity 37

8.8 The relationship between the frequency of an operator

task and the risk implied 38

8.9 Support of the operator in critical, infrequent plant

conditions 39

9. HUMAN RELIABILITY hZ

10. PREDICTION OF HUMAN RELIABILITY h5

11. CONCLUDING REMARKS k7

12. REFERENCES k8

1. INTRODUCTION

In previous reports the task of the human oper-ator as an information receiver and processor in man-machine systems has been discussed and the followingconclusions have been drawn:

- The human operator in modern technical systemshas to be considered mainly as a versatile data pro-cessor .

- The task of the operator will be to transformthe observed data describing the state of the systeminto a set of manipulations appropriate to his cur-rent goal.

- This data processing will be based upon a trans-formation model of the system and a data handling pro-cedure used to predict the system response, and thusto select and control the appropriate manipulations.

- A system operator normally has available severalsets of models and procedures which can be used for aspecific data handling task. The actual choice of theprocedure depends upon several factors: the work con-dition, his training and experience, as well as hissubjective formulation of the task and his performancecriterion.

- The various models and procedures used by anoperator may differ greatly with respect to severalbasic features, such as the amount of state infor-mation or the complexity of the transformation modeland procedure implied in the data handling, the cogni-tive strain imposed upon the operator, etc.

The present report attempts to present a sim-plified model of a human data processor in interac-tion with a physical system. It is not the aim todiscuss the internal neurological functioning ofman, but rather to create a frame of reference forfurther reading and discussions and for future ex-periments with man-machine interface systems; a ref-erence frame illustrating the different typicalmodes of human data processing as seen from a systemsengineer's point of view.

Rasmussen, 1968

Rasmussen, 1969

Rasmussen and Jensen, 1973

2. A MAN IS QUITE SIMPLE - BUT IN A COMPLEX WAY

A human data processor is an extremely adaptivesystem component and therefore the complexity of hisbehaviour follows directly from the complexity of hisworking conditions. His behaviour therefore has tobe studied in very close relation to the task he issupposed to do and to the features of his environment.H.A. Simon explores the hypothesis: !fA man, viewed asa behaving system, is quite simple. The apparent com-plexity of his behaviour over time is largely a re-

Simon, 1968,' p.2'3

flection of the complexity of the environment in whichhe finds himself". He also states: rf— his behaviourwill reflect characteristics largely of the outer en-vironment (in the light of his goals) and will revealonly a few limiting properties of his inner environ-ment, of the physiological machinery that enables himto think11.

This is a very fruitful hypothesis, but clear evi-dence exist that man has several basically differentinternal modes of data processing, which are put intooperation by different types of outer environments, andwhich have different limiting properties, such as datacapacity, speed,etc. It is therefore important to beable to characterize the different modes of data pro-cessing and to identify the generic features of tasksand environments which will initiate the differentmodes of data processing.

op. cit.

»_—Viewed as a geometric fig-ture, the ant's path is irreg-ular, complex, hard to describe.But its complexity is really acomplexity in the surface ofthe beach, not a complexity inthe ant. "

"—-An ant, viewed as a behav-ingtsystem, is quite simple.The apparent complexity of itsbehavior over time is largelya reflection of the complexityof the environment in which itfinds itself "

"„„-In this chapter, I shouldlike to explore this hypoth-esis, but with the word "man"substituted for "ant".

A man, viewed as a behav-ing system, is quite simple.The apparent complexity of hisbehavior over time is largelya reflection of the complexityof the environment in which hefinds himself.---"

Miller, 1956, p.87:

Such evidence is found, when psychological labora-tory experiments indicate a low data capacity, althoughthe man is able to recognize faces and to drive a car inheavy traffic.

" You may have noticed thatI have been careful to saythat this magical number sevenapplies to one-dimensionaljudgements. Everyday experienceteaches us that we can identifyaccurately any one of severalhundred faces, any one of severalthousand words, any one of sevralthousand objects, etc. The storycertainly would not be completeif we stopped this point. We musthave some understanding of whythe one-dimensional variableswe judge in the laboratory giveresults so far out of line withwhat we do constantly in our be-haviour outside the laboratory.

Accordingly, there is now a trend to distinguishbetween verbal and visual thinking-

Huggins, I97I, p.l60:

"-— Study of the relation be-tween visual imagery and sym-bolic forms such as naturallanguage is currently enjoyinga revival among American psy-chologists who at last seem tohave overcome the classical pre-judices which earlier in thiscentury led them to separateperception from thought. (Theinappropriatenens of this sep-aration is clear to anyone who

has studied L.G. Roberts1 im-portant, but neglected, doctoraldissertation (1963)«The mechanisms which Robertsdescribes provide an elegantmodel of the process of percep-tion. And since logical opera-tions are explicit in this model,it is evident that perceptionlikewise involves thought (albeitat a subconscious level) in muchthe same way as that discussedby Arnheim (1969). Of particularinterest is the major, recentwork of Bugelski (1970), who sup-ports Arnheim's thesis that visualimagery plays a central role inthinking when he writes: "Mycurrent leaning is toward thebelief that few of us, if any,actually think in abstractterms when we are doing oureveryday routine thinking. Irather suspect, with Mowrer (i960)and Bower (1970) that we think withour gut and some rather individ-ual imagery and that our idiosyn-cratic reactions are responsiblefor the considerable lack of com-munication in conversations thatrelate to serious social problems.Words have a way of arousingimages and, if the images arenot the same, communication fails.

Sloman, 1971, p.27O:

ational and intuitive thinking.-* and between » Within. Philosophy there has

long been a conflict betweenthose who, like Immanuel Kant(90, claim that there are somemodes of reasoning, or addingto our knowledge, which use"intuition", "insight", "ap-prehension of relations betweenuniversals", etc., and those whoclaim that the only valid modesof reasoning are those which uselogically valid inference pat-terns—-"

Newell, Shaw gud Si ru o n ,

The point has also been raised by Dreyfus, whocritizises the viewpoint of Simon that human thinkingcan be simulated by a sequential, heuristically guidedsearch of a digital computer.

(The theory )

" rests its claims on otherconsiderations:

1. It shows specificallyand in detail how the processesthat occur in human problem-solving can be compounded outof elementary information proc-esses, and hence how they canbe carried out by mechanisms.

2. It shows that a programincorporating such processes,with appropriate organization,can in fact solve problems.This aspect of problemsolvinghas been thought to be "mys-terious" and unexplained,, be-cause it was not understoodhow sequences of simple proc-esses could a c c o 11 n t f o r t li p»successful solution of com-plex problems» The theorydissolves the mystery by show-ing that nothing more need beadded to the constitution of asuccessful problem-solver.---"

Dreyfus, p-21:

'« We need not appeal to intro-spection to discover what a(chess-) player in fact does be-fore he begins to count out; theprotocol itself indicates it:the subject "zeroed in" on thepromising situation ("I noticethat one of his pieces is notdefended"). Often, of course,locating the promising or threat-ening area involves more thansimply noticing that a Rook isundefended. It may involve no-ticing that "here somethinginteresting seems to be goingon"; "he looks weak over here";"I look weak over there"; etc.

n a s

zeroed in on an area does hebegin to count out, to test,what he can do from there.---"

To conclude: the fact that a man is able to re-cognize faces, drive a car while unaware of his ma-nipulations and "see" generalized features of a dy-namic situation seems to indicate that a model of thehuman data processing should include more functionsthan a sequential processor of low capacity. Some kindof a parallel or holistic processing network of highcapacity and operating subconsciously seems to beneeded in the model.

If one supposes that the human data processingcapability is due to different cooperating subsystemshaving basically different functional characteristics,then it is without meaning to discuss codes, procedures,and information capacity, unless careful reference ismade to the particular subsystems.

mult ivari ab1e, pdata p T o c e s s i K, •;..

3. GENERAL OUTLINE OF A SIMPLE MODEL

As mentioned previously the only aim of the pres-ent discussion is to suggest a frame of reference forour future work related to design of man-machine in-terface systems. In this context man is considered adata processor through which input information receivedfrom the environment is connected to the output, whichis the physical actions upon the environment•

The model will be a very crude sketch. Man is farmore than a mechanistic data processor, but in man-*machine systems the designer has to consider him asa versatile but predictable data processor, and thedesigner has to plan the data processing allocated tothe man and to the instrumentation system as one inte-grated system. The man-machine interface has to bebased upon compatible models of the data processingin man and in instruments, and a model of man drawingheavily upon engineering analogies may be fruitful inthis respect.

The general outline of the model is illustratedon page 6 and comprises two basically different sub-systems. One is a programmable, sequential processorwhich operates consciously to the man. It has a ratherlimited capacity and speed and its main task is totake care of the data handling in unique conditionscalling for improvisations, rational deductions; andsymbolic reasoning. It functions as a higher-level co-ordinator or controller of the second part, the maindata processor.

The main processor has many features of a distri-buted, parallel processing analog computer with highprocessing capacity. It functions mainly^ subconscious-ly to the man, although the conscious processor mayto some extent control and monitor the data processing.The functions are distributed in a highly interconnectednetwork which is able to decode information from thesensing receptors, and to extract higher level featuresor patterns. These control the external actions throughseveral levels of coordinating functions in the motorsystem, which is an integral part of the processor. Animportant implicit function of this distributed networkis a continuously dynamic modelling of the environmentand the body. This internal model is up-dated or trainedslowly, but it operates in real time and serves to con-trol the actions and thus to synchronize the body withthe dynamic environment.

- 6 -

cd

J>3UCD

-P Ö

s o u(D +i O

O

edb

aX

Lo

cal

fe

4

bO

OU

4a

X CDÜ -pCÖ CQ

CD

Outline of the human data processor

7-

escapingfighting

. THE SUBCONSCIOUS MAIN PROCESSOR

It will be reasonable to expect that the basicnature of the information generalization and thederivation of control information to the motor sys-tem and thus the coding of the subconscious dynamicmodel, are closely related to the basic needs manhas met during his evolution. During the evolutionof man survival has been granted most probably tothose individuals who were best at identifying quick-ly by direct pattern recognition the appropriate goalimplied in a situation, such as approach, escape, eat,don't eat, etc., and who had the most efficient con-trol of fast critical sequences during hunting, fight-ing, escape, etc.

This dynamic interaction with the environmentcalls for a very efficient feature extraction andclassification and dynamic coordination of the motorsystem with the environment. The receptors and themuscels of the body serve as input and output systemsof a complex multivariable control system having ahigh data processing capacity. The subconscious mainprocessor may therefore be assumed to have evolvedinto a data processing system with particularly highcapacity in functions needed in the control of thebody, i.e. when operating in a spatial, temporaldomain. Johnson refers to the work of Heidbreder,which should be studied in this context.

This function may quite probably be'organizedsimilarly in man and in higher-order animals.

k.l Generalization and feature extraction

In engineering terms, the main processor maybe viewed as a complex, distributed analog comput-ing network although the signals of the nerve tis-sue are coded as pulse trains. It is simulating fea-tures of the control task rather than performingsymbolic calculations. The high data processing capa-city needed in its real time tasks is due to an exten-sive parallel processing of data in the network. Thefunction of this processing system is based upon ageneralization of the input information representedby the signals from a large number of receptor cells.Chis generalization takes place through, several, levels)f processing, informationally speaking, as well as;een from the organization of the processing network.b.en the determining, higher-level features of the'ontrol task are formed in this way, another multi-

Johnson., 1950, p.30.5:

"---To account for these re.-Heidbreder has put forth thehypothesis that the porcepti.of concrete objects is the chnant mode of c o gn i t i v e r e a c trelated to the phylogeneti.c ,o r i t y of 1o c omc t i v o and ma n:;tive capacities of the orgai:*.a n d t h a t c o n c - v ' \ • * l > ••••-. i r; n ••

therefore 5 easier' Lh'% more isembles the perception of acrete objects«,-—"

level processingtrol signals forsystem. See e.g.

decomposes these features into con-the individual parts of the motorWeiss.

In the task of controlling the body in the dy-namic environment, the most important generalizationsto be made from the input information will be to iso-late and identify spatial entities and their temporalcharacteristics. This means to identify objects whichare moving or which are able to move and for whichgeneric dynamic models are present in the internaldynamic modelling system. Furthermore, the generali-zing function has to extract those generic featuresfrom the background - the scenery in which the dyna-mic actions take place - which are determining thebehaviour of the movable objects.

