1. INTRODUCTION
When considered in the context of systematic studies of the design
process and its formalization, “design space” is an age-old
concept. The design methods movement of the 1960s has been one of the
first concerted efforts of exploring a problem domain, which we can
retroactively call the design space. These early efforts, at best, chipped
away at a difficult and poorly formulated problem. Attempts at exhaustive
and quantitative solutions did not go far (Jones,
1970; Wade, 1977). Simple-minded
formalizations merely revealed the complexities of the real challenge and
led to abandonment of these approaches even by their own authors (Alexander, 1964). More serious and less ambitious
steps were taken with the entry of artificial intelligence and cognitive
science approaches into the fray (Eastman, 1970;
Akın, 1986). These removed more layers of
ambiguity and misdirection from the larger research effort of
understanding the architecture of the design search process. Clear
explanations have not been easy to come by. From a distance, all of these
efforts appear as whittling away at the bark of, say, an old
growth trunk. If the paper by Woodbury and Burrow represents taking an axe
to it, then it promises to reveal more than has been accomplished before,
but the trunk would still remain massive and vast.
2. ANALOGY
The attempt to formally describe the concept of design space (taken in
its phenomenological sense, as that which the formalization effort
targets) is laudable and significant. I imagine that Woodbury and Burrow
imagine such a feat can yield the goods; goods that have eluded many
others attempting the same task, or other tasks of design formalization,
which in fact, may be rendered through this one.
Analogous goals have been attempted in a variety of domains. Chess,
medical diagnosis, mathematical discovery, and formal reasoning come to
mind. Some of these are even computational cousins of this effort. Some
attempts predate these efforts by centuries and do not even anticipate a
world of machine computing.
What would be the example that benchmarks the ultimate success of
formalized exploration spaces; one that would enable generations of future
scientists and designers to pluck dozens of both tractable and intractable
problems, as if they were an organized series of well-defined ordered
search acts in structured “space?”
How about the discovery of the Periodic Table of Elements?
Mendeleev's discovery has been nothing short of unveiling the
holy grail of exploration to material scientists (physicists,
chemists), who for many decades used this as their roadmap to elements
both known and unknown. A brief excursion to 1868 may be instructive.
The motivation for Mendeleev was the construction of the table of
contents of the second volume of his new text book on chemistry (Kedrov, 1966). In the first volume, he had already
covered the halogens and the alkaline metals. It was not clear as to which
group of elements should be covered next. In the absence of a logical
structure to organize all of the 64 known chemical elements of the day,
Mendeleev's exploration began with a search for a pattern that could
govern all of the elements. One can view this as the pattern that
structures the space of all elements.
Mendeleev started by comparing the atomic weights of elements.
Although this was a good start for organizing the problem space at hand,
there were two big obstacles. The first was the fact that the number of
comparisons of atomic weights of all known elements was far too large to
undertake exhaustively, not to mention the fact that the properties of
these elements were too imprecise, owing to the poor methods of
measurement available at the time, to make these comparisons accurately.
The second obstacle was due to the chemical elements not yet discovered at
the time, which made it difficult to see the global pattern in the
data.
Mendeleev started comparing groups of elements based on their
atomic properties and ordered them according to their atomic weights. This
reduced the space of comparisons considerably and started yielding some
consistent patterns. His second breakthrough came when he made a
modification in his representation of the elements to save time. He was
under pressure to go on a journey the next day and hand-written lists of
the elements became too cumbersome to manage. Thus, he decided to use
cards to represent elements ordered in a two-dimensional matrix space,
with one dimension representing the ordering of atomic weights and the
other general chemical properties (valences, and so forth) of the
elements. The cards containing the identities of chemical elements were
organized by Mendeleev in the same orthogonal fashion as the playing cards
of the solitaire called Patience, according to suit and
value.
This enabled Mendeleev to reduce the amount of clutter present in his
problem representation. Despite the unknown elements, the new
representation also made clearer the organizational principle of the
resulting construct: “properties of elements stand in periodic
relationship to atomic weights.” One of the greatest contributions
of the Periodic Table has been to facilitate, based on its underlying
logic, the further study of the known but poorly understood elements as
well as the discovery of ones that were unknown to scientists.
Mendeleev's discovery both provides an apt analogy for the
present task at hand and exhibits certain patterns that can constitute a
set of useful guidelines for making such discoveries. Mendeleev was
- working with a complete command of the knowledge in his
field,
- discriminating the “noise” in the data so that
the critical patterns are evident,
- identifying alternative approaches to the problem and
considering the most manageable ones first, and
- deriving general principles from specific relationships to
combine exploration in different problem domains1
In the original text, the term used is problem
spaces. I opted for substituting the word “domain” for it to
avoid confusion in my arguments, because I use this concept to guide the
exploration of a formal representation for problem spaces. Domain is a
problem–space–neutral term referencing roughly the same
semantic area.
