Download - Filter mediated design: generating coherence in collaborative design

Information growth in recent decades has drastically altered the struc-ture and location of the knowledge one uses to design. Individuals

increasingly specialize, relegating entire domains of knowledge toother specialists, computational agents and reference materials. The field

of collaborative design has evolved to make these different domains andagendas explicit, and to bring them back into dialogue. The goal is to

increase the capability to bring about innovation and cohesion from the

field of possibilities and constraints by communicating concerns and possi-bilities across disciplines.

Different domains have evolved their own languages and representations,making communication between disciplines difficult. The challenge is to

develop new methods for negotiating across different domains. This paperconsiders new approaches to collaborative design and describes opport-unities for new means of generating coherence and innovation by reformul-

ating the construction and flow of information. The work proposes some

www.elsevier.com/locate/destud0142-694X/00 $ - see front matterDesign Studies21 (2000) 205–220PII: S0142-694X(99)00042-3 205 2000 Elsevier Science Ltd All rights reserved Printed in Great Britain

Filter mediated design: generatingcoherence in collaborative design

John Haymaker, Paul Keel, Edith Ackermann and William Porter,Massachusetts Institute of Technology, Department of Architecture,Design Technology, Room 10-462M 77, Massachusetts Avenue,Cambridge MA 02139, USA

Architectural design involves the integration of diverse, sometimesconflicting, concepts and requirements into a coherent singlecomposition. This paper proposes a method for negotiating architecturaldesign across domains, by examining issues of perception, generationand evaluation, and detailing a prototype in which these mechanismsare augmented using computational agents for achieving coherence andinnovation in remote collaborative design. Filter Mediated Design isintended to explore the processes and strategies of constructingintelligent designs and design intelligence. 2000 Elsevier Science Ltd.All rights reserved

Keywords: collaborative design, teamwork, architectural design,collective intelligence

1 Simon, H A The sciences ofthe artificial MIT Press, Cam-bridge, MA (1981)2 Cross, N ‘Natural intelligencein design’ Design Studies Vol 20No 1 (1999) pp 25–39

fundamental processes at play in design—perception, generation and evalu-ation—and details a prototype in which these mechanisms are augmentedusing agents for achieving coherence and innovation in remote collabor-ative design. The paper proposes a common geometric and topologicaldatabase, from which multiple semantic models are constructed. ‘Filtermediated design’ is intended as a tool for design, and for thinking aboutdesign.

1 The changing nature of collaborative designIn recent decades, the designer’s work environment has changed. Infor-mation growth and advancements in information technology have transfor-med the location, structure, and flow of the knowledge one uses to design.A number of phenomena mark this transformation:

I Specialization—humans confine the scope of their knowledge in orderto go in depth on a particular subject.

I Delegation—humans increasingly rely on other specialists, compu-tational agents, and reference materials to augment their cognitive abili-ties1.

I Cooperation—an increasing number of people become involved in anyone project, often working in parallel.

I Distribution—increasingly people are working remotely.

Design, as a process of generating coherence and innovation, must operatein this altered information landscape. Some of the things designers do inthe process of generating coherence and innovation, according to NigelCross2, involve search for emergence, imposition of additional constraints,use of solution conjectures, and change of solution goals. New mediatingtechniques which support these design processes are needed to bring thedifferent disciplines back into dialogue and allow cohesion to emerge inthe changing information landscape. Negotiations, which traditionallycould occur internally (and often implicitly) in the designer’s mind areprojected outward, and thus are made explicit. Traditional collaborativedesign relies on verbal communication and artifacts representing aspectsof the design to share and negotiate this knowledge. However, the forcesof expansion, specialization and distribution of knowledge often make tra-ditional modes of collaboration difficult. Some of the issues traditionalcollaborative design faces in a world of increasing specialization and del-egation are:

I Communication—spatial and temporal constraints of verbal communi-cation inhibit the transfer of knowledge.

I Representation and meaning—different disciplines evolve differentrepresentations and languages for their domains.