UU UU UU

sensory data

intrasense integration

intersense integration

visceral activity

6 intramernb

7 muscle, co

oordination

ction

Diagra of levels

Research on visual perception has identifiedfixed programmed networks in several animals whichgeneralize the visual information into this kind ofdynamically determining features. In the retina ofa frog's eye at least four types of preprogrammeddetectors has been found: edge, bug, dimming, andevent detectors. Similar detectors identifying formand movements of objects against the background arealso found in higher order vertebrates, although thefeature extraction does not in all cases take placein the retina, but in preprogrammed parts of thenerve tissue at a higher functional level (visualcortex). A review of this research is given by Michael.

Lettvin, 1959

Michael, 1969

Hubel, 1963

Although the programming of the nerve tissue ofman may be more flexible, there is reason to believethat similar features have developed in the subcon-scious information handling of man during his longevolution. This is in good agreement with the state-ments of Lashley.

Lashley, 19̂ +2:

" there is some rcdso,believing that general!is one of the primitivefunctions of organized :tissue ". "In spite oenormous differences intural arrangement of thsystems of different anclasses, the fun c 'c; i. o 11,?. 1ty seems essentially th--- in the bird, the roand the man . — ••--- u

n 1za t, b

01"

i oö, S

nervöf t

»t:ii V

in; ane

e sd e n

h rr i.i

i >51

t; .:

amt

(from 1969 ed. p* 236)

Neurological research thus supports the prop-osition that the neural system has inherent proper-ties which result in a particularly high data pro-cessing capacity for generalization in the spatial,temporal domain, needed to control the body in a dyna-mic environment.

-escaping too

- 9 -

k.2 Generalization of input signals from man-madesystems

The hypothesis is then that the subconscious mainprocessor has a particularly high data capacity whenit is allowed to operate in a temporal spatial domainby extracting dynamically meaningful features and pat-terns directly from the receptor signals.If this istrue, it leads directly to implications for the de-sign of man-made systems.

Such systems fall into different typical groups,seen as part of man's environment.

In many instances man-made systems act as dynam-rical objects in the environment like some kind of ar-tificial animal, such as other peoples cars, vendingmachines, etc. In other cases they act as extensionsof man's motor system. Examples are tools-* also toolslike bulldozers or cars you are driving yourself. Inboth cases man's interaction with the systems can bebased upon the generalizing functions and the internaldynamic model of the subconscious processor. To thesesubconscious generalizing and modelling functions therewill be no great difference between the input informa-tion related to the body itself and to the environment,i.e. whether a tool is considered an extension of thebody or part of the environment may not be important.The important thing will be that it is possible forthe main processor to define a control surface in theenvironment; a surface from which the informationneeded for the task can be perceived and processed inthe temporal, spatial domain, i.e. the data should beavailable in a code fitting the main processor.

Another typical class is the systems in whicha communication system is inserted between the actualphysical process and the man, a system converting theinformation into symbolic representations of the stateof the process. This is typically the case in industrialprocess plants controlled from a control console presen-ting to the operator the data'collected by the measur-ing channels. Whether the main processor selects thesurface of the controle console as the control surfa'ceor accepts the measuring channels as an extension ofthe sensing system and thus places the control sur-face close to the internal physical functioning ofthe system, probably depends upon the compatibilityof the display coding and the code of the main pro-cessor.

This leads to the following considerations forthe design of man-machine interface systems:

To activate the subconscious main processor, in-formation should be presented in parallel by meansallowing feature extraction or pattern formation in a

- 10 -

time, space domain. The visual representation of theinformation then can be used directly by the proces-sor; it will not have to operate upon the symbolicvalue of the presented data.Sequential presentationof digital or alpha-numeric information will not be ac-cepted by the subconscious main processor, but willinstead initiate conscious reasoning because directvisual generalization is not possible.

When information is presented by analog devicessuch as meters aid recorders, the visual representationsform by themselves a spatial, temporal world, whichmay be accepted by the subconscious processor; theoperator may work "from the expression on the face ofthe system11. The control surface will be the face ofthe operator's console, and familiarffface expressions"or data patterns may initiate trained routines, whichare repetitions of actions earlier found successful.The closer the temporal, spatial world formed by thevisual representations of the information is relatedto the internal physical function of the system, themore capable the man will be to improvise subcon-sciously if new, unique situations appear. Everycar driver has met unique and critical situations intraffic which have only been managed because of theimprovisations of the subconscious main processor(- whew S) •

In many industrial systems man has to monitorand control a process which is a complex system ofenergy and mass flows. The balance of this systemmay be disturbed owing to faults, and the man then hasto improvise, especially if he has to take over froma failed automatic control. Will the operator be ableto improvise in the same efficient way as he does inbusy traffic if the visual representation of theinformation displayed to him is closely related tothe spatial, temporal properties of the-primary pro-cess? If the information is displayed in pictures show-ing how something, i.e. some kind of substance, massor material, or symbols substantiating energy or in-formation, moves through a physical system and pilesup if a balance is disturbed?

If so, abstract information such as temperatures,pressures, etc., should be transformed into physical,visual properties of the time space world formed bythe visual representation, rather than displayed di-rectly. This may for instance lead to display of a tem-perature as a visual indication of the level resultingfrom an integration - a pile up - due to a flow of en-ergy, which can also be visualized.

Primitive use of communication theory generallyeads to a statement of the need to reduce the amount

of information presented to the man to avoid satu-ration of his data capacity. Normally this is due to

Which of the two unusualfeatures in the drawing onpage 9 did you notice first

Guilford 1967, p. 197

This has been usedby Chernoff (1971)of mineral sampletion:

litterallyfor control

classifica-

I I I M I

THE USE OF FACES TO REPRESENT POINTS IN n-DIMENSIONAL

-SPACE GRAPHICALLY

- 11 -

use of the simple analogy of a transmission channel- depicting something like the transatlantic tele-graph cable. In our context, however, the principalproblem is not the information transmission, but thecapacity and reliability of information processing.

To facilitate the high processing capacity ofthe subconscious generalization function it is im-portant to display all the information possiblyavailable, but as an integrated, meaningful environ-ment in which to operate. This is important to givethe man the best opportunity to draw upon direct vis-ual generalization and to form his own generic fea-tures. Redundant information should not be omitted.Redundancy depends upon the reference frame applied.Information which, is redundant from a rational, de-ductive point of view may be extremely important ingeneral search procedures or in feature and patterngeneration. It does not sound reasonable to expecta car driver to operate more efficiently when pre-sented with only an abstract indication of the vi-tal physical parameters via a control console^reading a book will not be eased by removing allredundant letters and words in order to avoid over-load of the data input capacity.

In other words: the process operator livesin a symbolic world. The information presented to himis a code for the physical, dynamical process in theInterior of the plant. He is able to - and by thelesigner normally expected only to - operate uponühe physical meaning of the symbols by rational de-luctive reasoning. During frequent routine tasks,/lowever, he may be operating the control desk - notehe process - by his subconscious routines. He may;e able to improvise rapidly and subconsciously - asa car driver is able to do - if he is allowed tobreak down subconsciously the information-patterns>f new situations into familiar generic units. Thisis only possible if he can control the process di-•ectly - the display system therefore should be^transparent" and the physical process should betirectly "touchable11 on the control desk.

-1-.3 The internal dynamic model

The very efficient behaviour of man - and ani-mals - as multivariable control systems in dynamicsequences which are too fast to allow perceptualfeed-back correction is possible only if the sub-.onscious main processor includes an internal dy-namic model of the body and its environment. It isimportant, that the model includes the body itself,••trice only then can it operate as an efficient ref~r"ence frame for fast, predictive, open loop control; :" the body. The internal model is a dynamic model1 the engineering sense - a kind of functional ana-

'..• <-'gy to the external spatial, temporal world - cap-

Channel in communication theory.

I 1 -•*- Pro- fiction

Channel in Man-Machine systems.

-quick recoveryon slippery ground.

- 12 -

able of simulating the dynamic processes in a waysimilar to analog computer simulations.

In the present discussion of human data pro-cessing we distinguish between a number of discretefunctions with different data handling propertiesseen from an engineering point of View. This doesnot imply a similar separation of the neitye ̂ systemin corresponding biological parts performing" the *.functions.On the contrary the holographic descrip-tion of the functioning of the nerve tissue as inter-ference patterns of exitation waves (Lashley, Pribretm)leads to a distributed model in which generalization,^simulation and coordination may be various aspectsof the function of a distributed and highly inter-connected network o

The internal dynamic model stores spatial, per-ceptual patterns and trained patterns of Coordinatedmuscle movements and their interrelation throughtime. The state in time of the model is continuouslysynchronized by the information from the patternsperceived« The formation of generic features of themovable objects and the background scenery is dueto a slower updating of the model; a kind of averagingfunction adjusts the structure and parameters of theanalog model to the information perceived during thetraining and adaptation period«

The closely integrated nature of the entiresystem is also demonstrated by the way the internalmodel and the generic features build up. The modeland the generalization are not the result of passiveprocessing of the perceived information, but a re-sult of active, physical interplay with the en-vironment o This statement is. supported by exper-iments9 see for instance the experiments of Heldand Hein.

These aspects are not only of academic interestbut may have important practical implications. Themovements of the head needed to read instruments ina control room may take part in the identificationof the source of the information. The physically welldefined data, such as distance or speed which areused as discrete data for inputs in a mathematicaltreatment of a dynamic process, are not defined orused by the subconscious data processing as separateparameters, but implicitly as part of a pattern hav-ing a functional meaning. For instance distance isnot judged or measured by a two eye stereoscopicvision, but by the operational consequence experi-enced by walking or reaching along it.

Together with the need for data presented invets facilitating generalization, it may therefore,a important that abstract data are converted into

Pribram, .1966

1969 ed. p. 2351

M—._ The properties of the holo-gram are just those demanded byus to account for ordinary per-ception. I have already made thesuggestion that arrival patternsin the brain constitute wavefronts which by virtue of inter-ference effects can serve as in-stantaneous analogue cross corre-lators to produce a variety ofmoiré-type figures.---"

Lashley 19zf2 :

--- Details of the theory areof little immediate importance.But the principles of the estab-lishment of interference "patternsat successive levels in the ner-vous system, of the modificationof these patterns by superimposedpatterns from earlier stages inthe series of levels, from retinato motor cells, and the redupli-cation of memory traces as a con-sequence of the properties of theinterference pattern are, I be-lieve, reasonable conclusionsfrom the organization of behaviorand the structure of the nervoussystem. The visual system isprimarily concerned with spatialorientation and for it the tran-sition from a sensory to motorpattern can be most adequatelyconceived as an interplay ofpolarized systems or of inter-weaving dynamic patterns inwhich the spatial propertiesof the visual stimulus are trans-lated by intergration at a seriesof levels into modifications ofthe general pattern of posturalorganization.- -- "

(from 1969 ed. p.250)

Held and Hein, 1963

Barlow et al., 1972

'»„,„_ The organization of thevisual system hints that eachrelational feature is firstdetected by many separate unitson a local level, and is thengeneralized by combining theoutputs of these separate units.This would require very extensivereduplication of the "relationdetectors", but eye or headmovements might lessen this re-quirement. The extraordi nar11y '

sdisruptive consequences of dis-sociating or disturbing the re-lation between selfinitiatedmovements and visual experiencesuggests that spatial relationsin the visual field may be en-coded in terras of "motor schemata"the movement associated withproducing or removing a spatialrelation.---"

information man will be able to perceive directlyfrom its dynamic functional implications. He willthen be able to store them in his internal modelclosely related to the coordinating patterns of hismotor system. Practically speaking this leads todifferent preconditioning or encoding of the datapresented to a man if he is expected to control theprocess itself, compared to the case when control ofthe display surface is aimed at.

The function of the internal dynamic model isnot only clearly demonstrated in fast sequences ofopen-loop control of the body. Also in slow sequencesit supplies information to the processor and decreasesthe amount of information required from the environ-ment. This is indicated clearly by instances foundin our control room studies as well as by common,everyday experience.

Warning signals in process plant control roomsare generally meant by the designer to be signals toalert the operator, who is then supposed to consultthe instruments to identify the cause. In case ofrather frequent warning signals which appear as anormal part of a work sequence, during plant startup for instance, the internal model of the operatorapparently gives him a "process feeling" whichcontinuously keeps him "set" for the proper action.He often only needs to be triggered to take action.He "knows" from the general behaviour of the systemand from his prior actions what has to be done with-out consulting the instruments. He may even - withoutrealizing it - correctly respond differently to theoame warning signal in different operational phases- for instance during start-up.

