Reliability and Integrity of Computer Assisted Decision Making Process

Copyright © (FAC SAFECOMP '85 Como, Italy, 1985

RELIABILITY AND INTEGRITY OF COMPUTER ASSISTED DECISION MAKING PROCESS M. Hashim BlIrrollghs Machines Ltd" CUIII/Jel'llauid, Scotland, L'K

Abstract, Computer Assisted Decision Making (CADM) is finding its use more and more both in engineering and service industries. In our daily life real intelligence is being rapidly replaced by artificial intelligence (AI). This explosion of AI demands a heavy emphasis on meeting the expectations and requirements of all four components of CADM process i.e. hardware, software, man and society. The crucial part played by software in transporting the "thinking, judging and deciding" mechanism from human brain to computer brain (memory or logic) cannot be over emphasised. The nature of this role is no longer a purely technical matter, it must be now seen in more realistic i.e. Socio-technical context due to unique features of CADM process. Designing into a software the reliability and integrity depends very much on learning the behaviour of software through model based on Socio-technical considerations e.g. complexity. One such model is presented in this paper. This work suggests that CADMP is a Socio-technical entity which is going to influence each and every aspect of our society and, therefore, should be dealt with accordingly. The error prediction model based on complexity can be used as a step forward towards this understanding of Al explosion. Keywords. Computer Assisted Decision, Reliability Models, Software Complexity, Socio-Technical Nature of AI, Human Machine.

I NTHODUCT I ON This paper discusses the important elements of CADMP in terms of its socio-technical behaviour with emphasis on software. It also proposes a reliability prediction model based on complexity particularly applicable to CADMP and similar software. The paper also highlights the unique features of computer assisted decision making process and compares it with the conventional (manual) decision making process.

The bewildering rapid advances of our computer age . have brought us to a cross road of physical and intellectual illusions where 'real intelligence' (human brain) is being replaced by 'artificial intelligence' and in future may see real people replaced by artificial people (robotics). We also see th e wizardly machines called computers not only assisting us in our decision making process in our daily life but often making actual decisions for us even if we don't like them and have no option to change them either. The noble, refined and joyful experience of listening, pondering, thinking, judging and then d e ciding upon t.o do or say something seems to have been taken over by concepts such as expert systems, Al etc.

COMPUTER ASSISTED DECISION MAKING PROCESS

Figure 1 shows a conventional decision making process without the use of computer where the know-how resides in the human brain. If this know-how (the body of knowledge as the experts call it) is transferred from the human brain to il computer memory, the process of thinking also ~ransfers from one location (brain) to another (memory). This sounds simple and we want to leave it to such simplicity for the purpose of our Imderstanding.

On e of the most important elements of Computer Assisted Decision Making Process (CAD~lP) is the set of those unseen commands and instructions to the machines that do the so called magic work and what we term as software. The software forms the core of all Artificial Intelligence (AI) activitY with unique requirements. Though substantial ~fforts are being spent on developing the heaps of hard",are and piles of software for CADMP, little att e ntion seems to have been paid to the fundamental requirement of integrity and reliability of softwar e particularly used in decision making process. The impact of integrity and reliability of software o n the society has not yet been fully appr p ciated.

Figure 2 shows the above mentioned location transfer of know-how where apparently nothing '? lse seems to have changed except that the human brain is now behaving like a robot. Unfortunately '",hat seems simple is not what it is. The rundamental concept behind AI is the reproduction of 'thinking' mechanism which in spite of being

21

22

M. Hashim

imperfect in may respects, is still able to Jperform at least as good as a human being in at least one isolated ar",,-of knowledge. Computer .~ssisted Decision Making (CADM) is a complex ]process of : i.

presentation of a body of knowledge and rules and guidelines to use it appropriately

ii.

the retrieval of this knowledge when called upon to do so with full reliability.

iii.

stepwise and logical scanning through the whole body of knowledge (information) and

iv.

most difficult of all to make a decision,

a.

In a conventional decision making process

(Figure 1) man is always on-line with the problem solving system and his environment. In CADMP man as a user is not directly involved to the high degree of thinking. b. CADMP can only house a limited amount of information (no matter how large computer storage is) from a body of knowledge which is very commonly available and used in practice but may not possess such rules and information that may be of critical importance under special circumstances or in the solution of boundary line cases. Decisions in these cases are almost always urgent

requiring high degree of integrity and accuracy give a verdict or provide a sensible and

authorised answer to the problem posed by the user. These decisions may be, for example, if a person

is suffering from a heart disease (medicine); why a large sum (say £50K) should be spent on the construction of a warehouse (engineering); is Mr XYZ a criminal? (law) and so on. NATURE AND ElEMENTS OF CADMP Computer assisted decision making is finding its use more and more both in engineering and service industries. Its' particular use is seen recently in the diagnosis of medical ailments from symptoms

If we look at the entire process of CADM, we find it very democratic. It starts with people writing the intelligent software for other people to use it, where the former utilise their 'thinking and intelligence' to spare the latter from using theil 'thinking and intelligence'. Hardware or the machines only play a mere mechanical

role in this whole process. That is why concepts like artificial intelligence and expert systems

cannot be regarded as purely technical in nature. When Aleksander (1) wrote a book on Artificial Intelligence, he named it 'The Human Machine' and very appropriately so.