206 Design Studies Vol 21 No 2 March 2000

3 Eastman, C A and Fereshe-tian, N ‘Information models foruse in product design’ Com-puter-Aided Design Vol 26 No 7(1994) pp 551–5724 Rosenman, M A and Gero,JS ‘Modelling multiple views ofdesign objects in a collaborativeCAD environment’ CAD, SpecialIssue on AI in Design Vol 28 No3 (1996) pp 207–2215 MacKellar, B and Peckam, JMultiple perspectives of designobjects. In Gero, J andSudweeks F (eds) Artificial intel-ligence in design ’98 Klewer Aca-demic Publishers (1998) pp 87–1066 Pahng, F, Bae, S and Wal-lace, D A web-based collabor-ative design modeling environ-ment (1998) http://cadlab..mit.edu/publications/98-wetice-wallace7 Eastman, C, Jeng, T S andChowdbury, R ‘Integration ofdesign applications with buildingmodels’ in Junge, R (ed) CAADfutures ’97 Kluwer AcademicPublishers (1997) pp 45–598 Campbell, M, Cagan, J andKotovsky, K ‘A-design: theoryand implementation of an adapt-ive agent-based method of con-ceptual design’ in Gero, J andSudweeks, F (eds) Artificialintelligence in design ’98 KluwerAcademic Publishers (1998) pp579–5989 Kolarevic, Schmidt, Hirsch-berg, Kurman, and JohnsonAn experiment in design collab-oration, Acadia ’98 Associationfor Computer-Aided Design inArchitecture (1998) pp 91–98

I Memory and information processing—individuals and collaboratorsstruggle with large, combinatorial tasks.

I Coordination—distribution of intelligence generates extensive diffi-culties in the task of coordination.

One of the major issues in developing tools for remote collaborative designis how to structure, access, manipulate and coordinate data. Variousresearchers try to resolve these issues by constructing definitions of entitieswith attributes and by explicitly representing the relationships betweenentities3. Relationships between elements are often expressed through con-straints contained within the object definition. Other work addresses theneed for customized representations or ‘views’ of the data by labeling enti-ties based on purpose (i.e. ‘window’) or function (i.e. ‘provide—view’)4,5.The disciplines interested in these labels see the objects. Another area ofresearch is on the benefit of multiple versus centralized data models withfocus on the maintenance of consistency between such models, detailingupdate rules that propagate changes between views6,7.

Filter mediated design differs from similar research by focusing on theearly stages of design when ambiguity serves as a powerful heuristic forcreative design. This work explores a database design with as little seman-tic content as possible to allow for multiple readings. Semantic data isactively constructed from geometric and topological data rather than beingread from labels on database entities. ‘Filter mediated design’ proposes acommon geometric and topological database, with multiple dynamicallyreconstructed semantic models. To maintain consistency, this research pro-poses the imposition of external constraints to allow for a more flexiblestructuring of the data. In design, it is often desirable to dynamically relax,remove and impose constraints. Proposed changes to the common databaseby individual users are first negotiated between the semantic models.

This research is also concerned about design interaction, the protocols ofcooperation, and conflict management. Mathew Campbell, Jonathon Cagan,and Kenneth Kotovsky’s8 research involves multi-agent design processes,in which specialized agents focus on domain specific aspects of the designto construct a collective intelligence. The design world consists of compo-nents with inputs and outputs that are connected, sized and evaluated. ‘Fil-ter mediated design’ expands on the idea of multi-agent distributed design,considering schematic design in the architectural domain, where issues ofmultiple readings and criteria are negotiated. Urs Hirschberg et al’s9 workon PhaseX and Fake Space fosters collaborative design activity by blurringauthorship throughout the design process by encouraging users to exchangetheir models iteratively. Filter mediated design seeks to further blur author-ship by immersing users in a distributed virtual environment where users

Filter mediated design 207

contribute simultaneously and explores a design architecture for user/user,user/agent, and agent/agent interaction.

2 An opportunity for new means of constructingcoherence and innovationThis paper views the distribution of knowledge and the mediation of con-straints and possibilities through digital technology as an opportunity, notto facilitate traditional means of collaboration, but to construct new meansby which cohesion and innovation can be achieved in collaborative ende-avors far more complex than those of the past. Figure 1 shows a schematicdiagram of traditional collaborative design compared to ‘filter mediateddesign’. In traditional collaborative design, communication of possibilitiesand constraints occurs through conversations with others, with represen-tations of the project, and with others through representations of the project.Coherence emerges in our formulation when multiple points of view findsatisfaction in the artifact. ‘Filter mediated design’ proposes implementing‘filters’, which serve to reformulate the construction and flow of infor-mation in collaborative design. Users view and modify objects throughinterconnected computational devices that automatically exchange and dis-play information about user contributions. The intent is to supplant someof the verbal communication to improve efficiency and allow for morepeople to participate in the design process.