The importance of the internal model for thembconscious data processing is clearly demonstrated•vhen the internal model loses synchronism with thereal environment. This sometimes happens in dailylife when a very familiar routine sequence of ac-tions is disturbed, for instance if one occasion-ally has to enter the sequence not at the usual open-ing but at a later link downstream. This may be thecase if a very familiar city area is occasionallyentered along an unfamiliar route, resulting innistakes at familiar street corners until aconscious reorientation has taken place. Anotherexperience: Have you ever had a door in your housemoved to another position in the wall?

The output from the internal model controlsrhe physical activity of the body. The result of; tiis control is currently checked and the state of;ie model adjusted by the perceptual feed-back sys-

tem. If the internal model loses synchronism; i.e.i the environment does not proceed "as expected"

See also Bainbridge^ 1969

»__— In this way he apparentlymaintains a "mental picture" ofthe present and future status ofthe furnaces. There is evidencethat he often makes use of hisstored data rather than readingvalues from the display paneleach time they are needed. "

and Bainbridge, Beishon et.al. 1968

"-— This arises because peopleappear to work from an internal"model" of the system they arecontrolling. Very little isknown about the form of thisinternal model, but studies onthe subjects' running memoryfor variables in the systemshow that they keep mentaltrack of important variableseven through these may be dis-played in front of them. "

Consider also what Pribham (1969)calls "The Bowery-el Phenomenon":

"---For many years there was anelevated railway line (the "el")on Third Avenue in New York thatmade a fearful racket; when it wastorn down, people who had beenliving in apartments along theline awakened periodically out ofa sound sleep to call the policeabout some strange occurrence theycould not properly define. Manysuch calls came at the times thetrains had formerly rumbled past.The strange occurrences were ofcourse the deafening silence thathad replaced the expected noise.--"

- Ik -

or the model is incapable of simulating an unfarm-iliar environment, an interrupt signal will alertthe conscious data processor.

In a man-machine system it is important to con-sider the role of the internal model which gives the"process feeling11 and to display the informationneeded for the subconscious synchronization and up-dating of the model. It may not be feasible to haveinformation displayed only at the operator's order,some information may be needed in continuous dis-play.

It may further be important to consider thatthe state of the internal, model is not only con-trolling the actions based upon the subconsciousdata processing, but may also control the intuitiveguesses and hypotheses which underlie the consciousdataprocessing.

k.k Output system, motor functions

The output from the subconscious data proces-sing system is always a motor function. The motorsystem consists of a great number of muscles whichare controlled and coordinated by a complex, highlyinterconnected nerve system. It is generally acceptedthat this coordination and control is hierarchicallyorganized and experiments show that the modellingfunction of the subconscious data processing is dis-tributed also in the different levels of the motorsystem. This makes the motor system of the body anactive part of the human data processor. A few im-plications of this statement have been touched,see page 12.

The role of the perceptual part of the pro-cessing system is to generalize, to compose fromehe signals coming from the individual receptive>;ells the generic features characterizing the dy-namic Interaction with the environment. The coordi-nating function of the motor system is to decomposethrough several levels the appropriate complex re-sponse of the body into detailed control sequencestfhich are based upon information from the higher-level data processing as well as upon local motorfeed -back signals.

li. 3 Q v era.11 system functions

When discussing data handling functions andprocedures one normally has to define determiningjcirameters such as task, goals, performance criteria<-tc« These are not explicit parameters of the sub-conscious data processing, but are found implicitly.;. the features defining the state of the environ-

ment or In the trained patterns of the internal dy-namic model•

Weiss, 1952:

visual acoustic propriocepti

1 sensory data 5 intermember coordination

2 intrasense integration 6 intramember coordination

3 intersense integration 7 muscle contraction

4 visceral activity

Diagram of levels in the hierarchy of coordinating functions

- 15

The conscious decision to go home sets the in-itial conditions of your internal model and nofurther decisions are needed explicitly to leave theroom, take your coat, or start the car« Very oftenyou don't even have to decide to go home - you areshort of cigarettes, you get hungry, you are ffsetff

by your wife in the morning to go home at fouro'clock, etc. One action creates implicitly the scenefor the next.

However, the conscious data processor may over-rule the function of the subconscious processor atseveral different levels to keep track of a planfor a special sequence of actions, to evaluate thestate and decide the subsequent plan, etc. The func-tion of the conscious data processor is initiatedby interrupt signals when the internal model losessynchronism with the environment; when it provesinadequate to control the interaction with the en-vironment; or when the model reaches- predeterminedstates for which interrupt demands have beenassigned.

- 16 -

5. THE CONSCIOUS DATA PROCESSOR

The main data processor operates subconscious-ly upon information derived by generalization andfeature formation from the data supplied by the re-ceptive organs. The information treated and the modelused for the data processing have a direct relationto the external time, space world in the same way asan analog computer model has a direct relation instructure and variables to the process simulated.

The conscious data processor is capable of oper-*ating in several basically different modes. It cansimulate the external world in the domain of thesensed information - e.g. make "visual experiments"in which a visual imagination of a process is usedto "foresee" the response of the- process to plannedmanipulations. Or it can process symbolic data byfollowing a prescribed plan, a sequential procedure- e.g. to do numerical calculations and abstractlogical reasoning.

Whereas the subconscious processor from anengineering point of view has great similaritiesto a distributed analog computer, the consciousdata processor has many characteristics of the pro-grammed, sequential digital computer. The data pro-cessing is based upon a model or representation ofthe physical system underlying the problem to besolved, and upon a plan or procedure to make themodel operate upon the input data.

3.1 Input information

In the physical world the data which arechosen as observations by the conscious data pro-cessor have no meaning individually; they 'only gettheir meaning through the constraints imposed upontheir interrelationship by the physical processesof the environment.

The generalizing functions of the subconsciousdata processor derive these functional constraintsfrom large sets of input data. The conscious dataprocessor may scan the input data provided by thereceptive system and then select and process physi-cal variables directly by abstract reasoning. Howeverin most cases input information is selected from arather high generalization level in the subconsciousprocessor - you "see" directly that a man is angry^ij;<t at all the detailed physical changes of his face,or you "see" directly that the water in your tea-kettle is boiling. You hear a tune, not a sequence<f individual notes, or you see a bottle ready tof irn over, not the data describing the interactionbetween gravity and support. From an engineering

point of view this type of dataconditioning fits theholographic capabilities of a highly interconnectednerve tissue.

However, in some cases the conscious data pro-cessor cannot take the situational meaning of a setof physical variables as input information, but ithas to operate directly upon individually selectedand observed magnitudes of such variables, for in-stance numerical data obtained from instruments orwritten messages.

$.2 Processing models and procedures

To be able to interact successfully with aphysical system, the state of the system and itsresponse to the actions intended have to be iden-tified from the observations or data received fromthe system. Generally the observations have nomeaning individually, only the interrelationshipsin a set of data will define the state of the sys-tem and allow a prediction of future behaviour.This interrelationship is due to the constraintsimposed upon the set by the internal anatomy andprocesses of the system.

The human data processing has to be based upona representation of such system-given constraints.This representation is a data processing or trans-formation model of the system and can be present asan internal representation, i.e. a mental model; orexternal models, i.e. drawings, diagrams, etc., canbe used to support the data processing. The mentalmodel is not necessarily a visualization of thephysical system, but can be a more abstract dataprocessing model in the engineering sense. Suchmodels have been discussed from the psychologicalpoint of view by Craik.

Furthermore, a procedure is needed to use themodel for data processing. The conscious data pro-cessor has several different models and proceduresat its disposal and the choice in a specific taskdepends upon different characteristics of the tasks.

Some examples:

The problem is to judge the movement of amechanical lever when another lever is operated byhand. The levers are connected internally in thesystem by a mechanism of ropes and pulleys. Thisproblem may be solved by an internal visualizationcf the system or by a simple sketch on paper. Somemay be "experienced enough" to grasp the solutionas a whole, others may have to-proceed through aeries of intermediate steps - if I move this leverthis way, then and so forth. The procedure thenwill be a stepwise tracing through the system of the

Forundring er begyndelsen til tænkning.

"Mental Model"

Bo Bojesen in "Politiken"

Craik , 19^3

Rasmussen and Jensen , 1973

C——J 2w. I

- 18 -

flow of movements, a stepwise visualization of theeffects of a simulation. This problem has been dis-cussed in a more philosophical context by AaronSloman. The model as well as the procedure usedhas a close relation to the actual physical problem.

This sort of direct visual experiment is possibleif the physical system and the relevant variablesare directly visible; if not, some kind of visualanalogy can be used. For instance this is done whena primary school teacher uses ffwater-pipe-and-jar -explanations" to explain simple electrical circuitsand when mechanical mass-and-spring systems are used,to visualize electrical resonance phenomena.

In this type of data handling the model supply-ing the system constraints between the individualobservations is either the physical system itself orsome kind of physical or visual analogy, and the pro-cedure to obtain the desired resulting data is a visualexperiment; i.e. you visualize the process itself.

If the process or the system is too complicatedother types of models are generally used. A model ofthe constraints often used in a design or researchtask is the abstract or symbolic model constitutedby mathematical equations. A system of equationsmodels the constraints between the individual vari-ables or observations. The characteristic feature ofsuch a model is that it is generalized. The data-handling problem has been transferred to a symbolicworld where neither the model itself nor the proce-dure by which it is used now has a clear relation tothe physical world.

Several procedures are available and used toprocess data by equations. One is a sequence of logi-cal, symbolic decisions and manipulations%based upona set of mathematical theorems. The written equationsthen is a visual support for the symbolic mentaloperation on numbers, sets and relations. But thewritten equations live their own lives, and theprocedures used for treating familiar types of equa-tions look more like direct visual manipulation ofthe physical signs on the paper. You know how theybehave when you move them around. Signs change whensymbols are transferred from one side of the equa-tion to the other; dx^/dx changes almost by itself'o 2x and so forth. The display aspect of solvingalgebraic equations has been discussed by MymonGoldstein.

The power of the mathematical data processingi3 due to the generality of the model and existenceof a comprehensive system of established procedures,

are only related to the formal model itself,

Sloman, 1971

Goldstein, 1969'.

(3x-7)+2x = 6+(x+19)

3x-7+2x = 6+X+19

5x-7 = 25+x

5x = 32+x

kx = 32

x = 8

- 19 -

not to the physical problem. Other mental models re-lated to the physical world are of course needed totransform the problem to the formal domain. However,they do not represent the total system, but only thebehaviour of the parts at the level where the indi-vidual equations are formed.

Complex data processing by mathematical modelsand procedures are typically used by system plannersand designers rather than system users and operators.If the data processing implied in system operation ina certain situation call for a complex data handlingprocedure, e.g. solving of mathematical equations -this has very reasonably been foreseen by the system'designer, and he has by means of experiments in thelaboratory, by simulation using some sort of analogy- physical or mental - or by use of mathematicalequations, performed the data processing necessaryto decide upon the appropriate sequence of manipula-tions or actions. From the results of this data pro-cessing he can then extract a set of operating in-structions or operational procedures to be used bythe system operator.

The data handling procedures of the operatorare now simple - to execute a stored sequence of pre-determined actions. The system model supporting thedata processing now has to supply the operator withstate-defining data sets, to initiate the stored pro-gram, and to provide a description correlating theprescribed actions with the manipulation surface ofthe system, i.e. some sort of topographic map. Thefunctional model of the interior of the system isnow found implicitly in the stored program.

Whatever the nature of the transformationmodel has been in initial system operation, experi-enced system operators develop work procedures orrules by storing sets of state-defining observationsand goal-directed procedures in a kind of decisionmatrix, as discussed by Beishon. Even when operatinginstructions have been issued by the system designer,system operators often tend to form their work pro-cedures directly by an initial cut-and-try operation•~ i.e. to derive the transformation model neededdirectly from the system behaviour, - rather than•spend the effort reading the manual.

The aim here has not been a thorough discussioncif different typical conscious data processing methodsbut rather to illustrate that mental data processing::i s s e v e r a 1 basic elements:

- The observed information describing the statet the system to be operated, which can be used at

several levels of generalization as input to the

Beishon 1966 , p. 2^2:

" These are information-decis-ion-action procedures which utilizethe facts and results of the per-ceptual observations and directthe sequence of activities of theoperator in time. "

" This third part of the skillis the "central" program whichdirects activity and in a senseis the "skill" itself. This part,together with the first category,can be envisaged as a "rule book"which contains the necessary factsin "look up tables" and the pro-cedures in a hierarchically struc-tured network. "

PIP'BEFORE OPERATING'.' RFAD YOI/RWSTPUCTION MANUAL.WC

Instrument manufacturer's prayer

("Tektronix" tag)

- 20

- The system model used for the processing andsupplying the system constraints upon the interre-lationship of the observations. This model can bethe system itself, a visual model or analog of thesystem or a purely general or abstract model. Fur-thermore it may be found implicitly in a set ofinstructions or rules.