Based on this line of reasoning, we can define the task proposed by
Woodbury and Burrow as one that would commence with a complete
understanding of the literature in the field of design, eliminate the
noise in this body of sources, identify alternative approaches to the
solution to be proposed, and combine insights from other domains to
structure and synthesize a solution. The first one takes a lot of hard
work, the second a discerning focus, the third an inclusive foresight, and
the fourth, nothing less than brilliance coupled with good fortune. Lest
you think this final factor leaves things to chance to an extent greater
than that which would be warranted in the case of geniuses like Mendeleev,
his familiarity with the solitaire game Patience and the
circumstances surrounding his determined, almost obsessive, concentration
on his task, over a very short period of time, were nothing short of
serendipitous.
Consequently, the specific assignment I gave myself, in discussing the
paper entitled “Whither Design Space?” by Woodbury and Burrow,
has been to consider the presence of these four conditions in their
documented exploration.
3. COMPLETENESS
Woodbury and Burrow draw from an impressive canvas of literature
representing research done in various domains of spatial design ranging,
impressively, along dimensions of both chronology and discipline. They
sample broadly from linguistics to generative algorithms, from
Simon's early work in the 1950s to his and other latest in the area.
They define a comprehensive scope for spatial design particularly within
the confines of a spatiovisual world.
They also hint at things that may be missing from this world. Early
on, while discussing the paucity of research on the design space, they
remark: “It may be that the current state of our knowledge of
representation and generation is inadequate to the task” (Woodbury
& Burrow's section 1). I agree with this assessment. In
discussing the “prosaic” use of spreadsheets as a part of
practice, they hint at the limitations that are implicit in these
applications for assisting in the creation of formal representations for
design spaces (Woodbury & Burrow's section 2.1). I agree with
this assessment also. There are many more hints in the paper (Woodbury
& Burrow's sections 3.3 and 3.8), at the shortcomings of
formalisms and applications developed by many a design space explorer,
including, to their credit, their own work, which share a thread with
these two points.
The observation that the underlying media in all of these approaches
is primarily spatiovisual is worth restating. Owing perhaps to the central
metaphor of space, in the term “design space,” and the
primarily visual orientation of architectural design fields, to an
obsessive degree I might add, all of the constructs reviewed or proposed
are laden with graphic representations. The nongraphic content of design,
often included in the form of design requirement models, are all but
nonexistent. This is the blind spot of research in architecture, and
indeed research in design computation. The situation is more than ironic,
if you consider that computing is a medium particularly suitable for
symbolic representation and processing, which are more suitable for
nongraphic expressions. This preoccupation with graphics appears to be
nothing short of disciplinary idiosyncrasy.
It is not the fact that there is no research in the area. It is,
however, the fact that there is a lack of powerful paradigms that
represent the design space of requirements. Furthermore, those
that interface the graphic and nongraphic entities exactly in the way
designers formulate and use them during design (Özkaya et al., 2004) are absent. This is a missed
opportunity for design research in general and for the Woodbury and Burrow
paper specifically.
Any attempt that models requirements alongside physical entities of
design inevitably needs to address the duality of information types.
Specifically, it must distinguish between requirements and physical
entities; define the symbiotic relationship between them; and provide for
the distinctions of storing, displaying, and computing within either set
simultaneously.
The typed feature structures (TFS) paradigm adopted in the Woodbury
and Burrow paper hints at this duality but primarily remains a homogeneous
representation. Requirements and other nongraphic information are
relegated to labels at best (Woodbury & Burrow's section 3.8).
The bifurcation between generation and evaluation is another manifestation
of the singularity of view. The role of evaluation in TFS is not clear. Do
paths and tendrils (Woodbury & Burrow's fig. 10) grow from and
append to one another due to some evaluation mechanism? If so, how does
this influence the “morphology” or “genetic code,”
to introduce yet another metaphor, of the space that is created? These are
questions that may posit interesting venues of development of TFS in the
design realm.
4. DISCRIMINATION
This is where the Woodbury and Burrow paper shines. It takes an
immense scope of a problem and brings it down to succinct enumeration of
concepts and ideas. It begins by structuring and justifying a paradigm for
formalizing the quasi-formal and certainly softer definitions of the
design space found in the literature. It proposes a multitiered
abstraction of this space and a specific approach that structures all of
these tiers (Woodbury & Burrow's section 1). It sets a premise
that suggests a convergence between the
type–procedure–operation hierarchy and the
action–amplification–computation operations (Woodbury &
Burrow's section 1). It then goes on to specify the action
domain through three central properties: representation, size, and
intention (Woodbury & Burrow's section 2). The bulk of the
discrimination effort exhibited by the authors is dedicated to the next
category of operations: amplification (Woodbury &
Burrow's section 3). It subsumes eight dimensions (representational
prowess, codification, the explicit space, implicature, speed, backup,
recall, and replay) that present an insightful and comprehensive view of
design spaces. Computation in comparison to amplification takes a
back seat in its generality. Clearly, this is not given detailed coverage
by the authors, at least in this paper. They primarily refer to work that
they have published elsewhere without getting into details. They conclude
by proposing an approach to resolve remaining issues, through analogy,
homology, and taxonomy.