When knowledge becomes distributed, the challenge becomes how to sup-port and develop the ability to integrate diverse perspectives in new andinteresting ways and to begin exploring new means of supporting and gen-erating design intelligence. This paper addresses this challenge through amulti-disciplinary exploration of the design process, considering a number


Figure 1 (a) (left-hand side): traditional collaborative design, (b) (right-hand side): filter mediated design

of issues from the domains of Design Inquiry, Developmental Psychology,Artificial Intelligence, and Communications Technology. Several mech-anisms are suggested for an individual to construct design coherence inthe world: Perception, Generation, and Evaluation. The paper discusseshow these mechanisms can then be employed to construct coherence andinnovation across individuals in the field of design. This approach doesnot privilege particular design strategies; rather, it proposes a mediatingmechanism that can support various strategies. This reconstruction directlyaddresses the issues of communication, representation, memory and infor-mation processing, and coordination, with which traditional collaborativedesign struggles.

3 Making explicit the mechanisms of designCollaborative Design is a Constructive Dialogic process. This means thatcoherence evolves through multiple cycles of concept generation and test-ing in ‘feedback loops’. The feedback allows for the generation and testprocesses to be reconsidered in light of the other. In an attempt to modelthis process, this project proposes a dialogue between three mechanisms:

I Filters—the ability to sense attributes and construct readings in theworld. ‘Filters’ can also communicate their readings to other ‘filters’.

I Constructors—the ability to judge, and the ability to act.I Managers—the ability to coordinate ‘constructors’ and ‘filters’.

Figure 2 illustrates an overview of ‘filter mediated design’. This conceptualoverview proposes the integration of Users, computational ‘constructor’agents and ‘manager’ agents in a collectively intelligent module, mediatedby ‘filter’ agents. An overview is provided here to provide the reader witha sense of the issues addressed in this paper and explain the individualmechanisms in more detail below. Coherence in design, it is postulated,emerges from the dialogue between the mechanisms proposed.

Design is a dialogue, through the object, with the contexts for which theobject is intended. During this process the designer adopts many perspec-tives. Each perspective focuses on different aspects of the project, andrequires specialized representations suited for the task. When testing poss-ible solutions, the designer simulates different aspects of the environment,imagining how the project will respond to issues like wind, light, andhuman use; when generating possibilities, the designer manipulates differ-ent aspects of the object, trying possible solutions. Coherence emergesfrom these transactions when the object satisfies multiple points of view.These ideas are explored from the perspective that generation of coherenceand innovation within an individual could well be analogous to the gener-ation of innovation and coherence across individuals. In a collaborative


exercise, the individual designers distribute and specialize on the differentperspectives of the project, but the fundamental dynamic of cycles between

generation and evaluation is fundamentally unchanged, although the

negotiation between perspectives is made more explicit.

3.1 Filters: addressing the issue of representation,meaning and communication‘Filters’ are computational agents that provide customized views of the

design world under consideration, and support non-verbal communicationby relaying information to other ‘filters’ in the system. ‘Filters’ are

expected to cultivate the range of design solutions by allowing individual

views and modifications of the design world, while automatically managing

the flow of ideas to and from a shared solution space.

Specific disciplines have evolved special means of representing, manipulat-

ing, and communicating about their specific concerns. That is to say they

have developed their own language. For example, the word ‘window’ foran architect carries different connotations and issues from those of the

mechanical or structural engineer. Because of the different worlds thatthese three disciplines deal with, they perceive different realities when

experiencing the same artifact. Figure 3 illustrates that architects might

associate the word ‘window’, with issues of view and light, the mechanical


Figure 2 Filter mediated

design—an overview

10 Lettvin, J Y ‘What the frogseye tells the frogs brain’ inCorning, W C and Balaban M(eds) The mind: biologicalapproaches to its functions Inter-science Publishers, New York,(1968) pp 233–25811 Porter, W L ‘Notes on theinner logic of designing: twothought experiments’ DesignStudies Vol 9 No 3 (1988)pp 169–180

engineer is primarily concerned with opening size, and orientation, whilethe structural engineer is concerned with span and loading.