- The procedure or plan to let the model oper-ate upon the input. The procedures can be exper-iments, visual or real, with the system or an analog«it can be a set of general rules closely connectedto the representation used, or it can be a set ofprocedures closely connected to the specific system .and its actual operational state.

Furthermore, it is important to realize thatan operator may have several basically differentsets of procedures and models available for a task,and that his choice depends upon his subjectiveformulation of the task and of the performancecriterion. This has been discussed in detail fora specific task by Rasmussen and Jensen.

5*3 Memory functions

Owing to its distributed nature the subconsciousdata processor can perform several subtasks simul-taneously. "Consciousness" however, has to be switchedbetween the different subtasks, and the consciousprocessor therefore is a sequential processor, whichnormally runs different subtasks on a time sharingbasis. Every task implies the subtasks of selectinginput data or observations, of generating or memoriz-ing the system model needed, of operating the modelby a procedure which may not be immediately avail-able, but may require formulation or memorizing.Furthermore, quite different tasks may run simulta-neously, and the decisions resulting from the dataprocessing may need conscious coordination and con-trol of the motor system.

To perform the processing' a memory is needed,and it is generally accepted that two different mem-ory mechanisms are implied. A short-term buffer mem-ory with low capacity for storage of input data, of•intermediate decisions and results and of modelsind procedures, and a long-term memory with a veryhigh capacity.

The capacity of the memory system is an import-ant limitation of the processing capability, and theresults published from psychological laboratory ex-eriments have to be studied in our context to relateI'ae capacity to the different 'means of encoding andformatting displays which are now available to sys-tem engineers. The influence upon memory capacityi.vom the context in which the information is stored

Bruner et al. 1956

Rasmussen and Jensen, 1973

has long been known - see for instance Fuller's (1898)description of the mnemotechnics of Simonides (500B.G») mentioned by Miller. This indicates the influenceof the mental data processing model used by the con-scious processor upon the capacity of the memory.

The effect of recoding has also been discussedby Miller. In his paper Miller exemplifies his viewmentioning perception of telegraph coded signals. How-ever, in the present reference frame the learning ofthe Morse code by a highly trained operator may bedue to training of his subconscious modelling system.This raises the question of the cooperation or inter-action between the internal dynamic model of the sub-conscious processor and the long-term memory of theconscious processor. Recall of a poem learned longago may only be possible if it is spoken out loud inthe proper rythm - is it stored in the subconsciousdynamic model which also includes trained patternsin the motor system of speech?

Miller, 1968 :

" People like to locateinformation spatially, andthe fact tells us somethingimportant about the way menand information interact. "

mnemotechnics, , isconcerned with devices andprocedures to improve ourmemory . There is one basicpsychological principle under-lying all these schemes, butit can be exploited in manydifferent ways. The prin-ciple is that concrete visualimages are easy to remember,especially if they are re-markable or even a littlebizarre. M

Miller, 1956 :

" Since the memory span is afixed number of chunks, we canincrease the number of bits ofinformation that it containssimply by building larger andlarger chunks, each chunkcontaining more informationthan before "

" A man just beginning tolearn radiotelegraphic codehears each dit and dah as aseparate chunk. Soon he isable to organize these soundsinto letters and then he candeal with the letters as chunks.Then the letters organize them-selves as words, which are stilllarger chunks, and he begins tohear whole phrases. "

22

6. COOPERATION BETWEEN CONSCIOUS ANDSUBCONSCIOUS PROCESSING

The task allocation between the conscious andsubconscious processor does not take place on an allor nothing basis, but depends upon the nature of thetask. The conscious processor is only taking care ofthat part of the processing for which the correspond-ing internal, subconscious dynamic model of the situ-ation has not yet had an opportunity to evolve.

A person starting to learn to drive a car has tomake very detailed observations, to formulate the de-tailed subtasks, to identify the prescribed procedureand to manipulate consciously: Now! I'm accelerating- in 2nd gear - have to change at 30 mph - whatfs thespeed? - Now - ok - release the throttle, that's theright foot - then disengage the clutch -- etc. -- etc.--- now the throttle, whew! ok - where am I now?Later you hear directly that itfs time to shift gears,you take the decision, and the rest happens automat-ically. Finally, you just get an interrupt - oh, Ihave to go to that meeting - and your consciousprocessor may continue to plan the meeting except forstrategically important decisions called for by inter-rupts during driving - e.g. to damn the idiots whosecars did not behave as your internal model predicted.

The conscious processor is the high-level con-troller, taking care of the unique, special tasks,controlling the body as long as it takes to have thesubconscious system adapt to the situation. Duringthe adaptation the internal model is updated, the inputsystem forms efficient features and the motor systemappropriate coordinating patterns. The input to andoutput from the conscious processor move to higherand higher data and coordination levels typical forthe situation.

An important aspect is that the input data usedby the subconscious processor may not at all be thesame as those used consciously, and you may not beable to tell what information you use in a routinetask. The conscious processing may only be used dur-ing training, and when high routine skill has beenattained the initial conscious procedure may degene-rate completely.

In some cases interference may obstruct thetask if one tries to take over by conscious control.An amusing experience - known at least to Danes ofsome age - Is encountered with the coded locks usedon bicycles. If you occasionally try to remember thecode after a long period of trouble-free operation,you may then find it completely impossible to unlockthe bicycle. After several unsuccessful attempts the

only way to succeed may be to turn your back to thelock and to think of something else - and then sud-denly to see your chance to operate the lock beforethe conscious control has time to interfere again.

This possibility of interference between theconscious and the subconscious processing has to bestudied in relation to the reliability of human oper-ators. Is there, for instance, a critical time inter-val during the learning phase, when the transitionfrom conscious to subconscious processing takes place,and is "mental black-out" during operation due tosuch interference?

Several of the aspects of modelling of humanbehaviour discussed here have been recognized andtreated by Miller et al., who also relate the problemto generally established psychological concepts.From a systems engineer's point of view their ap-proach leads to several difficulties. Their descrip-tion of human behaviour is based upon a hierarchy of"Plans", having the "TOTE-unit" (Test-Operate-Test--Exit) as a modular unit. This focuses the interestupon the sequential aspects of behaviour and incor-porates the "internal representation" as an implicitproperty of the "Plan". This may be an appropriatescheme for a description of a particular behaviour,but it seems as inadequate for predicting behaviouras a sequential description of the behaviour of amultivariable continuous control system would beto a control engineer. A model should rather dealexplicitly with the internal representation as wellas the sequential aspects of "procedures" or "Plans".In some cases it will be the "Plans" which are givenimplicitly, as when the internal dynamic model pro-cesses the input under the constraints of the actualinitial conditions and control parameters ("inten-tions", "goals"). Furthermore, the description tendsto be a "one channel" model covering all types ofhuman behaviour. Although the similarities of con-scious behaviour and automated skills to digital re-spectively analog computers are mentioned, the skillsare related to an analog output system connectedto a sequential processor rather than to a separatesystem operating parallel to the sequential pro-cessor.

Miller et al. i960 , p. 90,91:

" Once the subplan is masteredand turned over to his muscles,however, it can operate as ifit were a subprogram in an ana-logue computer. "

" That is to say, planningat the higher levels looks likethe sort of information-proces-sing we see in digital computers,whereas the execution of thePlan at the lowest levels lookslike the sort of process we seein analogue computers. The devel-opment of a skill has an effectsimilar to providing a digital-to-analogue converter on theoutput of a digital computingmachine . •"

7. LIMITATIONS OF THE MENTAL DATAPROCESSING

The mental data processing of man is very flex-ible and adaptive and the internal organisation isof interest especially when task demands violate theinternal constraints on the functions. In the planningof complex man-machine systems the limitations ofman's data processing capacity have to be given dueconsideration. It is, however, not very sensible todiscuss the limits of this capacity without referringcarefully to the mode of operation under considera-tion.

There Is a current trend in the discussion ofman-machine communication and of man's data capacityto adopt the concepts and measures developed byShannon in his theory on the information capacity ofnoisy communication channels. The adoption and useof concepts like information, entropy, and redundancyoften seems to be very crude, and the results havenot been convincing. There seems to be several funda-mental problems underlying the current use of infor-mation theory, problems which have to be studied care-fully before the final conclusion can be made:

- The information theory is developed to handleproblems related to the transmission of informationby a sequence of symbols through a channel of limitedcapacity* In our case of man-machine communication,the transmission of information will only in specialcases be a sequence of symbols, and the limitationsin man's data capacity are probably due to limitationsin his data processing capacity rather than limita-tions in the transmission capacity.

- The concepts of the information theory, suchas information, redundancy an entropy, are closelyrelated to the probablity of the relevant symbolsand sequence of symbols. In human communication, thebasic concepts should rather be related to the struc-ture of the information source and to the structureof the data processing models and procedures usedby the man.

- With reference to Miller's classical paperon the "magical number seven", it is often stated,that man has a low data capacity. In his paper Milleris aware that this is not in accordance with everyday experience, as already mentioned.

This controversy is important in our context.It is a general viewpoint that the amount of datain modern complex plants is so great that a data re-duction should be attempted so as not to overload theInput capacity of the operators. This may not beappropriate under all circumstances as also discussed

Shannon and Weaver, 19^9

Chapanis5 1971 :

" I have yet to find a singleinstance in which psychologicalresearch on communication theoryhas contributed to the solutionof any practical psychologicalproblem. For one thing, the bits,bytes or chunks of communicationtheory are like mouthfuls of saw-dust. They are as mindless as theyare tasteless. "

Miller, 1956

on page 11. It is important to examine by a reviewof the literature and by experiments whether thecontroversy is due to the setting for experimentson human data capacity in psychological laboratories.Very often such experiments deal with conscious classi-fication of information according to prescribed andverbally formulated definitions of the attributes ofthe different items. Such experiments indicate alow data capacity. However, in every-day situations,experience indicates a high data capacity when deal-ing with recognitions (e.g. of faces) or operationsin familiar routine environments (car driving).

Is it important that man in the second caseoperates by "information units11, "chunks" with avery high degree of redundancy in the information,and has a possibility to choose his own character-izing attributes? And operate subconsciously? Textis difficult and slow to interpret if redundancy isremoved; you may recognize a person, but not recallif he wears glasses. Experiments on recognition ofnon-verbal sound pictures tend to show a low capacity- because the subjects are asked to verbalize thecharacteristic attributes? You may recognize peopleby their footfall, but not verbalize why. In otherwords: Does the laboratory experiments measure the -data processing capacity of the conscious processor,while every-day experience indicates a high proces-sing capacity of the subconscious processor?

- 26 -

8. THE SEQUENCE OF MENTAL ACTIVITIES INOPERATOR TASKS.

The aim of the model of a human data processordiscussed in the present report in engineering termsis to facilitate the matching of the formatting andencoding of data displays to the different modes ofperception and processing used by human process con-trollers. To do this, the work situation of the pro-cess operator necessarily has to be structured insome kind of generalized model and the tasks he issupposed to perform have to be identified and charac-terized in relation to his mental operations as wellas to the system he operates.

This is a difficult problem. Very often the pro-cess operator has a sequence of interacting activitieswhich it is difficult to break down into identifiedsub-tasks without crude generalizations; the lengthydiscussion in the literature of task taxonomy has notresulted in the structurization needed in our context.However, model or no model, display systems are de-signed all the time, and a crude model is better thanno model, even if it is only useful for giving apreliminary break down of our engineering probleminto sub-problems more manageable in interdisci-plinary discussions.

The adaptability of the human process operatorto routine situations makes the interest of displaydesigners focus upon the problem of supporting theoperator in his diagnostic task during abnormal plantconditions. However, this task cannot be studied inisolation, as experience from accidents clearlyshows that performance in abnormal task conditionsis strongly influenced by the normal routine tasks.A description of the diagnostic task must be imbeddedin a general description, also covering routine tasks.

