Invited Paper - WIT Press...Stirling and Wilford's work on museums, art galleries and university buildings is remarkable for its exploration of new territory and the development of

Invited Paper

The quest for the parrot on the shoulder:

knowledge about emerging design solutions

and its representation in a CAD system

B.R. Lawson

School of Architectural Studies, University of

In this paper I shall consider the way knowledge about a solution to a designproblem emerges during the design process and the implications of this for itsrepresentation in any computer aided design system. In the context of thisconference I shall concentrate mainly on the computer as a graphical device,whilst recognising that increasingly many researchers are interested in thepotential it offers in applications which are only incidentally dependent ongraphics. Consequently I shall therefore also be concentrating on knowledgeabout design solutions rather than problems although an essential part of theargument concerns their close inter-relationship. However, I believe thisdistinction can still usefully be made when discussing CAD which Galle hasdivided into what he calls CAMP or computer-aided modelling and presentationas opposed to CAPS or computer-aided problem-solving [1].

I shall explore this matter through an examination of that area of design Iunderstand best which is architecture. However the issues raised here seem to meto be generic enough certainly to apply to other similar design fields, whichwould include industrial and product design, interior design, urban design and atleast some areas of structural engineering. These fields all share in common athree-dimensional end product which will be lived in or used directly by peoplewho are likely to be concerned about both its appearance and physicalperformance. I leave others to judge the extent to which these ideas are capable ofapplication in other design fields.

These three-dimensional design fields share another feature in common. They areall areas where by now we might have expected to see much more real progressin computer aided design than has actually taken place in the professionsthemselves. After early enthusiasm for computer aided design in architecture forexample, architectural practice has largely accepted the use of the computer inplace of the drafting machine but not yet in favour of the sketch pad. Why shouldthis be? There is no shortage of things computers can do which undeniably formimportant parts of the architectural design process. For example good reliablesoftware exists so that computers can visualise three-dimensional forms, they canmodel energy flow, they can calculate structural forces and design structuralmembers, they can model the effects of both artificial lighting and natural lightwhether in the form of sunlight or daylight, they can be used as cost-estimatingtools, they can offer expert advice on a whole range of technical and legislativematters, and they can even plan albeit in a limited way. In fact they can do mostof these things more reliably and in less time than most architects. Why thenwhen the three-dimensional appearance, energy consumption, structural stability,

Transactions on Information and Communications Technologies vol 5, © 1993 WIT Press, www.witpress.com, ISSN 1743-3517

422 Visualization and Intelligent Design

lighting, capital cost and planning of buildings are so critical do architects fail soconsistendy to take advantage of these tools.

I have recently been conducting a study [2] in which I have interviewed a numberof the world's leading architects about their design process, and I have found analmost unanimous view that we have yet to provide them with computer-aidedtools which they are prepared to use in the design studio. As an example let usexamine the views of just three; Michael Wilford, partner of James Stirling forover twenty years, the remarkable and highly acclaimed young Spanisharchitect/engineer Santiago Calatrava and Herman Hertzberger, one of the fatherfigures of modern Dutch architecture. These three are all acknowledged asamongst the great designers of our time, and yet they could not be more different.Stirling and Wilford's work on museums, art galleries and university buildings isremarkable for its exploration of new territory and the development of urban formin the public domain. Calatrava has designed some of the most astonishinglysophisticated bridges and railway stations of our era, and Hertzberger is wellknown for his structuralist philosophy and humanistic attitude towards the usersof his buildings. All three use computers as word-processors, and sometimes asdrafting tools. Calatrava of course also uses them for engineering calculations, sothere is no general antagonism to computers here. However none of these threegreat designers use or foresee using computers as design tools. Of course we canall say in these times of widespread criticism of the architectural profession thatarchitects are just stupid and incompetent, but I prefer the alternative and rathermore plausible theory that there may just be something wrong with the way weare designing CAD tools. In particular I am going to suggest to you that implicitin most such tools is a view of the design process which is based on several keyfallacies.