In nature, filtering has evolved to highlight relevant information. JerryLettvin10 analyzed the signals which a frog’s eye sends to the frog’s brain.There are four separate operations in the frog’s eye:

I Sustained contrast detectionI Net convexity detectionI Moving edge detectionI Net dimming detection

The frog’s perception of the world is highly constrained. Before the brainever receives information to process, the eye has produced informationrelevant to the frog’s needs. From the available information the frog candetermine the location of flies and the approach of large intruders. Thereis a whole array of other signals in the world that the frog simply ignores,but that other species have evolved to detect. Human designers also benefitfrom the mechanism of ‘filters’. ‘Filters’ abstract and reconstruct the worldinto features that are relevant to the organism’s success.

In design the need to create selective representations of a greater com-plexity is also apparent. William Porter11 suggests that at least three classesof description are necessary for the designer:

I Object description—elements of the design and their locationI Formal description—edge or boundaryI Semantic description—ideas that reside outside

The filtering mechanism goes a long way towards problem framing. Omit-ting the importance of different representations for different tasks privi-


Figure 3 Different representations and meaning for the same objects

leges one representation over others and greatly diminishes problem solv-ing ability. Designers invest a lot of time producing differentrepresentations to solve different tasks. The important role of perceptionin an intelligent system is the issue. A paradigm often found in computationhas been primarily one of direct encoding. A window is represented in thecomputer as a composite object with attributes and relations. This paradigmignores two aspects of visual perception:

I Perception selects, transforms, and reconstructs.I Perception is informed and directed by ‘concepts’—one sees what one

knows to look for.

A representation is needed which will allow the flexibility of perceptionto rapidly construct new readings. One way to satisfy this need is to tryto anticipate the possible readings and provide a database full of constraintsand relations that can address these readings. This work explores anothersolution where these relations might be imposed by perception, not in thestructure of the database.

‘Filter mediated design’ takes the position that geometric and topologicaldata is the information available in the world, before filtering occurs. Anysemantic information is generated in the process of filtering. This paradigmallows for multiple, overlapping and conflicting meaning structures to beread into the same design world. In the database, a window is describedonly in terms of the geometric and relational properties of its materials. Itbecomes a window only when a particular ‘filter’ views it as such. Figure4 shows how a database of geometrical and topological data can be subjectto multiple readings. The first image is read as nothing, The second imagesatisfies the “Chair” ‘filter’, the third image satisfies the “Window” ‘filter’,and the fourth image satisfies both the “Window” and the “Chair” ‘filter’.

Human perception is capable of rapidly switching focus. The ability toreframe a problem in a number of ways is a key to building coherence.


Figure 4 Filters determine the semantic content of the world

12 Stiny, G ‘Introduction toshape and shape grammars’Environment and Planning B Vol7 (1980) pp 343–35113 Goodman, N Ways ofworldmaking ch 1 Hackett PubCo, Indianapolis (c1978)14 Piaget, J The origins ofintelligence in children Inter-national Universities Press, NewYork (c1952) pp 352–35415 Minsky, M Society of mindSimon and Shuster, New York(1985) p 72

George Stiny’s12 work on Shape Grammars represents a breakthrough fromthe computational paradigm of representing the world as finite components,asserting that designers’ visual perception has the capacity to constructmultiple readings from the same world. The ability to adopt and switchbetween different ways of looking at things is a powerful heuristic in thecreative process. George Stiny’s work states that the world is not an absol-ute from which one cuts out pieces, but that one constantly engages inreconstructing the world for different purposes, and that computationaltools should, and can, work the same way (see note 1).

Our concept of ‘Filter’ contains not just the ability to select but also theability to transform. ‘Filters’ are the changing interfaces in constant negoti-ation between the environment and the organism that the ‘filter’ is serving(see note 2). Marvin Minsky concurs with the importance of ‘filters’ in theconstruction of intelligent agencies. ‘Each type of knowledge needs someform of representation and a body of skills to use that style of represen-tation. Once that investment has been made, it is relatively easy for aspecialist to accumulate further knowledge, provided the additional expert-ise is uniform enough to suit the same style of representation’15. In designa ‘filter’ serves to select, focus attention on, and reconstruct informationin the project.