The model of the sequence of mental activitiessuggested below has evolved during different attemptsto analyse verbal protocols and observations of thestart up sequence in a power plant, made in a controlroom. It is based upon a previous crude model, which Rasmussen, 1969resulted from analyses of accident records. The lineof reasoning is based upon the following assumption:the characteristic steps of a mental task-I.e. thesequence of steps between the initiating cue and thefinal manipulation of the system - can be identifiedas the steps a novice must necessarily take to carryout the sub-task. Study of actual, trained perfor-mance may then result in a description of this per-formance i#\ terms of shunting leaps within this basicsequence.

Data processingactivities

O

EVALUATE

performancecriteria

which goal to choose ?

States of knowledgeresulting fro« dataprocessing

the effect ?

INTEftPRETE consequen-ces for current task,safety, efficiency,

etc.

which isthe goal

then thestate ?

IDENTIFY

present state ofthe system

what*s lies behind ?

which is the appropriatechange in oper. cond.?

appropriateof syst. cond

OBSERV

information anddata

what's going on ?

ACTIVATION

Detection of needfor data processing

plan sequence ofactions

how to do it ?

FORMULATE PROCEDURE

•release of preset response —EXECUTE

coordinatemanipulations

Simple model of the sequence of human processing activities betweenInitiation of the response and the manual activity.

The diagram shows consequetive states of knowledge separated by mentalactivities to transform one state of knowledge to the next. Typical shuntingeffects evolving during training, resulting in symbolic data processing lead-ing directly to a state of knowledge later in the sequence, are shown» Directassociation leading to leaps directly between states of knowledge is notshown*

Study of the verbal protocols shows that it isdifficult in real control room environments withhighly practised tasks to identify the mental activityas such« The verbal statements are normally more likestatements of different states of knowledge resultingfrom covert data processing.

The model of the activity sequence is shown onpage 27* The basic sequence is based on a set of con-secutive states of knowledge about the system stateand the task requirements, separated by mental ac-tivities to transform one state of knowledge to thenext« This is clearly a very idealized description«The individual mental activities may not be clearlyseparated in time, and leaps backwards and forwardsin the sequence may often occur.

A trained operator will only occasionally haveto move through all the steps of the basic sequence«Two types of shunting effects - often resulting fromsubconscious or intuitive data processing - normallytake place:

- Mental activity may take place at a symboliclevel, resulting directly in a knowledge state laterin the sequence. E.g. holistic perception may leaddirectly to an observation of the system state oridentification of the relevant task, rather than toan observation of a set of separate items of informa-tion.

- An association based upon previous experiencemay result in a leap directly from one state of know-ledge to another one in the sequence.

A given conscious mental activity will be init-iated only if an uncertainty is recognized at thepreceding state of knowledge. If there is no uncer-tainty, either the state is simply accepted and thecurrent sequence stopped, or a leap in the sequencetakes place. The leaps are equivalent to a shuntingout of activities at higher levels of abstractionwhich call for complex conscious reasoning; as a re-sult they give rise to a considerable increase ofdata handling capacity.

8.1 Initiation of the operator's activity

The activity of an operator can be initiated indifferent ways. He may be alerted by an externalsignal or message from the system or he may receivean interrupt from his subconscious dynamic model -i.e. his nprocess-feeling!f tells him that the timehas come to act. Only if he is caught by an externalinterrupt when his dynamic model is not properly up-aated or is unsynchronized, will he have to start thebasic sequence at its entry. Depending upon his degreeof training, i.e. the development of his subconsciousnode!, the internal interrupt will arrive at higher

see also Gagne, 1963

op. cit. p. 5k

Detection of needfor data processing

— release of preset response-

- 29 -

levels of abstraction in terms of time for an expec-ted functional state of the system or a familiar taskto perform. He may then proceed from this state ofknowledge in the sequence or leap back to the obser-vational activity to verify or elaborate on the inter-rupt .

It is important to realize that the subconsciousmodel not only depends upon instrument readings, butupon all sources of information available, e.g. noisefrom the system like relay clicks in controllers ornoise from recorders; the previous operations uponthe system; slight changes in the response of thesystem to manipulations, etc.

In an automated process plant his subconscious mo-del may readily lose its proper synchronization, andhe is generally alerted by a warning signal from thecontrol system, especially in case of abnormal plan bconditions.

8.2 Observation and selection of information

Being alerted, the question in the basic sequenceis: "What's going on?", and the operator must lookaround to get information,from the system. In a controlroom he basically obtains readings on temperatures,flows etc.

what's going

He can not possibly read or store all the avail-able information; he must select relevant sourcesand the reference system for this choice will be hissubconscious model. If this is not developed, he hasto rely on learned rules or to base the choice uponhigher level identification of the functional impli-cations of different sets of data. When the subcon-scious model is highly developed, it will not onlyefficiently control the selection of data* but itsgeneralizing influence upon perception leads to obser-vation not of individual symbolic data, but of inte-grated sets in terms of their higher level implica-tions as functional status of the system or presenceof the need for a familiar task.

Our records from process-control rooms indicatethat the subconscious model or process feeling ofhighly trained operators is developed to a degreeüfhere he is not "reading instruments", but ratherasking yes/no questions to the control panel to test/hether his intuitive guesses of plant state arecorrect or not» This leads to a very efficient shunt-rig out of low-capacity, higher-level conscious ac-ivities* However, at the same time it results in

"ery selective attention, which does not leave theoperator open to unexpected variations of the entireask situation. The extreme case is when the operator

'.at-- a very high expectance of the task to come; the

informationdata

-* '"o«1'«*

Vv

external interrupt may then release an immediate presetresponse without any consideration of further informa-tion from the environment.

On the other hand, it should be realized that alsothe limited data capacity of higher level mental ac-tivities decreases the sensitivity and span of attent-tion when operation moves to such higher levels in thesequence•

The large data processing capacity of the intuit-ive shunt paths is also needed during periods when theoperator is constantly alert and monitoring plant oper-ation to detect abnormalities. In our experience, theprocess operator may be left with a critical monitoringtask5 for instance during initial plant operation orperiods following major repair or modifications. Dur-ing such periods the plant may not be completely pro-tected by the automatic monitoring system.

Detection of an abnormal situation takes placewhen man's environment deviates from the expectedcourse and thus initiates his action. The operatorhas to know the "expected course", i.e. to ha:ve areference frame defining the normal state. To havespecified limiting values of individual variables istrivial, and is the way automatic alarm systems nor-mally operate.

To follow or monitor a start up to see " if every-thing is normal" will rather be a check of plantstructure and parameters by monitoring the constraintsinterrelating the displayed data. If he has to resortto higher level mental activities such as consciousidentification and interpretation, his capacity, i.e.speed of operation will seriously decrease; he will be-come selective and his attention will then be focusedupon preconceived hypotheses. The monitoring task con-sists instead of being open to the unexpected events.

To be open to unexpected events and not locked bypreconceived ideas may imply that the data presentedby the environment should be related to the subconsc-ious dynamic model of the environment. If so, the in-formation displayed should be arranged in graphicalpatterns for which general subconscious models areeasily formed, as for instance the display suggestedüj Wohl for monitoring servo loops. Or the displayconfigurations should be related to subconscious models/hich are maintained by the man during his normal pro-fessional activity. Depending upon the staffing andcask allocation during check-out, the normal activityof the man given the monitoring task during plantotart-up, or restart following' major repair, may beplant design or maintenance as well as plant operation.?he subconscious model in use may therefore vary ints relation to the different subconscious "control

Coekin, 1969:

"A versatile representation ofparameters for rapid recognitionof total state"

Rasmussen and Goodstein, 1972:

Size of crosses indicatesradiation levels aroundreactor DR-2.

Wahl, I965:

Irregularities of a tiansientare easier to see in a spiralthan in an oscillation

7

2

Display osuggested(internal

2

t U r\ t

by 1

3

•. Ska?.: ond i t ion

sjburgi

see also Gooastein v<> •al*»-";l9$|

surfaces", located at different levels in the system,as discussed on page 9«

A problem in the design of displays is to presentthe information so as to support such holistic percep-tion and at the same time to find coding formats en-suring that the information patterns adopted as state-characterizing patterns are also adequate as state-de-fining patterns; i.e. that variables only relevant inspecial, but important states are significant .parts ofthe pattern.

8*3 Identification of system state

Activity within the basic sequence following theobservation and selection of information phase isinitiated by the question: What lies behind this in-formation?11 • In order to judge the situation and laterplan the relevant action, the operator has to identifythe internal state of the system, from the constraintsinterrelating the observed information.

This is a complicated cognitive process which isvery poorly understood in real-life tasks. Basically,different data processing procedures may be used bythe operator depending upon the mental processingmodels he has available and chooses, and upon his pre-conception of his immediate goal:

-He may correlate sets of observations to prelearn-ed or previously experienced sets of data characteriz-ing - and hopefully defining - specific, known systemstates.

-He may use information on system topography oranatomy to trace abnormal states through the system.

-He may use information on physical functioningof the system to trace cause-and-effect chains throughthe system.

-He may base his activity upon intuitive guessesand use his mental models to test such hypotheses.

It is obvious that he will frequently need sup-plementary observations during, this processing andtherefore he will return to the observational phase,but such observations will be either very specific orsimple yes/no questions to the system, not open-mindedobservations.

The shunting effects leading to direct percep-tion of the system state from the observations havebeen discussed previously. Such identifications may beused without questioning, or they may be taken as hy-potheses and tested by conscious data processing.

Basically different modes of human data processinguay thus be in operation during a specific task, not un-ike the ffzeroing-intf and ffcount-outn phases of ahess player's procedure, as discussed by Dreyfus, whichange from subconscious pattern recognition and appre-

IDENTIFY

present state ofthe systea

what's lies behind ?

"•

** /«*

Dreyfus, .196.5

elation of a dynamic situation to consciously guidedsequential search or reasoning. Correspondingly,quite different types of data conditioning and pre-sentation may be appropriate for support.

In abnormal plant conditions, the diagnostictask of an operator can be very critical and there-fore the aid given him through suitable design ofdisplay equipment is important. In this respect, theenvironment and task of the operator have specialfeatures which will be considered in more detail.

The plant operator is only able to protect the •plant, or plant operation, against the consequences ofa fault if there is a reasonable delay between the pri-mary fault and its possible consequences. Thus diagno-sis by the operator during plant operation will be relatedto events characterised by a delay, typically due topile-up of energy in the plant system (i.e. "integra-tion" due to a disturbed mass or energy balance - suchas high temperature due to a power imbalance or highpressure due to a mass flow imbalance), to transportdelays (such as cold plugs in once-through boilers)or to delay in control system data processing.

Relevant to the present discussion is the featurethat an important group of events is related to aspatial, temporal system - something moves around ina stationary scenery - not unlike systems, e.g.traffic systems, which man handles very efficientlyin the proper setting. The problem is that the "mov-ing something" is inside the system or an abstractquantity - e.g. energy - and has to be visualized ina proper way. Can computer graphics so visualize theprocess that it matches the ability of the subcons-cious processor to handle spatial-temporal systems?

This depends presumably upon the actual goal ofthe operator in a specific situation; i.e. whether heis considering the continuation of normal plant oper-ation, the security of the plant syster^ or he is aim-ing at a repair of the system. The ability to effic-iently handle spatial-temporal systems is presumablyof most importance when the operator considers plantsecurity. He then has to identify ."what is wrong" inorder to correct the operational state. The task isto identify the balance which has been upset so as tobe able to re-establish the balance before dangerous•'pile up" has taken place. In the first instance, anIdentification of "why" is not of importance. A poweralance, for instance, can be re-established by aecrease of power input irrespective of whether theause of the upset is abnormal input, load, or coolantlow.

- 33 -

When the plant has been brought to a prelimi-nary, stable and safe state of operation, theoperator can turn to another mode of diagnosis anddetermine "where is the fault" and localize the faulty-component in the system. This is another type of task.Typically the time pressure is lower and the operatorwill probably have available different types of con-scious search procedures based upon different mentalmodels of the system, as discussed for electronic sys-stems in a previous report.

The mental model and data processing procedureneeded by the operator depend then upon the situationin which he has to identify the system state, and thedisplay system should be able to support his modelby information related to the relevant type of con-straints, e *g« :

»prelearned sets of descriptors,-plant topography (locations),-plant anatomy (connections),--plant function (cause and effect relations).

If the system condition identified is familiarto the operator, the shunting and associating effectnormally leads to later phases in the sequence, asillustrated by the figure (p. 27 )• If not, his un-certainty will lead him to ask the next question withinthe basic sequence: "What are the implications of thisstate?" and thus to proceed to the next activity -interpretation of the system state.