It has long been recognised that the design process involves a peculiarly difficultcombination of thinking skills [3]. The design of large systems such as buildingsinvolves finding as well as solving problems in a context which has becameknown as 'wicked problem solving' [4]. Design problems are also notoriously'knowledge rich' as opposed to 'knowledge lean'. By this is understood thatsuch problems require considerable amounts of knowledge beyond that which isstated in the problem description to arrive at a satisfactory solution. Suchknowledge comes in the form of expertise or the accumulated wisdom of thedesigner, and thus raises at least two important sets of issues for design research.These concern the way in which this expertise is acquired and the way in which itcan be retrieved and applied to new design problems. Cross [5] has pointed outthat knowledge itself is no guarantee of success but that 'a designerly way ofknowing' is critical here. Lawson [6] has also shown how students ofarchitecture may acquire knowledge about environmental science but remainunable successfully to apply this to design problems.

In fact there is now substantially more expertise in areas such as architecturalscience than is normally acquired by any one architect. For example in theBuilding Science Unit in my department at Sheffield University alone there havebeen numerous research students producing information and tools of use toarchitects, but their work fails to get into a format which makes it accessible todesigners as opposed to research workers. There is a considerable differencebetween an analytical tool suitable for use by an environmental engineer and thekind of tool needed by an architect who is in the early stages of design [7], Theanalytical tool thus tends to evaluate the proposed design in some way and if weare lucky it may check its performance against some declared criteria such asbuilding regulations. This then is the first fallacy implicit in so much software


Visualization and Intelligent Design 423

design for use by designers like architects. It is based on the mistaken view thatsuch design can proceed by a series of one dimensional analyses of successiveattempts to form a solution. In fact there are two problems here. Firstly suchsoftware tends on the whole to be evaluative and analytical rather than goaldirected or synthetically in its style, and secondly that each aspect of the design istreated in a discrete rather than integrated manner. Let us examine these twoproblems in turn.

The sequence of events in the design process is much less clear than wasoriginally thought in the early days of design research which tended to favour'analysis-synthesis-evaluation' type models in which it was assumed thatproblems must be analysed before solutions could be synthesised [8]. Bycontrast more recent empirical research has shown how designers tend to developsolutions in order to understand aspects of the problem, although this is stillneglected in the references of many who write about CAD. The author in a seriesof laboratory experiments showed that by their fifth year of educationarchitectural students had developed a cognitive style quite dissimilar to thatadopted by post-graduate science students [9]. Whilst the science studentspreferred to analyse a design-like problem before attempting synthesis, thearchitects were more likely to develop solutions long before they had understoodthe problem. We might say that whereas their scientific peers seem to be problem-focused, design students appear to be solution-focused in their approach. Sincewe have already seen that such design problems are not capable of completedescription, and since design is a prescriptive rather than descriptive activity thisprobably represents a reasonably sensible adaptation to the situation. Furtherevidence supporting this general cognitive strategy came from a study by Darkewho conducted a "series of interviews with award winning architects [10].She found that when designing housing these architects frequently usedsome major form generating idea well before they could be expected to knowwhat form would work best. This 'primary generator' acted as a sort ofhypothesis enabling her architects to explore the inadequacies of their yet ill-formed solution and through this to learn more about the problem. In a yetunpublished study of a group of internationally famous architects the author hasfound evidence which while redefining the idea of the 'primary generator'strongly supports this solution-focused approach.

What appears to be the case is that much designer behaviour is based upon thegeneral strategy of groping towards an understanding of both problem andsolution together. Heuristic procedures are thus more likely to assist this processthan purely analytical ones. For example, knowing with great precision just howmuch energy a proposed building will lose is much less help than knowingwhether a change of orientation or double glazing would reduce consumptionmore.

All of this suggests that considerable attention must be paid to makinginformation available in a form which is usable in a design process. Our failure tounderstand the significance of this has dogged the teaching of scientific materialin architectural education. Moreover the problem continues beyond educationinto practice where particular problems may only be encountered infrequentlythus causing designers at best to be unable to solve the problem unaided and atworst to forget to consider the matter at all. In fact this criticism can equally wellbe levelled at tools intended for designers to use which do not even requirecomputers. For example the author has shown that most schools of architecture inthis country have taught students to use daylight protractors, and yet those samestudents report virtually no use of this tool in practice [6].