‘Filters’ also can be used to address the issues of communication in collab-orative design. When the number of participants in a design task increasesdue to increased distribution of knowledge, when the sharing of knowledgebecomes increasingly mediated by communication technology, and whenthe methods of representation between disciplines become too highly speci-alized, traditional collaborative methods break down. Verbal communi-cation, as a spatially and temporally constrained medium of exchange, canno longer serve as the preferred medium to connect this vastly distributednetwork because only limited numbers of people can talk at once. Inaddition, the mediation of face to face communication tends to diminishspontaneous interaction, as phone messages and e-mails disrupt the flowand opportunistic nature of discussion and discovery. In ‘filter mediateddesign’, much of the communication that is traditionally expressed verballycould be transmitted through ‘filters’. Also, by means of the ‘filters’, someof what is traditionally expressed silently, through the experience of theobject, can be focused and sharpened. Communication through ‘filters’could alleviate some of difficulties which different languages evolved indifferent disciplines place on understanding and generation of coherence.

Detecting both conflict and agreement is critical to the success of collabor-ative design. Figure 5 shows one application of using a ‘filter’ for com-


16 Kahneman, D and Tver-sky, A ‘Judgment under uncer-tainty: heuristics and biases’Science Vol 185 (1974)pp 1124–1131

munication: that of displaying agreement. The ‘filter’s’ interface could besimilar to an overlay. In this scenario, the ‘Filter’ has determined throughthe actions on these elements of other ‘constructors’ and users where agree-ment is being generated. Many users have deleted the overhang, proposedby one of the ‘constructors’, so the overhang is fading out, and its decliningvalue could be signaled with a black outline. As more users have acceptedthe parapet, it becomes more solid, and a white outline denotes gainingacceptance. Other filtered views could communicate specific concerns likestructural instability. A structural engineer’s display could express an archi-tect’s desire to have the view unobstructed, warning the engineer whenevera move violates this principle, or an individual designer could alternatebetween different ways of considering the design under consideration. The‘filter’ will consult with other ‘filters’ to analyze all user modificationsand negotiate what is being displayed on the individual’s view, such ashighlighting the most important changes under consideration or abstractingextraneous information to the task at hand.

3.2 Constructor—addressing the issue of memory andinformation processingOur research is concerned with how to merge human users and computersinto a collective intelligence. Many combinatorial strategies place unbear-able strains on people when too many constraints must be tracked. Inaddition, Humans often fail to determine clever strategies, due to immenseor unperceived solution spaces. Finally, while the human perceptual systemcan switch between representations with remarkable flexibility, our abilityto consider more than one representation at any time is constrained16. In


Figure 5 A ‘filtered’ view,

as seen by a user

17 Ackermann, E ‘Circularreactions and sensori-motorintelligence: when Piaget’stheory meets cognitive models’Archives de Psychologie Vol 38(1990) pp 65–78

a world of specialization and distribution of knowledge, the role of compu-tational agents can become greater in addressing some of the issuesdescribed above. ‘Constructors’ become external memory and informationprocessing aids when the knowledge becomes too extensive and distributedfor the unaided mind to process.

In our conceptualization of ‘filter mediated design’, ‘constructors’ are anal-ogous to Marvin Minsky’s agents15, and ‘filters’ are the lenses throughwhich they perceive the world. ‘Filters’ and ‘constructors’ are in constantdialogue. If a given filter/‘constructor’ pair is not performing well twosolutions are possible. One can either alter the ‘filter’, or alter the ‘construc-tor’. In our formulation, the ‘constructor’ evaluates a Constructed Viewprovided by the ‘filter’ and proposes a modification to the world. Just asthere is a recursive, interdependent relationship between ‘constructor’ and‘filter’, there is also an interdependent relationship contained within the‘constructor’, between the evaluative and generative functions. Evaluativeand generative strategies evolve in light of each other (Figure 6).

A great deal of the richness in design arises from just this interdependence.Design is a creative and open-ended process. Generative and evaluativestrategies are in constant dialogue. This tension in turn informs the wayone ‘filters’ the world, and preferred ‘filters’ settle in over time. As Piagetstates ‘behavior becomes intelligent as pathways between the subject andthe environment cease to be simple, rigid and unidirectional, but becomemore reversible and mobile, as the scope of the interaction goes beyondimmediate and momentary contacts to achieve far reaching and stablerelations’17.