The CADMP has four basic elements namely: Hardware, Software, Man (as user and as well as manufacturer)

as the balance lies in the fact that a small piece of information is present or not. Man in conventional process has a ready access to such information or is known to him, computer can't

think what it requires to fill the gap or what is missing at the first place. c. Rational behaviour, constraints, pr e judices, sympathy, considerations, change of mind etc. all play important parts in the overall process. These factors, however, cannot be programmed into a computerised decision process h e nce lacks the

integrity and reliability of decision to be of immense us e decision.

to those concerned and affect e d by th e

d. Before a man makes a final d e cision h e is at liberty to it e rate and r e -iterate his findings in view of social, economical and behavioural aspects going in favour or against his decision. This

gives him the fre e dom and opportunity to judge the overall impact of the decision. He is not reluctant to change if need be. During the proc e ss of thinking and deciding a man gathe rs more and more information, wisdom, scre e ning logic and balance between emotions and nee ds. These factors contribut e immensely to the integrity and reliability of d e cision. A 'colnputer brain' is, unfortunately unaware of thes e scenarios and cannot, therefor e , b e relied upon c ompletely with d e cisions made by it.

the impact of CADMP in the form of ripe fruit and

e. There is always a small sub-s e t of corr e ct decisions to a problem out of a univ e rsal solution set. Man is capable of and has access to this sub-set in entirity. Comput e r is bond e d and can reach only a part of th e ahov e mentioned sub-set. This limitation is on e of th e g reatest sources of unreliability and lack of int e grity of comput e r assiste d decision proc e ss.

is not directly involved in the actual process of computer assisted decision making. We must, however, remember that man as a manufacturer of

f. CADMP exhihits a l ac k of fr ee dom of thought thus r e ducing the int p grity o f th e proc e ss of

and his environment or society. The hardware has

already taken a giant step forward in the reliability field. Man as user is being asked to keep his real intelligence away from the process

so he is behaving just like robotics in the expert system or AI environment. The society is receiving

CADMP is very much involved. The fourth component, the software is the most crucial of all and results

dec ision making. Th e above featur e s Cl e arly indicate th e ne e d for each e leme nt (particularly

from the efforts and interaction of the other thre e

software bein g the major component of CAVMP) to bp

elements to the maximum, as it is established that

design e d to r e vok e th e se limitat]or~ &i"'u l int e ntional efforts to b e ma de to achi e v e the d e sired reliability and int e grity of th e whol e syst e m.

CADMP would become an impossible task without software explosion (2).

Without this th e du e ben e fits of CADMP cannot be appreciat e d.

The reliability and integrity of each of the above mentioned four components is not merely a technical concept or a known mathematical number. It is on the contrary a state of confidence and

trustworthiness of each component that adds up to give the overall integrity and reliability of CADMP. THE UNIQUE FEATURES OF COMPUTER ASSISTED DICISION PROCESS It seems appropriate to highlight some of the unique features of CADMP before discussing the integrity and relaiability of this concept. These features are summarised as follows

:

INTEGRITY AND RELIABILITY OF CAVMP Let us go back to the fundampntal question of defining th e terms 'integrity' and 'r e liability' when us e d in relation to c omput e r assist e d decision making process e s and its compon e nts.

Integri ty of

CAD~lP

Lit e rary me aning of word 'int e grity' is th e stat e of b e ing entire; wholen e ss; hon e sty an d uprightnesS

But I think it is mor e than that. In t e rms of CADN it means providing a decision whi c h is

'timely'

Reliabilit y a nd Integ-rit \' with respect to nature, type and r eq uirement of the problem. (Functioning timely in physical time fram e is called reliability/avai lability). Let us illustrat e this by a well known example. A six minute alert in a USA nuclear base following a fault in th e system software that almost took the world at the doorsteps of a third ..-orld war may b e absolutely r e liabl e from technical or operational vi e w but it certainly had no int eg rity and r espec t attached to it. It was indeed not an hon est manufc station of computer assisted decision making process. Int eg rity also means to r es pond quickly and strongl, to th e wrong input to the process thus avoiding unwa nt ed or und es ir e d decisions. The balk of th e r esponsibilit y to inculcate the int eg rity into the computer assisted decision proc ess falls upon the decision proc ess designers i.e. software expertise. ~iability

of CADHP

Reliability is often thought in technical terms, a numbe r as sociat ed to the system confidence o r trustworthin ess . I t should, hO"" l ~ \' (~ r, be looke d into broad e r fram e work of social and en vironmental scen ar i os too. Th e high mat hematical probability tha t a system will 'd ecide' rightly to stop functioning if th e re occurs a malfunction in one of th e SUb-systems and compo nnnts withouth causing a fault or a f ailu re is considered to be the reliability of the decision proc ess . This becomes a crucial parameter if such a decision has to be take n under a correct function and time fram e at the same time . A fault in the system hardware or software could be more deadly and fatal th a n input or data induced faults. This in turn throws the whol e burden on designing in th e reliability i n th e system components lik e software and hardwa r e . Th e total effectiveness of CADMP is therefore a function of socio-technical reliability component of each element. In very simple nota tion it can be ex presse d as follows : wh e r e Rp' rn' rs' r