This should not be taken to be a criticism of the idea behind the tool or itsimportance in design, quite the reverse. Others have shown that during thedesign process the techniques used are highly likely to influence the problemsfound and therefore solved [11] [12]. It is therefore critical that tools are designedwhich are indeed usable in the design process. To do this we must understandmore about the nature of the generative ideas used by designers during the earlystages of the design process and the language through which these ideas areexpressed.

A study of the primary generators used by designers in both Darke and Lawson'sstudies reveal that these ideas are almost always integrative in their nature. That isto say they do not deal exclusively or uniquely with a single dimension of theproblem. They do not therefore represent attempts at sub-optimisation. Suchdesigners do not habitually try to design a solution which optimises energyefficiency, or structural rationality, or planning organisation. Take for examplethe attitude of Richard Burton who developed the first ever RIB A policy onenergy conservation in buildings. Even this enthusiastic campaigner was toremark that "energy in building has had a fanfare lately and maybe it will have tocontinue for some time, but soon I hope the subject will take its correct placeamong the 20 other major issues a designer of buildings has to consider". Thisconcern to ensure balance in the design process is nicely echoed by Ian Ritchiewho describes technology as a 'beautiful parrot sitting on his shoulder' whoconstantly overlooks the process but occasionally gets out of control and 'jumpsdown and shits all over the paper before we've had a chance to think properly andthen you have to get hold of him and stick him back up there'. This image of theparrot sitting on the shoulder is I believe one which should drive softwaredevelopers in their search for the right kind of relationship between designer andcomputer, hence the tide of this paper.

Most of the descriptions used by these designers about their early generative ideasare couched in the language of solutions rather than problems. Often thesedescriptions also reveal a process of discovery which again seems solution ratherthan problem-focused. "He suddenly said 'it's a wall!', and I said 'yes it's notjust a lot of little houses, it's a great wall two hundred metres long and threestoreys high'. The problems which have already been discovered are then fittedinto this idea, "...but we'll make it a bit higher to get in the plant and servicecirculation and then we'll punch the residential elements through the wall as aseries of glazed bays which come through and stand on legs."

Thus we can see here a process which consists of ranging across a series ofproblem issues very quickly through focusing on and developing a singlesolution idea. Many of these designers talk of the need to work very rapidly andintensively at certain key stages in the process. One of the main reasons for thisseems to be the need to consider a wide range of problem issues, or as oneleading architect described it 'keep a lot of things in play'. A common image usedby designers to describe this is that of juggling. Michael Wilford, for exampletalks of 'a juggler who's got say six balls in the air...an architect is similarlyoperating on at least six fronts simultaneously and if you take your eye off one ofthem and drop it you're in trouble." The concentration required for such a complexjuggling act is so intensive that one cannot imagine it lasting for long, or asRichard MacCormac puts it, 'one couldn't juggle very slowly over a longperiod'.



So how can computer aided design tools fit into such a process. One of the keyfeatures of this kind of design lies in its multi-dimensional problems. Someaspects are relatively objective and numerical, while others are more subjective.However as we have seen tools already exist to aid the designer with many ofthese problems. However our difficulty is that we have seen how good designersat least need to work extremely quickly, and many tend to dismiss computer-aided design for this reason. The whole process is destroyed by having to breakoff and enter data into a computer before getting an answer. This correspondswith the experience of studio tutors in design schools. Students often need andwant advice very quickly. If they have to wait more than a few hours before thetutor can find time their thoughts have moved on and they may again be in thatrecurring phase of the process when they do not want to discuss it.

If we are trying to develop computer aids for such a process then it is simply nogood writing individual programs, which while they may be able to giveanswers to their own questions will not be used by designers. Many years agoJohn Lansdown pointed out that we could either have what he called 'ad-hoc'programs or 'integrated systems' [13]. By this he meant we either have a seriesof quite separate programs each with their own inputs and outputs or we have asingle computer description of the proposed design which a whole series ofroutines can operate upon. Clearly the 'ad-hoc' programs provide far too great adelay to be satisfactory for the design process as we observe it to be practised.