Each ‘constructor’ requires a specialized representation of the world, pro-vided by a customized ‘filter’. From this representation, ‘constructors’ cantest for satisfaction, and generate to improve satisfaction. Figure 7 showsschematic representations for four ‘constructors’. The upper left shows a‘light constructor’, whose algorithm analyzes the amount of light cast on


Figure 6 Interdependent,

evolving relationship

between ‘filter’ and ‘con-

structor’

the floor from any south facing windows. The upper right is for the ‘space

constructor’, which is interested in determining the amount of usable floor

area. The lower left helps the ‘structural constructor’ assure that any span-

ning planes are appropriately supported. The bottom right image is used

by the ‘weather constructor’ to confirm that a roof covers all floor area.

Figure 8 shows a schematic scenario of how coherence might evolve from

specialized ‘constructors’. Interactions of ‘space’, ‘economy’, ‘light’,

‘structure’, and ‘enclosure’ ‘constructors’ collaborate in 12 steps to con-

struct a somewhat coherent structure, addressing the concerns of each. First

an empty site is presented to all ‘constructors’ (1). Then, the ‘space con-structor’ provides rooms to its liking (2), after which the ‘enclosure con-

structor’ covers the spaces (3). The ‘structure constructor’ recognizes

unsupported roof area, and provides support (roof not shown—4). The

‘light constructor’ eliminates the North, East, and West walls to allow for

more light (5), and the ‘economy constructor’ eliminates any roof not

covering its conception of space (#6). The ‘structure constructor’ shores

up the roof (7), while the ‘space constructor’ tries to reclaim space and to

assure that all rooms are connected (8). The ‘enclosure constructor’ covers

the new space (9) which causes a lighting conflict. Because of this, the


Figure 7 Filtered views for ‘constructors’

‘light constructor’ raises the roof (10), and the ‘structure constructor’ sup-ports the raised roof (11). Finally the ‘enclosure constructor’ fills in exteriorwalls with glass (#12).

Finding ways to integrate human and computational designers is a majorfocus of the work. The ‘filter’ mechanism provides a useful interface toallow multiple design strategies and representations to coexist cohesively.At times, users may want to use their ability to appreciate, allowing compu-tational agents to generate possible solutions. At other times, the users maywant to freely generate, allowing the computational agents to keep trackof the various constraints on the project and advise users of particulartransgressions. The application proposed should ultimately allow users to‘switch gears’ between different dialogic modes depending on how muchcontrol they wish to keep or delegate.

3.3 Manager—addressing the issue of coordinationIn ‘filter mediated design’, ‘managers’ are behind the scene agents thatfacilitate the construction of coherence out of a plurality of views, by


Figure 8 Emergent coherence from ‘constructors’

organizing the communication flow. Figure 8 suggests that coherence couldbe generated with a minimum of hierarchical control. When a particular‘constructor’ becomes notably interested, it produces a proposal. Other‘constructors’ are then left to consider this production in terms of their ownevaluative function. Marvin Minsky proposes the need for a heterarchy, ordynamically changing hierarchy, in the construction of intelligence15. Her-bert Simon states that this administrative power can come at a cost notincurred in decentralized systems, and concludes that different kinds oforganization suit different tasks1. Benefits and costs are inherent in anykind of organizational system.

When information becomes specialized and increasingly distributed, whatorganizational strategies best serve our desire to construct coherence andinnovation? The ‘managers’ are intended to explore this idea. Thereforeany system constructed should be capable of implementing both hier-archical and decentralized structures and, most importantly, hybrids thatcontain both. Which aspects of coordination are best left to humans andwhich aspects could be handled through computation is a subject ‘filtermediated design’ is intended to explore. ‘Managers’ can be equipped with‘filters’, so that they can alter their perception of the world as well. Whatthey are interested in observing, however, is less characteristics of theobject and more characteristics of the dialog between ‘constructors’, Users,and other ‘managers’. What they evaluate and manipulate is not the objectbut the communication flow.