This is a long, optimistic, and fairly written paper. Some of the more
obtuse passages and expressions notwithstanding, it does an excellent job
of discriminating thoughts, constructs, and relevant models for the design
space. It accomplishes more through the parsimonious and powerful TFS
model than those adopted in earlier attempts. It manages to explain and
meaningfully connect a vast space of research gathered from diverse
disciplines.
Digressing for a moment into the self-conscious realm let me observe
that Woodbury and Burrow, wittingly or not, fall prey to the cognitive
parameters by which we all abide. Many of their enumerations and
categorizations number in conformance with George Miller's (1956) maxim: “the span of STM, 7 ±
2.” This is certainly acceptable as we all conceive and communicate
within these cognitive parameters and exceeding them is neither expected
nor useful. The connection is not lost, at least on me, that the paper
does not take a particularly cognitive view of the phenomenon
under examination (design navigations within a space metaphor), which is
the focus of the next section.
5. ALTERNATIVES
Woodbury and Burrow pay attention to cognitive factors in design. They
not only display a basic understanding of designers' cognitive
parameters in building their arguments, but they also take it into account
in making proposals. Yet, this view of the world of design spaces does not
rise to the level of a significant alternative to their central
paradigm.
In enumerating sources for “computational access to the design
space,” for instance (Woodbury & Burrow's section 1), they
exclude perhaps the only source that treats the subject matter from a
cognitive viewpoint (Akın & Sen,
1996). In discussing amplification, they overlook a body of
literature that specifically deals with designer limitations (Flemming et al., 1997). These may not, in fact, be
oversights. They may have been intentional exclusions, as the
contributions of particular sources are assessed differently by different
researchers. The demonstration of the alternative view then should be
sought in the contents of the proposals included in the paper rather than
its specific citations. Does the Woodbury and Burrow paper incorporate the
cognitive view of the design space in its proposal? To examine this I
considered their “path” diagrams in their figure 10.
The path diagrams are pictures of metaphoric paths that represent
design exploration. Each tendril represents a “design event.”
Mapped into a literal space these events have certain characteristics such
as beginning and ending points, lengths, and internal state sequences.
There are interesting questions, in a formal representation, of preserving
these properties as well as manipulating them to advantage while
designing. Woodbury and Burrow discuss and discover important properties
towards formalizing this concept while leaving out some of the important
characteristics of cognitive events.
Reusing these tendrils in the course of design (which may be
equivalent to reusing a particular sill detail, for instance) is
represented as a mere “transplant” of the original tendril
onto the body of the state space path. Within a cognitive framework this
would play out differently. Cognitively speaking, an event (tendril)
previously experienced is never the same, another time around. There would
be economies of action, information, and representation that would be
realized with repetition. There may also be parsing of paths into
fragments or lateral connections between parallel paths, all of which have
precedents in protocol analysis literature (Akın
& Dave, 1986; Akın, 2002).
Furthermore, there are current formal models that go well beyond the
much maligned copy–paste models of which Woodbury and Burrow are
rightfully critical. It has been shown that the equivalents of their
tendrils of arbitrary length, beginning point, ending point, and state
sequence, can be formally represented from scratch as well as from
prerecorded tendrils. Such a formalism was recently tested using
cognitively based ethnographic records of design episodes (Akın & Moustapha, 2004).
6. COMBINED DOMAINS
A useful construct in studying design has been the
structure–behavior–function (SBF) paradigm (Chandrasekaran & Milne, 1985), which can provide a
valuable metaphor with which to fashion a formal ontology for design
spaces. This metaphor would present numerous fruitful correspondences to
constructs proposed in the Woodbury and Burrow paper, such as the
“designer action,” “amplification,” and
“symbolic representation” (Woodbury & Burrow's
section 2), which correspond respectively to behavior, function, and
structure. Because of the generality and power of its reach, however, the
SBF model has been interpreted liberally in a variety of fields.
Consequently, one of the immediate challenges in using it here is to
define its components unambiguously.
Although a great deal has been written about it in the design domain
(Gero & Kannengiesser, 2003), the genesis of
the SBF model belongs to chemistry. Returning to this domain, let us
consider the water molecule in Figure 1. It
illustrates some of the central structural properties of water.