It was hoped that integrated CAD systems would offer the prospect of embracingat least all technical knowledge into a design tool, and the author has himselfdevoted many years to developing such a system for architects [14]. Howeverthis has proved something of a mirage I think for several reasons. Firstly theexpense of such systems and the skill required to operate them has so farprevented them from becoming commonly used, although architects now makeregular use of drafting systems. The second reason for the failure of suchsystems to make the predicted impact is of a more technical nature and concernsthe difficulties of enshrining knowledge which needs updating and challenginginto software which can only be updated by system suppliers. There is hope thattechniques such as object oriented programming and the use of declarativelanguages may alleviate some of these problems but little real progress has beenmade to date. Even were this progress to be made such integrated systems allhave a recognisable style with many built in preconceptions that designers findtoo inflexible in use. Creative and sensitive designers simply do not want to giveup so much control of their process to software developers. The system feels as ifit suffocates the designer. The tail wags the dog!

An alternative scenario might be for designers to be able to buy individualpackages which can draw their data from a common and agreed conventionaldescription of the design. The popularity of the Apple Macintosh amongst suchdesigners suggests this idea is appealing. What the Mac basically offers is acommon interface which is firstly relatively intuitive but also so disciplined that itforces software developers to construct programs with sufficiently similarinterfaces for the user to be able to operate programs from several supplierswithout suffering 'manual overload'. Unfortunately even though much work hasbeen done I think there is little hope of a. satisfactory common description in theforeseeable future. Indeed such an idea is positively dangerous in my view untilwe understand much more about what is required. Ill-considered and restrictivestandards can impose serious constraints on innovative development We alreadyhave enough obstacles posed by the early development of standards for micro-



computer operating systems and data exchange formats for two-dimensionalgraphics.

However this discussion is already in danger of becoming as unrealistic as aremany computer design aids, for it implies that a database of known facts about asolution can really exist during the design process. The problem before us is lessa matter of how evaluative routines could be fed from a common database, butrather how that database grows and interacts with the designer as he or shegradually comes to understand the solution, which until the final decisions aremade remains a nebulous and ever changing mirage rather than any concretereality. Personality studies of designers have shown them to be prepared to accepthigh levels of uncertainty and change and to be capable of looking at things inmany different ways. These then are also the characteristics we need from anycomputer system to which we might be describing our design solution.

By contrast the characteristics of the computer have lead software developers tocreate programs offering ever increasing levels of precision, accuracy andrealism. Unfortunately these programs require huge amounts of input and canreally only be justified once the design is more or less finalised. Perhaps the mostobvious example of this can be seen in the world of visualisation tools. We nowhave the most astonishing levels of realism available using sophisticatingrendering techniques of ray-tracing or radiosity. However these algorithms canproduce nothing without data governing not only the geometry of the design butalso surface colours and textures to say nothing of light sources andcharacteristics. All this is very nice when you want to convince the client thatyour design is just what he wants, but quite useless while you are still at thedoubting stage yourself. What designers need are tools which give a roughindication. There is no point in being precise about something which is as yetonly a vague idea.

This analysis leads me to suggest that the area in which we need to do more workis the way in which the designer describes this vague, imprecise, nebulous,shifting idea to the computer. I do not refer here however to the minutiae ofergonomics, to the detail of windows, icons, pull-down menus and pointingdevices. These are in my view only the superficialities of interface design. What Irefer to here is the way in which the computer must be able to infer and acceptambiguity, to live with incomplete descriptions and perhaps fill in the gaps, andmost importantly interpret the constantly changing language of the designer.

We must finally learn the lesson that if computer aids are to play a useful role indesign, as I believe they can, they must not require of their users to learn aforeign language. One simply cannot expect a creative co-operation to emergefrom a situation where the key participant is continually reducing to interruptingthe train of thought and argument by having to look up words in a dictionary! Ifwe are to develop suitable interfaces then we must listen to how designersdescribe their solutions. We should set as our goal the objective of a systemwhich can at least understand as much about a design as would another designer.The immediacy, subtlety and often vagueness of these conversations are not aneccentricity, but an essential component of the creative process. Many designers Ihave interviewed make reference to a phenomenon well known anecdotally, andbest illustrated in Richard MacCormac's words. "I cannot say anything until I'vegot a pencil or pen in my hand." This supports the delightful concept developedby Donald Schon who describes designers as 'having a conversation with thedrawing' [15]. Over the years I have developed such a relationship with my wordprocessor. I wrote this paper by 'conversing with my keyboard and screen'. I can



now type faster than I can write, and the liberating feeling I get from knowinghow easily I can edit somehow enables me to work in a less inhibited and moreproductive way. Even the slight pause required to print offers that few momentsto get up and walk around which somehow facilitates the creative process. I haveyet to find any computer aided design tool which even approaches this quality ofcreative interaction. We have hardly begun in this field.