A ‘manager’ could, for example, observe the activities of ‘filters’ and groupthose that successfully match proposed modifications multiple times. The‘manager’ would then ensure that the selected ‘filters’ would then proposetheir modifications primarily to their group of ‘filters’ and secondarily,after agreement has been determined within the group, to other groups of‘filters’. This procedure would allow users and ‘constructors’ with commoninterests to focus on a specific issue for a certain period of time withoutinterference from those with other interests. ‘Managers’ might also noticedynamics of system communication and try to divert the flow when pro-gress is not being made.

4 Filter mediated design is a tool for design, and forthinking about design

4.1 As a tool for designThis work is interested in improving the communication required to gener-ate innovation and coherence in design. When multiple participants andcomputational agents come together to evolve a design, the exchange of


meaningful information becomes the primary concern. Generating coher-ence and innovation is the goal which, when left to an individual relyingon verbal communication to understand the concerns of others, can bedifficult. This work proposes that communication and generation of coher-ence could be mediated with ‘filters’.

Among the things ‘filter mediated design’ will allow Users to:

I Adjust the ‘filters’ of various ‘constructors’, obliging them to focus onalternate aspects of the project,

I Alter the ‘constructors’ themselves, obliging them to adhere to differingevaluative and generative strategies.

I Manage the ‘managers’, privileging different ‘constructors’ or negoti-ation strategies over others.

I Share ‘filters’, adopting the point of view of other Users of ‘construc-tors’

This work describes ‘filter mediated design’ in the field of collaborativearchitectural design. However, if one were to replace architecturally spe-cific ‘constructors’ and ‘filters’, the underlying function of negotiation overdata based on the specific points of view of individual ‘constructors’, ‘man-agers’, and Users could serve other fields where coordination of multipleperspectives is a concern.

4.2 As a tool for researchIn addition to studying the processes and interactions at play in design,‘filter mediated design’ will provide a useful platform for exploring ideasranging from notions of self, to the location of meaning, to the understand-ing of ‘complex systems’. The generation of coherence from distributedprocesses draws from the work of Marvin Minsky who proposes that con-cepts of Self emerge from the interactions of many individual processes.When a design is collaboratively generated, issues of self and authorshipare called into question.

Generation of Coherence is a process of building shared Meaning. Meaningdoes not exist in the world, but rather is constructed by individuals andgroups. Privileging perception, as a process instrumental in the constructionof meaning, may help to understand the origins and maintenance of mean-ing. A computational understanding of the concept of emergence can beginto be explored.

Finally the mechanism, with the introduction of ‘managers’, is an idealmodule in which to study the interactions of ‘complex adaptive systems’.


This paper stated that ‘filters’ and ‘constructors’ evolve in relation to eachother and the world in which they operate. The method of coordinatingthese processes and the interactions between agents constitutes a complexsystem. ‘Filter mediated design’s’ modular nature should facilitate thestudy of organizational systems capable of generating coherence and inno-vation from a distributed system.

AcknowledgmentsMitsubishi Electric Research Lab has provided their support for this pro-ject. An implementation of ‘filter mediated design’ is being built on topof Mitsubishi Electric’s ‘Open Community’ (Mitsubishi Electric ResearchLaboratory, http://www.meitca.com/opencom/ (1999)), a decentralized dat-abase in which every user contributes a portion of the model. Open Com-munity provides the platform for sharing and maintaining consistency oflarge amounts of data over the web.

Notes1 Nelson Goodman elaborates on this subject by proposing five ways in which we construct our world forour own ends:

(1) Composition and Decomposition—world-making consists of taking apart and putting together.(2) Weighting—ratings of relevance, importance, utility, and value often yield hierarchies rather than dichot-

omies.(3) Ordering—nothing is primitive or is derivationally prior to anything apart from a constructional system.(4) Deletion and Supplementation—we find what we are prepared to find.(5) Deformation—the physicist smoothes out the simplest rough curve that fits all his data.

2 The question of what kinds of reconstruction are appropriate should be addressed. Piaget speaks ofassimilation and accommodation as two facets of the adaptive process. These are also at play in the construc-tion and maintenance of filters. Assimilation has to do with altering the perception of the world to makethings fit what we already know. The complementary function, Accommodation, alters the filter when theexperience of the world no longer supports the current representation14.


Download - Filter mediated design: generating coherence in collaborative design

Top Related