8«A Interpretation of the situation and the perform-ance criteria

When the operational state of the system has beenidentified, the trained operator will generally knowwhat to do and therefore he may bypass the interpretivephase.

Rasmussen and Jensen, 1973

The novice, or the trained operator finding aunique plant state, must go through an interpretationof the operational consequences of the plant state.This implies that he has to make decisions. The identi-fication of the state will probably be ambiguous, leav-ing him with different possible hypotheses. Furthermore,the consequence of a plant state has several aspectssuch as the effect upon efficiency of plant operation,upon the risk of plant damage, or the influence uponother current tasks.

The operator will have to predict such different-•/spec t s of the consequences, and find a system stateminimizing them. The different aspects may very readily.Yead to conflicting requirements for the proper system;tate, and the operator must decide on the relevant goalo persue. This initiates a higher level task of inter-re tat ion of his performance criteria, i.e. his own

which goal to choos«

*t'« tS« ftfffttft t

evaluation of what is expected from him in terms of costand risk consciousness and of efficient and dependableperformance.

The mental model required in this interpretationand prediction phase can be based upon different typesof system constraints, such as basic knowledge of theanatomy and functioning of the system, or sets of de-scriptive and defining data which may be prelearnedfrom the system designer or collected by the operatorduring his prior activities«

Only in very unique situations will a trainedoperator have to move to this high level of abstraction."However, it can be crucial in critical situations thathe does not bypass this activity, and therefore suit-able support from the display system is important. Thisis especially true because the mental data processingmay be very complex and stressing since the operatorwill have to consider several lines of reasoning toarrive at his decisions.

The display should not only attract his atten-tion and initiate his conscious consideration in uniquesituations so as to avoid leaps, but it must also sup-port the relevant mental models by relating the datato the proper sets of system constraints. It willalso be important to call the attention of the opera-tor to possible ambiguities in the situation demandinghis consideration of alternative sets of states, con-sequences or goals in his hypotheses and judgements.

8.5 Evaluation of the appropriate task and procedure

Knowing the operational state to which the con-trol actions should lead the plant, the operator hasto decide upon the change of operating conditionswhich will bring the plant from the actual to thedesired state; i.e. he has to formulate his task withreference to the internal state of' the plant.

The operator is now about to move down from thehigher levels of abstraction towards more specifictechnological details: he has to identify the possiblechanges in the anatomy of the plant (like switchingor valve manipulating) and the operating conditionsof the plant and its sub-systems which will lead tothe desired plant state.

To support this mental task, the display systemshould relate not only measured state data but alsoi/he means available for changing the operating con-lit ion (valves, switches, control parameters) to theplant anatomy and function. This may be important ifhe operator has to improvise in unique situations andthus also has to consider the use of control meanshxch are only occasionally used*

see also Rasmussen, 1969

which ia th« appropriatechange in op«r. cond.?

•eloet appropriatecha'af* of ay at* cond

Venda, 197o:

Data displays and control keysarranged for different mentalmodels

The final mental task will be to plan the proce-dure for manipulations; i.e. to relate the desiredinternal changes in the system to a seuqence of mani-pulations at the surface of the system, e.g. via thecontrol panels. The suitable support of the operatorin planning and execution of the manual procedureswill depend upon the specific tasks and work condi-tions; e.g. whether the step from an identified taskto the proper procedure is based upon his knowledgeof plant anatomy and functioning or instead should bebased upon work orders and instructions. Equallyrelevant is the consideration of whether he is in anirreversible situation, where mistakes cannot becorrected immediately, or whether the manipulationsare reversible and experimentation possible.

The novice will have to compose his sequence ofmanipulations consciously from elementary autonomousmanual routines. As training expands, he developsmore and more integrated and complex autonomous se-quences which will be.executed subconsciously wheninitiated and which will finally be labelled inhigher level terms, such as task or even plant state.

8.6 Preferred mental sequence in familiar tasks

The proceeding discussion illustrates how train-ing gradually changes the very detailed functions ofa novice? who extracts his knowledge about the systemstate from individual observations and coordinates theelementary operations of a manual sequence, into the hol-istic perception of states and the repertoire of auton-omous manipulations, available to a skilled operator inroutine tasks. The low-level functions become increas-ingly integrated as they are transferred to subcon-scious data processing.

Analysis of a protocol from routine work con-ditions — start up of a power plant boiler - indicatesthis clearly, and shows frequent use of the followingsteps in the mental activity sequence:Observations directly in terms of plant state-* statementof the task to perform -K execution of a manual sequence.

The model of the mental activity sequence is onlyto be seen as an illustration of basic aspects of themental sequence and, as expected, variations arefound in our protocols. A frequently used variation inthe protocol is:

-The internal model, which is updated subconscious-ly by previous tasks and observations, interrupts andmakes the operator aware that the time has come toexecute a specific task that is directly stated interms of the task or sometimes the procedure.

-The operator then makes some observations andthe information perceived is stated in terms ident-ifying plant state; often the observation is a yes/noquestion to the system as to whether the appropriatet)lant state is present or not.

v^Related to system anatomy.

Related to system response (controlsensitivity and time constants).

)—Bis -J!\*Related to procedure.

The work situation during whilthe protocols are collected,|£gdescribed by Pedersen; 197^

- 36 .

-If not, he goes on to a more detailed identifyingactivity; if yes, he executes a trained sequence.

-The operator next makes some observations interms of plant state to verify whether the executionof the task brought the system to the expected state.

The activity of a man "operating by experience"can be seen as a conscious control of the relationbetween sets of familiar plant states and tasks,while the transformation from the displayed informa-tion to plant state and the coordination of the associ-ated manual task, sequence is taken care of by thesubconscious modelling system.

Clearly it is difficult to study the mental pro-cedures and models used by trained operators. Thehighly practised routines are not immediately access-ible through verbalization, and it is difficult toobtain detailed verbal records during rapid worksequences in abnormal situations. The relationshipbetween the level of training and the correspond-ing mental procedure available to an operator suggestsa study of the change of the procedures during thelearning of a technical skill. A study illustratingthe change from a procedure based upon the under-standing of the internal properties of the problemto a procedure based upon the external visual featureshas been reported by Nielsen, who studied countingsupported by pencil marks. This study indicates thata shift in mental procedure may be identified inverbal protocols and interviews during training per-iods»

Short-hand diagram used for quick access to pro-tocols during analysis. X indicates activity.0 indicates statement of knowledge. The sampleillustrates the most frequently used states inthe sequence and the use of leaps.

Nielsen, 1964:

Count:1-2-3-4-5, 1-2-3-4-5,-—

Then try to count:1-2-3-4, 1-2-3-4,—but continue marking insets of five.

M In the beginning, the subjectsdid not give the strokes any sig-nificance beyond the fact that oneof them was drawn each time a numberwas spoken. The subjects then expe-rienced some pencil marks whichevolved visually and/or throughmotor responses into certain figu-rative entities which had nothingto do with the spoken numbers" .

op. cit. p.75.

Evaluation ofperf. criteria

Activity mentionedTilt, goal stated

- 37-

8*7 Shunting effects and mentål capacity

It is often stated that the amount of informationpresented to the operators of a modern plant has reacheda size which makes it important to find means for reducingthe number of data, so as not to overload his input capac-ity. But the problem is not that simple. It has pre-vously been touched upon (p. 11 ) that the limitingfactor in real-life environments may not be input butprocessing capacity.

Analysis of trouble-shooting records indicates acomplementary relationship between the complexity of •the conscious processing and the amount of data usedfor the processing. The same relationship is found inthe processing sequence discussed above. Intuitive dataprocessing performed by subconscious leaps and as-sociations bypassing more complex higher level reason-ing relies upon the large capacity of holistic percep-tion and recognition, as well as the large data proces-sing capacity of the subconscious functions of general-ization and modelling. However, this mode of proces-sing depends upon simultaneous presentation of a setof "data" large enough to allow for discrimination byholistic identification.

The problem here is not to limit the amount ofinformation, but to present a large amount of informa-tion in a way suited to holistic perception, and at thesame time to ensure that the information used bythe subconscious processing as state-characterizinginformation will also be sufficient as state-defininginformation. This is important to avoid routine leapsin a non-routine situation.

The capacity of higher level mental operationsis less due to the use of conscious reasoning withthe sequential use of individual observations, andthe use of the small capacity, short term memory, etc.If the amount of data presented is large, the capa-city may decrease due to the selection problem. Thereliability of processing may be low, due to the needto limit the number of observations taken into accountto within the limits of the short-term memory capacity,thus narrowing the attention to a subset of the state-defining data. Again the solution may not be to reducethe information given to the operator, but to presentthe available measurements and data in suitable setsand configurations allowing perception and processingat an appropriate level of information or data?lchunks"»

It is not unrealistic to expect a relationships shown on the sketch between performance andmount of data available in a specific task. If theumber of data in a given presentation is reducedt lie performance may increase, but it may very well

Rasmussen and Jensen, 1973

Newman, 1966:

"(It was) found that properlyformatted displays allowed peopleto tolerate and absorb much moreinformation than would normallybe expected. There seems to be animportant principle operatinghere, one of considerable generali-ty: People don't mind dealing withcomplexity if they have some wayof controlling or handling it(If) a person is allowed to struc-ture a complex situation accordingto his perceptual and conceptualmeeds, sheer complexity is no barto effective performance."

Performance(speed, reliability)

Presentation calls for ;sequential

Amount of information

be that performance is increased even more if theamount of data presented is increased.

8.8 The relationship between the frequency of anoperator task and the risk implied

As discussed in the previous sections, the waysand means used by an operator in his data processingactivities vary fundamentally with the task conditionsand the level of training of the operator. To matchthe display coding to the way an operator will solvea specific task, it is therefore necessary to corre-late at least the basic characteristics of the oper-ator's mode of data processing to his work conditionsin the different types of tasks.

Generally, it can be assumed in a well designedplant, that the events calling for the operator'sintervention will have a frequency which is inverse-ly related to the risk related to the event. Thisleads to the assumption of the relationship shownbetween the frequency of a task and thus, implicitly,the level of training, the basic features of dataprocessing and the risk involved. In the very fre-quent tasks, such as set point adjustments, switchingoperations during frequent start-stop operations, etc.the operator will act by a holistic perception and anintegrated manual sequence forming a complete autonom-ous sequence. This sequence of subconscious leaps willdevelop during training , irrespective of whetherlearning was based upon an education in plant anatomyand function, or on work instructions. Normally therisk involved in such tasks will be low. The designerwill protect the system against simple mistakes andmaloperations because the task will be recognisedand considered in the design phase. Furthermore, thetask will very often have a reversible effect upon theplant and the operator may readily correct malopera-tions.

Frequencyof the task

Sub-conscious,automated routines

Experience

Instructions,rules

Improvisations,functional

reasoning

Risk related tothe task

In less frequent tasks, such as infrequent startprocedures, calibrations,and response to more trivialfault conditions (e.g. in instruments), the operatormay operate by direct pairing a perceived plant statewith familiar trained manual routines from his pre-vious operating experience,

Normally the plant designer find some operatortasks to be too infrequent and important to rely uponroutines developed by the operator and he thereforeissues work instructions. The data processing of thehuman operator then is expected to follow a predeter-mined sequence and the data presentation has to berelated to this mode of operation. Theffdata proces-sing model" is then given implicitly in the prescribedsequence, and the presentation should support theidentification of the attributes defining the eventcovered by the instruction and the physical items to

- 39 -

be operated. In addition, the recall of the sequenceshould be supported.

It is a general experience that operators tendto "improve" instructions operationally, but can thisbe avoided without presenting the instructions to him?If the instructed sequence is not stored in his sub-conscious internal model by proper training throughexercises, how does he then store the sequence? Bytraining of his speech motor system like learning anursery rhyme in primary school or by visualizationof the text? Or most probably by deducing the necess-ary steps from the supposed aim of the instruction? .

However, situations involving high risks in awell-designed plant may be due to a coincidence ofseveral abnormal events. Normally such situations willhave a very low probability, wherefore it is extremelydifficult for the designer to envisage all relevantsituations. As a rule he will protect the systemagainst this type of event by an automaticsafety system which detects events in a later and moregeneral state of the cause-and-effeet-chain. However,experience shows that due to the variability of humanfaults, human actions may lead to system states notcovered by the automatic protection system. Especiallyduring periods following major repairs and modifica-tions of the system, when faults leading to such uniquesituations may be present. In these situations theoperator will be forced to improvise and base hisintervention upon his fundamental knowledge of plantanatomy and functioning. Therefore he must work con-sciously through several steps of the basic mentaltask sequence.