, re are th e reliabilitip.s of the

m dec i s ion proc ess, hardwa r e, software, man ann environment. With the advances in technology the values of most of these paramet ers have already been achieved as high as possible. The software, however, has not yet achi e v ed th e r equi r ed r e liability. This is the component that shares most of the burden of unreliability a nd disrespect. One of the r eason s for this is th~ lack of s uit ab l p mod e l s that take into consideration the true nature of software development in r e lati o n to its socioteChni ca l behaviour particularly in the design phase. We shall now try to conc e ntrate o n this compon e nt.

TilE CADMP 'BUG'

: THE SOFTWARE

23

This allows the manag e ment to for esee the trend and thus to mobilis e th e various r esou rces to the best us e ac cor d in g ly. The above mode l has been applied to data obtained from many large modul ar software de velopments in telecommunication, medi cal diagnosis software and fr o m r ea l time programmes deve loped for the flexibl e manufactu ring systems (FMS). The r e sults on o ne of the softwar e modul es are shown in Figure 3 . CONCWSIONS Th e ove rall objectives of comput e r assisted decision makin g is to provide accuracy, efficiency spe ed and reli ab ility of decision. The e ffici e ncy in terms (.' f n."l.ncrseconds may be sure ly achieved through ha rdwar e but other parameters me ntion ed above and th e int egr ity cannot be ha mme red into as it depends on the software of CADMP that r e sid es inside th e machin e as a huma n brain with limit ed me mory. Th e profe s sional s in th e fi e ld hav e a h eavy r esponsibilit y nn th~ir s ho ulde rs to make th e so ftwar e as r e liabl e and respectful as possible within the human limits. The manufacturer us er a nd e nviron ment of CAm·tp is a ll sociotechnical and its impact in ca se of failure to provide a right decision ca n be e normous and ca tastroph ic. This component of CADMP is mos t critical and needs most attention. This pape r has suggested a compl ex ity based error prediction mode1 for the software during its various phases o f development (including ea rly phases) for the manage me nt and des igners a lik e .

REFERENCES I. Al e ksander , I (1977), The Human ~lachin e , Georg i a Publishing Company, USA. 2. Hayes , J .E. and Nichie, D. (J 983) , Int e lli gent Systoms, Thp Unprecedented Opportunity, Ellis Howard Ltd., U.K.

3.

Has him, H i 1 984) , 'A Soc io-techni ca l Model for the Pl a nning and Prl "diction o f Soft wa r e ', Nat iona l Conference on ~ality (NCQH-'84) Bombay, Ind ia .

FIG.l

Conventional Decision Making Process

Compl ex it y plays a very significant part in the r e li ab ility and int eg rit y o f a softwar"e specially if th e so ftwar e is e v e n tua lly to be us e d in compl ex and strateg ic decision making sittJations. Many works ex i st in th e li te rature that provide models of one type or th e othe r for software r eliabi lity but few take the social a nd t ~ch nic a l compl~xity loo k together. One such work for prediction of software r el iability is df'cr~bp.( ; "~n (3) and provides a suitable workin g mode l. It is summarised in APPENDIX A alon g with th e mod e l e quations a nd uses a compl ex it y factor ' C l in pr e dicting th e number of e rrors r e mai nin g at various stages of software life cyc l e and development phas es .

sccs- c

FIG.2

Computer Assisted Decision Making

24

M. Hashim f

Rc: N\Al""'1ING N'

t

FIG.) Comparison of Observed and Predicted result

APPENDIX A Symbols: Qo Nf Nr Nm NR Np C K L To f Z p P s m U Ed V

,Estimated initial faults ,Faults found to time t ,Faults removed in j ,Maintenance induced faults ,Total faults removed ,Faults remaining at t ,Complexity factor ,Residual factor ,Programme (S/W) size ,Mean time to failure ,Fix factor ,Complexity shadow ,Patching factor ,Programmability factor ,Team profile index ,S/W modularity ,Utility factor ,Error density .Severity factor j ,Maintenance number MTTR,Mean time to repair

Model Eguations U x Ed x MTTR

C

:;

P x s

x m

Z

JC /

f

~

Qo =

2

Qo

3

2P:-

P K

1

-------------

4

[Log((L x Nf)/(T o x'1) Nf x K

U-

1

5 6

V = j2

7

Nr

( Nf x f)

8

Nm

( Nr x V ) p

9

Np

Qo

Nr

Nm

10