Of course it is far easier to develop a word processing tool than a design aid. Wedo not expect our word processor to understand and evaluate the quality of thearguments in the text. That is just what we do expect from design tools, and wewould like this even before the argument is cogently put. This is a tall order, but Ibelieve we can already see some of the things that will be necessary if we are toachieve this, and I shall devote the remainder of this paper to illustrating thisthrough one example.

Let us then finally get down to asking ourselves what we want from tools whichwe might genuinely call computer design aids? Firstly let me say that I see almostno point at all in designers using computer graphics systems except for quitespecialised purposes. Most designers have had to learn to draw and sketch andhave a great facility with the pencil which is currently a far more flexible andsensitive tool than any computer drafting system I have ever seen. Donald Schonhas taught us to understand that design seems to progress through what he calls aconversation with the drawing. That is to say the designer sees the drawing andmakes a change and sees the result. Most usually the designer can and doescompare the result very directly with what went before. Sketches are made eitherby drawing over the original on the same piece of paper, laying another semi-transparent sheet over it or starting a new sheet. In all cases the original and newcan be compared. How many computer drawing systems allow this absolutelyessential but simple requirement without enormously tedious and complexinstructions? However computer drawing systems are often superbly efficient,accurate and reliable ways of constructing complex geometry and transforming itSantiago Calatrava tells us that "I like to sketch quickly by hand", but that "Iadmire very much the precision of the computer maybe if you want to draw an arcbetween two points and you want to see immediately, you could not do that byhand". [2] This again reminds us about this need for immediacy, only if thecomputer is more direct and immediate will it be preferred by designers.

If we are really to produce useful CAMP tools then we must be able to interact withthem in a very immediate and flexible way. However any observation ofarchitects working shows that they frequently load their design drawings withmeaning. These meanings are usually taken to be self explanatory andunderstandable by other designers. These can be seen at two levels. Firstly thekind of representation offered by the drawing and its relation to other drawings,and secondly the relationship between the drawing and the building it givesinformation about

At the first level we therefore expect a plan to be recognisable as a plan rather thana section, elevation or perspective. If one of my students starts drawing during adiscussion they will almost never say what kind of projection they are using andindeed bits of plans, sections and perspectives may well all appear on the samesheet and I am expected to understand this. At the second level my student willexpect me to understand all sorts of things about the building from these drawings.For example as a plan is drawn it would be reasonable and normal to expect thatthe viewer could distinguish indoors from outdoors.



This last point about implicit information can be further elaborated to show thatoften what must be inferred by the reader of drawings depends on the ability touse multiple modes of thought about buildings or other design objects. It hasbeen argued that architects organise knowledge about their emerging buildingdesigns into a number of quite distinct and structurally different forms. We maysee these as 'modes' of thought each having their own 'features'. For example abuilding may be seen as a collection of components such as walls, floors, roofs,windows, doors and so on. We may thus call this 'component mode', andidentify each building element as a feature of that system of knowledge. Howeverthe same building may also be seen as a collection of envelopes such as spaces,courtyards and external skins. What we might call 'envelope mode' thenrepresents a quite different system of knowledge about the building to that whichwe see in 'component mode'. Unfortunately the mapping between these and othermodes is rather messy, and the relationships vary from feature to feature. A wallwhich we would be inclined to think of as a single component may , along itslength separate several rooms, so we have a one to many mapping from componentto space. A door by comparison can reasonably always be considered to connecttwo spaces, or a single space with the outdoors.

We can go on to consider other modes such as 'system mode' where a building isthought of in terms of features such as structure, cladding, circulation, servicesand so on, or perhaps 'strata mode' where the features would include floorlevels, facades and sections. Of course, all these ways of organising knowledgeabout a building are mutually interdependent and architects in conversation abouttheir solutions frequently flip from one mode to another in quick succession.Regrettably however few computer-aided modelling systems for use by architectsrecognise this and require a form of communication in only one mode and that ofthe software designer's devising. There are two reasons why this poses a seriousobstacle to the successful development of such systems. One is to do withdrawing or input and the other to do with evaluation or output.