Water is the simplest compound of the two most common reactive elements in
the Universe. This structure results in specific behaviors that
we can readily recognize.
The structure of the water molecule.
For instance, when the temperature of water exceeds 100°C or falls
below 0°C, under normal atmospheric pressure, it evaporates or,
respectively icing occurs. In the latter case, crystallization takes
place, increasing its volume. This is how ice can damage other materials
that have fine cracks into which it can enter in liquid form. These
behaviors, along with its structure, account for some of the very
important functions it serves: we drink it, wash with it, swim in
it, and cook with it.
Let us try to use this construct as the vessel into which we can place
some of the concepts proposed by Woodbury and Burrow for design spaces.
The “actions of the designer” would correspond to the
behaviors encountered while observing designers in action.
“Amplifying” these actions to serve specific purposes would
give us the palette of purposes, or functions, designs serve. The
constructs we would form through computable, symbolic representations,
which correspond to design structures, would help us realize
these amplifications.
Such an approach can well subsume the full complexity of
designer's actions, their amplification, and their computable
structures. The SBF metaphor can provide a clear set of the following:
- criteria for categories into which known or to be discovered design
phenomenon (just as in the case of the Periodic Table) can be placed,
including design requirements;
- formal relationships that structure the interaction between these
categories, including partial and cognitively morphed events, or
“tendrils,” and
- semantic threads that reach into a plethora of formal and case based
examples in different domains of problem solving and search.
Although the Woodbury and Burrow paper presents powerful ideas about
the domain of search spaces, its implicit metaphor of “searching in
a space” remains excessively generic and open ended. Consider the
path metaphor. What evidence is there that the parameters that define the
characteristics of a spatial path are the same as those that take place in
design? A spatial path is defined by dimensions that are continuous. Space
contains physical objects that obey the rules of physical dimension and
continuity. They possess properties of mass, occupancy, and the physical
laws of statics and dynamics. Is there any reason to believe that the
properties present in the space of design are isomorphic to those of
physical space? If not, then how far should this metaphor go in providing
the guidance toward formalizing design spaces?
Conversely, their explicit mapping into the TFS model developed for
computational linguistics (Carpenter, 1992)
appears to be specific and well defined. This mapping begs the question of
range. How successfully and to what extent can we map the world of design
into the world of linguistics? The field of architectural design is
populated with a fair number of language metaphors, similes, analogies,
and the like. Some have captured the popular imagination of architects who
are always eager to use loose reference, while, by and large, these
efforts have failed to produce formal models of design that behave as they
do in natural language.
A construct that provides both generality and power, such as the SBF
paradigm, is more likely to provide a fruitful breakthrough, as Mendeleev
was able to achieve. In summary, the SBF, when applied in a reverse
engineering mode,2
As one reviewer of this
paper put it “the analogy of the water molecule seems to indicate
that the structural properties of water result in specific behaviors that
we recognize and that account for the functions that it serves. Projected
onto Woodbury and Burrow's concepts, it could be interpreted as the
structural properties of computable, symbolic representations, and their
constructs give rise to specific processing techniques and methods
(behaviors) that account for designer action amplification
(functions).” At first (as suggested by this reviewer), the
causality appears to be counterintuitive. Computable representations do
not drive design behavior; they are created in response to them. Indeed,
this is so. However, the requirement for creating these representations is
to make sure that they are able to give rise to the appropriate behaviors
and functions, that is, some sort of reverse engineering is
required.
going from known behaviors to structures that
accommodate these behaviors and associated functions, can help tap into
the challenges and opportunities present in a very large body of relevant
literature in a very diverse set of design domains: engineering, sciences,
industrial design, and so forth. This may seem like an overly ambitious
strategy at first; yet the ambitious agenda of creating a formal and
domain-independent representation for design spaces needs an equally
ambitious strategy.
7. CONCLUSIONS
Woodbury and Burrow's work represents a significant step forward
in understanding and formally describing design. Its use of the space
metaphor, although unremarkable, is useful. Its utility arises principally
from raising interesting questions rather than answering them. Here are
some of them that provide food for thought.
- Is it possible to formalize design, with all of its phenomenological
complications, through a single, parsimonious model?
- What role should metaphors play in developing such a model, during
its conception, its formalization, versus its promotion through products
like papers and prototypes?
- What performance goals in the realm of design motivate and justify
such efforts?
- What role should human cognition play, if any, in defining these
performance goals, not to mention the paradigm underlying these
formalisms?
- Can a comprehensive objective, such as this one, scale beyond the
members of the inner group who produce it? What can be gained by
broadcasting it to other researchers, such as the effort that this journal
issue represents?
Although I am optimistic about the possibility of answering these
questions positively and constructively, particularly the last one, I
remain curious about the outcome, which should be manifest in this
collection of articles.