8.9 Support of the operator in critical^infrequentplant conditions

The general conclusion from analysis of accidentrecords tends to be that an important problem in plan-ning of the man-machine interface is to make it possiblefor the operator to correctly identify the work condi-tion in cases of infrequent, but critical operatingstates of the system by:

-counteracting the operator's inappropriate leapsand associations based upon subsets of the informationpresented by initiating and reinforcing in the oper-ator a state of uncertainty which will thus lead tohigher level mental activities«

-supporting the mental models and proceduresneeded by the operator in accordance with the mentaltask he is supposed to execute, especially in identify-ing plant state and evaluating consequences and choos-ing the goal to pursue. This includes assistance inidentifying and recalling existing, relevant, writtenwork instructions.

The digital computer provides a very flexiblemeans of supporting the operator but in order toavoid conflicts of responsibility, a clear generalphilosophy should be followed. If the operator isresponsible and is supposed to make real decisions,the data processing and display system should supporthim in a way which will not interfere with the waysof reasoning and reacting which are felt to be rea-sonable and natural in the actual situation.

In a process plant, especially in abnormalsituations, plant dynamics will be forcing the oper-ator's work sequences. He will be in time stress, andit may be unrealistic to try to compel him to con-sider very unlikely, but possible and dangerous sys-tem states, if hypotheses suggesting more probablecauses are at hand. His procedures will very likelybe controlled by the preference for "the way ofleast resistance" and have "points of no return" asdiscussed for electronic trouble shooters.

In other words, the trained operator can beassumed to try out hypotheses from the high probabili-ty end of the curve shown (page 38), whereas, thesafety officer would tend to be conservative and wanthim to first consider hypotheses covering possiblehigh risk situations. If safety considerations there-fore lead the designer to introduce automatic diag-nostic programmes to counteract the operator's natu-ral tendency, the designer should assume full respon-sibility and the programmes should lead to automaticactions or presentation of clear orders.

Such programmes will have to be conservativesince they cannot possibly take into account allpossible plant and operating conditions. Thereforethey should not lead to presentation of hypothesesto be considered by the operator leaving the deci-sion to him. There is a good chance that such hypoth-eses may either be trivial or too'conservative. Theoperator should be allowed to trust the informationpresented to him, and rather than force him to con-sider possible, but a priory improbable,hypothesesresulting from conservative automatic diagnoses, thecomputer and the display system should assist him inreferring his attention to the relevant part of thesystem and supporting him in a rapid verificationof the hypothesis he considers most reasonable andprobable.

Facilities to assist the operator's tests of hishypothesis can be based on the designers analysis ofnormal plant function! which is far more realistic thana complete analysis of the plant under conditions offailure.

Rasmussen and Jensen, 1973, P» 5^:

11 A main rule for the structur-ing of the procedures seems to bethat as soon as an observation isfound to give a topographic refer-ence to a more restricted field forwhich a familiar search routine isat hand, a decision is taken toswitch to that field, to followtheMway of least resistance".

There seems to be a "point ofno return" in the attention of theman the moment he takes such a de-cision, as discussed by Bartlett.Although more information, indicatingpossible short-cut methods or import-ant hints for the next search, isclearly available from the obser-vation and is recorded by the man,the decision prevents any influencefrom such information; hence thenext search is a routine, startingfrom scratch.---"

The large storage capacity of the digital com-puter can also assist the operator in a systematicuse of plant operating experience; e.g. by storingsets of data related to verified plant states whichthe operator can use later to check the consistencyof his diagnosis in the event of reoccurrence of therelated hypothesis.

The operator1 s "LADDER OF ABSTRACTION*'

fi•Æmurtducépou>er input at

cooling of —

3 start—syster»- by-I*.

f,_'_50p»i«

i r „ n>,k„ «>ch..».. tloo* •<

7

J \ \ata

Si

/ . / • • • •

40 60 80

-2WF,

/ :

- / :

100

9. HUMAN RELIABILITY

Post-event analysis very often identifies the hu-man element as a critical factor in serious systemfailures and, too often, the conclusions of the analy-sis classify the cause as operator error and recommend"Stricter administrative control'1 of the work proce-dures. There is, however, a great need for a far moredifferentiated view of human failures in system designas well as in post-event analysis if effective counter-measures are to be found.

System design and system operation seem tradi-tionally to be the activities of two different worlds.If it has been demonstrated during system develop-ment and commissioning that the operating staff canoperate the system satisfactorily it is generallyassumed that they will be able to continue satisfactoryoperation througout the life of the plant if they aresufficiently careful. Therefore, maloperations not dueto technical faults are classified as operator errorsand this seems to be a general definition of operatorerrors today.

To improve system reliability, more integratedconsideration must be given to human errors during sys-tem design and operation. If operator failure is dueto normal psychological mechanisms during the actualevent, the cause should rather be ascribed to thework situation and therefore possibly to a human faultduring design. Therefore, to require the operators to??be more careful" has only little effect. This point CRigbyi.3-969of view has been advocated strongly by Rigby and Swain.

Swain, 1969

The appropriate means for counteracting operatorfailures depend upon which step in the mental dataprocessing sequence is the most vulnerable in aspecific work situation. The present model of the se-quence suggest the following examples of failuremechanisms:

- Conscious mental activity is not activated., ata step in the sequence needed in the specific task,due to leaps bypassing the step. The cause maytypically be that features of the perceived informa-tion closely match features related to anotherfamiliar situation, for which the operator has afixed response available.

- Conscious mental activity is activated at theappropriate level, but performed as a test of a hy-pothesis generated from an intuitive leap. The observa-tions made for the test may then be in the form ofyes/no questions to the system, and use of an inappro-priate subset of information may readily follow«

See also Russel Davis, 1958.

- Conscious mental activity is activated at theappropriate level and the relevant observations made,but the data processing is incorrectly performed, dueto the difficulty -following e.g. from the limitedcapacity of the short-term memory- in taking into ac-count all the available information. This tendency mayvery likely be increased during periods of stress.

- Incorrect performance can follow when the appro-priate mental model for conscious processing is insuf-iciently developed or has degraded. If this is recog-nized by the operator he will have to update the modelby memorizing, consulting diagrams and manuals, etc.ra subtask which will take time and further load hisshort memory.

In process plant environments it is of importancethat incorrect data processing normally not only leavesan operator task undone, but leads to the execution ofanother task, which may be inappropriate in the speci-fic situation and thus lead to a coincidence of twoundesired events in the plant.

These are only a few illustrative examples offailure mechanisms following from the proposed modelof the sequence of mental activities. The importanceof these mechanisms and other possible sources of hu-man failures can only be judged from detailed analysisof real life events.

The main issue at the present state will be thatthe likelihood of human failure and the appropriatecounter measure does not only depend upon the workcondition during the specific task but, equally impor-tant, on the individual operator's data processing dur-ing his daily routine work which results ina great repertoire of normally extremely efficientleaps and associations in the data processing. If asub-set of features in a state of his mental activityin a specific task closely matches a step in a highlytrained sequence, the attraction from such a sequencevery likely causes a switch-over, resulting in a bypassof higher level activities.

Such affinities between menta,l procedures leadingto erroneous performance in a specific task cannot befound from analysis of this task alone. Only a modelstructuring the sequence of mental activities andstates and enabling a comparison of the states passedin different tasks will make it possible to judge whenand how a switch-over may take place.

It is not always realised that the appropriatemeans for counteracting human failures depend stronglyupon the mechanisms involved in the fault. Does it,for instance, improve the performance of a traineddriver involved in a traffic accident to force him back

Discussing human errors and trans-port accidents:

The false hypothesis: " Theseexperiments, with many others, showthat one tends to perceive what oneexpects to perceive." In othercases what is seen may be deter-mined by the observer's previousexperience, recent or remote, "" It is usual for a person tohave expectations, or to hold towhat may be called an hypothesis,about every situation he meets,even the information is notablyincomplete. This hypothesis, whichis in some degree the product ofhis previous experience of similarsituations, governs the way inwhich he perceives the situationand the way in which he perceivesthe perceptual material availableto him. ""—People may fail to see whatstares them in the face if theirlooking; is guided by a false hy-pothesis. "

Preoccupation: " When anxiety wasat a low level at the beginning ofa test, the subject tended to re-spond to the instrument panel asan integrated whole. As his anxietyincreased, the several componentsof his task tended to become dis-articulated. His attention thentended to be directed, more and more'to the particular component of thetask which had gained a specialimportance. "

- /fif -

to driver's school for verbal rehearsal of the instruc-tions, e.g., how the distance between cars should berelated to the braking distance, as derived from speed,reaction time and road conditions? In other words,attempts to improve performance of a highly trainedman by updating his knowledge of system function andof operating instructions mainly relevant for conscioushigher level processing, may not help at all since suchprocessing is very likely to be bypassed in the actualwork situation.

A general, basic way of changing the work situa-tion of the operating staff is to automate a task bymeans of suitable control equipment. This does notremove the operator from the system, but, used in aproper way, it may improve the reliability of hisperformance. Automation of routine tasks during plantoperation and start-stop operations will reduce hisrepertoire of stereotyped leaps and associations, and,by replacing his routine tasks with'other meaningfultechnical tasks, it is possible to give him a betteropportunity to maintain the basic knowledge of thesystem needed under abnormal task conditions.

A plant can be automatically brought to a safeoperating condition, when it is malfunctioning, bymeans of automated protective systems, if the necess-ary automatic diagnosis can be based upon rather gen-eral criteria related to overall effects of failuresupon mass and energy flow balances. This removes theoperator from task conditions calling for improvisa-tions in stressed situations when he is forced by thesystem dynamics; instead, he is given the role ofinspector and repairman. If the protective systemkeeps the system operating under safe conditions withequipment within operating range, the operator willhave the best possible working conditions.

Generally speaking, proper use of automation willnot remove the operator from the system, but eliminatea wide range of possible tasks calling for improvisa-tions during stressed situations and replace them bya much more uniform class of tasks related to inspec-tion, test, fault-finding and maintenance. Theseare more easily foreseen and planned in advance andtherefore more accessible for human reliability pre-diction. Ideally, no critical task for which the humanreliability can not be reasonably well estimatedshould be left to the operator, but rather transformedby the proper use of automation to predictable tasks.

10. PREDICTION OF HUMAN RELAIBILITY

Several important approaches have been madetowards the development of methods for predicting hu-man reliability. Such methods have recently been re-viewed by Meister. The basic assumptions of thesemethods are typically:1. The task is well defined and the procedure followedcan be formulated in detail.2. The procedure can be broken down into a sequence ofbehavioural units, i.e. subtasks or task elements.3* Data on the reliability of the individual sub-taskare available together with the parameters character-izing the relevant task situation.

Typically these assumptions do not fit the workprocedures found in process control room environments:

Re. 1: The work procedures may be known in detailunder task conditions, where the physical environmentpaces the man in a way which forces him to use a knownsequence of subtasks, as is the case in e.g. manualassembly processes. In automated process plants today,however, the human function is typically higher levelmental data processing and decision making, and thehuman work procedures are far less constrained by thephysical environment. As discussed in this report, thecreativity and adaptability of man often results in theevolution of several basically different mental proce-dures for the same type of task, all capable of endingup with the same result. The mental procedure adoptedin the actual task is therefore difficult to predict.

Re. 2: A prerequisite to be able to use the classicalreliability methods for evaluation of human behaviouris the breakdown of the procedures used into a se-quence of typical and generally used units. This break-down also can be done for a task in which the worksteps are defined and cued by the environment as itis the case for manual tasks in production. But againthis is not the case for higher level data processingtasks in plant environments. As we have seen, importantand complex steps in the mental work sequence may bebased upon holistic or integrated processes causingintuitive leaps and associations which cannot bebroken down into elementary and general elements.Furthermore, the "process feeling" of the operatorconstantly keeps him prepared for the normal tasks tocome. This means that he may only be prepared to lookfor very little information at the actual time of atask, and it is not possible to predict whether theinformation actually underlying his decisions isproperly updated. Furthermore, the source of informa-tion chosen may be a convenient source during the nor-mal working condition, such as noise from the system,

Meister, 1972

Rasmussen, 1973

e.g. relay clicks, rather than the information plannedby the designer to be task defining and therefore dis-played to the operator and considered in a prediction.Fundamentally this effect of the "process feeling" linksthe elements of a task sequence together, and theseelements cannot be treated individually. A control roomoperator typically does not perform isolated actionson well-specified bits of information; he is an inte-grated part of a dynamic situation. This causes diffi-culties which are hard to predict when abnormal plantconditions suddenly demand that the operator switches toother tasks and performance criteria. The reaction toabnormal plant conditions can only be treated in the•light of the normal working conditions prior to theevent which establish the process feeling and thus theexpectations of the operator. The reactions can onlybe treated in isolation if the man-machine interfacecan be designed to disrupt routined responses of theoperator and to set the initial conditions of hisdata processing in a predictable way at the start ofa mental activity.