Observation of an architect sketching will show that as the 'conversation with thedrawing' takes place editing is not done all in one mode. For example a sketchmay show some rooms in plan draw in envelope mode which is then alteredthrough a wall being added in component mode which might well alter severalrooms.

However it is also the case that evaluative procedures may require data indifferent modes. Simple thermal evaluations need to be able to distinguishindoors from outdoors and thus need envelope data. Structural calculations needto be able to recognise those components which are part of the structural loadbearing system, dayiighting assessments are performed on rooms, routines tocheck for satisfactory means of escape in case of fire need envelope and stratumbased data and so on.

It therefore seems necessary to envisage systems which are able to makeinferences based on input in one mode about the implications to data representedin another mode. We have already built one prototype system at Sheffield whichallows the user to draw and edit either components like walls and windows orenvelopes like external skins and rooms. The System known as DAILYautomatically keeps both modal descriptions of the building up to date [16]. Thereare however two serious weaknesses in this system. Firstly that the knowledgeneeded to make the inference remains in the program rather than in a dataknowledge base, and secondly that the user must still decide before drawingwhich mode he is using.



This however seems to me to be the area most in need of work if we are really tomake progress towards the 'parrot sitting on the shoulder'. Better and fasterroutines for rendering or more sophisticated algorithms for assessing energy flowwill make no difference to the already disappointingly low take up of computer-aided design in architecture and its allied design fields. We need intelligentmodelling modules which can automatically produce multiple modal descriptionsof designed objects before we can sensibly feed design appraisal tools. Then andonly then will we know whether such a 'parrot' is sufficiently tame and knowsenough words to be useful in the design studio.

References

l.Galle, P. 'Computer methods in architectural problem solving: critique andproposals' The Journal of Architectural Planning and Research Vol.6, pp.34-54,19892.Lawson, B.R. Design in Mind Butterworth's Architecture, London, in press.3.Lawson, B.R. How Designers Think Butterworth Architecture, London,1990.4.Rittel, H.W.J. & Webber, M.M. 'Dilemmas in a general theory of planning'Policy Sciences Vol.4, 19735.Cross, N. 'Designerly ways of knowing' Design Studies Vol.3, pp.221-227,19826.Lawson, B.R. 'Science, legislation and architecture' in Changing Design ed.Evans, B.N., Powell, J.A. & Talbot, R.J. John Wiley, New York, 19827.Brown, A.G.P., Wood, D.J. & Brown, G.Z. A sunlight design toolECAADE, Aarhus Denmark, 1989.8.Markus, T.A. 'A doughnut model of the environment and its design' inDesign Participation ed. Cross, N. Academy Editions, London, 19729.Lawson, B.R. 'Cognitive strategies in architectural design' in Developmentsin Design Methodology ed. Cross, N. pp.209-220 John Wiley, Chicester, 1984lO.Darke, J. The primary generator and the design process' in New Directionsin Environmental Design Research: procedings ofEDRA 9 ed. Rogers, W.E. &Ittleson, W.H. pp.325-337 EDRA, Washington, 197811.Eastman, C.M. 'On the analysis of the intuitive design process' in EmergingMethods in Environmental Design and Planning ed. Moore, G.T. MIT Press,Cambridge Mass, 197012.Rowe, P.G. Design Thinking MIT Press, Cambridge Mass, 1987.13.Lansdown, J. 'Computer-aided building design: the next steps' RIB A JournalVol.76, pp.138-140, 1969M.Lawson, B.R. Gable: an integrated approach to interactive graphicaltechniques for modelling buildings On-Line Publications, London, 1981.15.Schon, D.A. The Reflective Practitioner: How professionals think in actionTemple Smith, London*, 1983.16.Lawson, B.R. & Roberts, S. 'Modes and features: the organization of data inCAD supporting the early phases of design' Design Studies Vol.12, pp. 102-108,1991


Documents

Invited Paper - WIT Press...Stirling and Wilford's work on museums, art galleries and university buildings is remarkable for its exploration of new territory and the development of