Re. 3« Data for reliability prediction in control roomenvironments are very scarce, and it is questionablewhether the field data collected from process plantenvironments according to present schemes are suffic-iently related to the characteristic features of thetask conditions present during the abnormal eventsgenerating the data. The variety of different modes ofmental data processing used by the operator and therelation between their use and the familiarity witha specific task, may very readily lead to the con-clusion that fault data related to a specific sub-task collected from minor, isolated events andincidents will not at all be representative of theperformance of the same sub-task in more complex andunique situations, due to the fundamental shift inmental mechanisms.

Finally, it should be repeated that the greatflexibility of man as a data processor and a systemcomponent leads to another fundamental difficulty inprediction of systems reliability: the real risk maynot be due to his not doing a specific task, but tothe consequences of what he selects to do instead.

Before a large-scale collection of human failuredata is planned, it is necessary to formulate a tax-onomy of tasks and work conditions related to differentmodes of mental activities, and to verify the founda-tion by careful analysis of the work conditions inprocess plants and by analysis of detailed post-eventrecords from the same plants.\

11. CONCLUDING REMARKS

Being a working hypothesis, the model isbased upon premature generalizations from ratherunsystematic reading and from preliminary expe-riments and analyses. However, the work has sup-ported the experience that a general model coveringmany aspects of the great variety of human abili-ties is useful as a frame of reference in inter-disciplinary exchange. For analysis of real-lifeworking conditions it is important to have a modelwhich is related to a specific task situation,rather than to specific human abilities.

To a systems engineer concerned with theeveryday tasks and working conditions of, forinstance a process operator, it seems as if thecontroversy between different schools withinpsychology is often due to generalizations basedupon different models or experiments consideringonly subsets of human abilities. During the studyit has sometimes been fruitful to study the con-clusion emerging if both parties in such a contro-versy were considered to be right, but on thebasis of different premises.

Also it is important to consider the experi-mental methods of different schools. For instance,the approach taken by behaviourists is a properway to study the aspects of subconscious data pro-cessing in tasks when a man is operating as amultivariable control system, whereas the differ-entiation between the various mental proceduresemployed in a conscious processing task can beproperly studied by listening to a man verbalisinghis internal activities.

The discussions with colleagues, operators,and technicians at and outside the department havebeen highly appreciated as have L.P. Goodstein'smany hints about good references.

REFERENCES

Arnheim, R., Visual Thinking. (University of California Press,Berkeley, Calif., 1969) 345 pp.

Bainbridge, L., Beishon, J., Hemming, I.H., and Splaine, M. ,A Study of Real-time Human Decisionmaking using aPlant Simulator. Oper. Res. Quart. Spec.Conf. Iss. 19. (1968) 9I-IO6.

Bainbridge, L., The Nature of Mental Model in Process Control.In:International Symposium on Man-Machine Systems,Cambridge 8-12 September 1969- IEEE ConferenceRecord No. 69 (58- MMS Vol. 3) •

Barlow, H.B., Narasimhan, R. and Rosenfeld, A., Visual PatternAnalysis in Machines anu Animals. Science 177(1972) 567-575-

Beishon, R.J., A Study of Some Aspects of Mental Skill in thePerformance of Laboratory and Industrial Tasks.Thesis(Ph. D.), University of Oxford, 1966293 PP.

Bower, 6.H., Imagery as a Relational Organizer in AssociativeLearning. J. Verbal Learn. Verbal Behav. £(1970)529-533.

Bruner, Jerome S., A Study of Thinking (Wiley, New York, 1956)330 pp.

Bugelski, B.R., Words and Things and Images. Am. Psychol. 25(197O). 1002-1012.

Chapanis, A., Prelude to 2001: Explorations in Human Communica-tion. Am. Psychol. 26(1971) 949-961.

Chernoff, H., The Use of Faces to Represent Points in n-Dimen-sional Space Graphically. AD 738 473(1971) 48 pp.

Coekin, J.A., A Versatile Presentation of Parameters for RapidRecognition of Total State. In: International Symposiumon Man-Machine Systems, Cambridge, 8-12 September1969. IEEE Conference Record No. 69 (58-MM'S Vol. 4 ) .

Craik, K.J.W.The Nature of Explanation (Macmillan, Cambridge, 1943)I23 pp.Excerpt in: Thinking and Reasoning. SelectedReadings. Edited by P.C. Wason and P.N. Johnson-Laird(Penguin, Harmondsworth, 1968) 431 pp.

Davis, D. Russel, Human Errors and Transport Accidents Ergoiiomies2 (1958) 24-33-

Dreyfus, Hubert L., Alchemy .and Artificial Intelligence.RAND-P-3244 (1965) 98 pp.

Gagné, R.M., Human Functions in Systems. In: PsychologicalPrinciples in System Development. Edited by K.M. Gagn$(Holt, Rinehart and Winston, New York, 1963) 35-73•

Goldstein, Myron, Display Aspects of Algebra. Psychol. Rep. 2̂±.(1969) 937-938.

Goodstein, L.P., Pedersen, O.M., Rasmussen, J., and Skanborg, P.Z.The Operator's Diagnosis Task under Abnormal OperatingConditions in Industrial Process Plant. Risø-M-1729(197*0 16 pp,

Guilford, J.P., The Nature of Human Intelligence. (McGraw-Hill New1967) 538 pp.

Hammond, Kenneth R., Computer Graphics as an Aid to Learning. Science172 (1971) 903-908.

Held, Richard and Hein, Alan, Movement-Produced Stimulation in theDevelopment of Visually Guided Behaviour. J. Comp.Physiol. Psychol. %. (1963) 872-876. Reprinted in Per-ceptual Learning and Adaptation. Edited by P.C. Dodwell(Penguin, Harmondsworth, 1970) 502 pp.

Hoffman, Paul J., Slovic, Paul, and Rorer, Leonard G., An Analysis-of--Variance Model for the Assessment of Configural CueUtilization in Clinical Judgement. Psychol. Bull. 69_(1968), 338-3^9.

Hubel, David H., The Visual Cortex of the Brain. Scient. Am., 209No. 5 (1963) 5^-62.

Huggins, William H., Iconic Communications. IEEE Trans. Educ. E-l*+,(I97I) 158-163-

Johnson, Donald M., Problem Solving and Symbolic Processes.Annu. Rev. Psychol •. 1_ (1950) 297-310.

Lashley, K.S., The Problem of Cerebral Organisation in Vision.In: Visual Mechanisms. Edited by H. Klüver (Cattell,Lancaster, 19^2) (Biological Symposia, 7) 301-22. Reprintedin K.H. Pribram, Perception and Action (Harmondsworth,Penguin, 1969) (Brain Behaviour, 2) 575 PP*

Lettvin, J.V., Maturana, H.R., McCulloch, W.S., What the Frog's EyeTells the Frog's Brain (Proc. I.R.E., V7 (1959) 19^0-1951•

Meister, D., Comparative Analysis of Human Reliability Models.AD-731* 432 (1971) h&l pp,

Michael, Charles R., Retinal Processing of Visual Images.Scient. Am., 220 : No. 5 C1969). 1O5-112*.

Miller, George A., The Magical Number Seven Plus or Minus Two.Psychol. Rev. >G]>_ (1956) 81-97.

Miller, G.Å., Psychology and Information. Am. Docum. 19 (1968)286.-289. '

Miller, G.A., Galanter, E, and Pribram, K.H., Plans and theStructure of Behaviour. (Holt, Reinhart, andWinston, London, 1970) 226 pp.

Mowrer, O.H., Learning Theory and the Symbolic Processes.(Wiley, New York, i960) 473 pp.

Newell, Allen, Shaw, I.C., and Simon, Herbert A,., Elements ofa Theory of Human Problem-Solving. Psychol. Rev.65 (1958) I5I-I66.

Newman, Robert J., Extension of Human Capability through Informa-tion Processing and Display Systems. AD 6^5^35(1966) Ik pp.

Nielsen, Gerhard, Færdighed og indsigt. Nogle emner indenforindlæringens og problemløsningens psykologi-(Munksgaard, København 1964) 277 pp. (in Danish).

Petersen, O.M., An Analysis of Control Room Operators' Needs forInformation from Plant Instrumentation During Start-Upof a Power Plant Boiler. Risø-M-1738 (197*0 (to bepublished) (Danish edition, Risø-M-l693 (1973)).

Pribram, K.H., Some Dimensions of Remembering: Steps Toward aNeuropsychological Model of Memory. In: Macromoleculesand Behaviour. Edited by J. Gaito. (Appleton, New York,1966) 165-187 Excerpt in: K.H. Pribram, Perceptionand Action (Penguin, Harmonusworth, 1969)(Brain andBehaviour, 2) 575 PP-

Pribram, K.H., The Neurophysiology of Remembering. Sci.Amer. 220No. 1 (1969) 73-82.

Rasmussen, Jens, *0n the Communication between Operators and Instrumen-tation in Automatic Process Plants. Risø-M-686 (1968)Zk pp.

Rasmussen, Jens, Man-Machine Communication in the Light of AccidentRecords. International Symposium on Man-Machine SystemsCambridge, 8-12 September 1969« IEEE Conference RecordNo. 69 (58-MMS. Vol. 3).

Rasmussen, Jens and Goodstein, L.P., Experiment on Data Presentationto Process Operators in Diagnostic Tasks.In: Aspects of Research at Risø. A Øollection of Papersdedicated to Professor T. Bjerge on His SeventiethBirthday. Risø Report No. 256 (1972) 69-83

Rasmussen, Jens, and Jensen, Aage, A Study of Mental Procedures inElectronic Trouble Shooting. Risø-M-1582 (1973) 71 PP«

Rasmussen, Jens, The Role of the Man-Machine Interface in SystemsReliability. Paper presented at the NATO Conference onGeneric Techniques in Systems Reliability Assessment«Liverpool, 17-28 July 1973- Published as Risø-M-l673

. (1973) llpp.

Rigby, L.V., The"Nature of Human Error. SC-DC-69-2062 (1969) 12 pp.

- 51 -

Roberts, L.G., Machine perception of three-dimensional solids , Mass.Inst. Technol. Lincoln LaboratoryTech. Rep. 315, May 22, 1963- Reprinted in Optical andElectro-Optical Information Processing, Tippett, J.T.et al. (M.I.T. Press, Cambridge, Mass., 1965) 159-197-

Shannon, C.E., and Weawer, W., The Mathematical Theory of Communica-tion (Univ. of Illinois Press, Urbana, 111., 19^9)117 PP.

Simon, Herbert, A., The Sciences of the Artificial. (M.I.T. PressCambridge, Mass. 1969) (Karl Taylor Compton Lectures,1968) 123 pp.

Sloman, Aron, Interaction between Philosophy and Artificial Intelli-gence: The Role of Intuition and Non-Logical Reasoningin Intelligence. In: Proceesings of the InternationalArtificial Intelligence Conference (The British ComputerSociety, London, 1971) 270-278.

Swain, Alan D., Human Reliability•Assessment in Nuclear Reactor Plants.SC-R-69-l236(1969) 29 pp.

Venda, Valeriy Fedorovich, Informational Technology and Ergonomics.JPRS 51123 (197O) k7 pp.

Weiss, P., Central Versus Peripheral Factors in the Developmentof Coordination. In: Patterns of Organisation in theCentral Nervous System. Proceedings of the Association, DDecember 1 5 - 1 6 , 1950. Edited by P. Bard (Williamsand Wilkins, Baltimore, Md., 1952) 3-23* Reprint in:Perception and Action. Edited by K.H. Pribram(Penguin, Harmondsworth, 1969) (Brain and Behaviour, 2)575 pp.

(Wohl, Joseph G.) Man-Machine Relationship in Prelaunch Check-outof Advanced Space Vechicles.Dunlap and Associates 510-QR-5, SSD-65-236,1965•