Derix 2010: mediating spatial phenomena through computational heuristics

61

ACADIA 2010

life in:formation

Christian Derix
Aedas|R&D
United Kingdom

| 412

:author

:organization

:country

Mediating Spatial Phenomena
through Computational Heuristics
1 From Theory and Practice
A decade ago, a group of people was researching and studying at the University of East London for their
MSc Computing & Design that focused on the production of spatial organization and perception through
the medium of computation. Computing was regarded as an instrument for perceiving, representing,
and generating spatial phenomena, and thus challenged traditional production of architecture. Projects
were linked to design briefs because the master’s program was coupled with a diploma design studio.
Projects were to balance computing with design requirements as the Royal Institute of British Architects
(RIBA) were examining students as well as a representative of the academic field of design computation
(then John Frazer). A series of computational approaches were investigated for spatial properties of
architectural design for the first time such as L-systems and genetic programming (Coates et al., 1999),
swarms (Miranda, 2000) or neural networks (Derix et al., 2001). Theories were applied to explain the use
of computation within a scientific context with logical consequence on spatial and social phenomena
such as Maturana’s autopoiesis, Luhmann’s theory of social systems, or van Foerster’s second order
cybernetics. The reading of spatial theories was encouraged to inform the design of algorithms like
Perec, Hillier, Arnheim, or Norberg-Schulz. The course became a template for many university programs
such as UCL’s MSc Virtual Environment or the AA’s DRL. The projects were largely seen as academic
research with little relevance to the industry.
1.1 Tools vs Paradigms

From 2004, some members from UEL’s Computing & Design program formed the Aedas|R&D
Computational Design and Research group (CDR) to apply these academic approaches to industrial
briefs and open up the industry itself (‘educating the client’) to the potentials of generating spatial
organizations through computing.(1) The notion of space as a component of architecture was hardly
relevant as a driver for design over the last 15 years in the global industry. Especially, computation
is seen as a ‘problem-solving’ tool to support structural, geometric, climatic, or statistical aspects
of traditional workstages, not questioning the stages through the new medium per se. Architects use
computing as a way to erase differences with engineers, not to enhance their own knowledge of the
key aspects of architecture: occupation and space.
This is not surprising as academia deals with computation in an either completely mystified or demystified way, reinforcing the existing roles within design disciplines. The former sells computation as
artistic expressions of the new paradigm of complexity but offering only geometry via existing software
(for example ‘morphogenesis’ by Generative Components or Grasshopper) and pseudo-philosophy. The
latter develops serious scientific simulations for discrete aspects of design through research funding
that often produce opaque mathematical models, inadequate for design practice. The former being

62

ACADIA 2010

life in:formation

regarded by industry as a marketing tool, the latter as interesting research,
supporting distinct design stages – none as an applicable design methodology
or paradigm.

2 Losing the Jargon
First attempts to implement academic holistic models representing appropriate
simulations of cybernetic systems failed as they did not synchronize with existing
design workflows and shut out other designers and stakeholders. Such allinclusive models are still currency and in fact, it seems the trend is increasing
to build simulations that integrate as many scales, parameters and stakeholder
participations as possible. This Feature Creep is driven by industry-philosophy
such as ‘life cycle design’ or ‘building information models’ and tends to look at
a project from an observer point of view rather than the designer’s (top-down
rather than bottom-up).
As a reaction to this separation into simulation and (not with) design,
developments became increasingly bespoke to design stages, acting as
automation of design mechanism. This initially allowed for active participation
and synchronization in design stages but also submitted computation to the role
of tools, not methodology (Miranda and Derix, 2009). Consequently, the language
of communication about computational developments changed to professional
terms, losing the theoretic terminology applied in academia. This was reflected in
the teaching at the master’s at UEL where theory became subordinate to crafting:
learning to read and write code and techniques before learning to philosophize. The
change of curriculum at UEL bore immediate fruits as students had un-mitigated
access to theories of computation by understanding how to craft notions of
complexity themselves from scratch before mapping them into abstract paradigms.

3 Emerging Performances
The distinction into tools and methodology appears important to convey the
problem of function that is usually attributed to tools. Tools could be perceived as
automation of mechanism with fixed functions. While tools could be used for other
purposes, they have been designed with a specific purpose in mind. Computation
on the other hand doesn’t serve a specific function but as a universal machine
could simulate any mental tool if its function is abstracted adequately through a
computable representation. Computation is about the design of representations
and their dynamic organization. It doesn’t have properties that extend anything
else other than itself (binary representation and states). As such, a designer
could simulate any condition via computation if he knows how to create adequate
representations and organize their states, including spatial phenomena.
Initially, this might sound plausible and many seminal works about the nature
of artificial design (Simon, 1969) and the complexity of design representation
exist (Rittel & Webber, 1973). But attempts of knowledge transfer from sophisticated
academic research into design computation (as opposed to technology!) to the
design industry have so far left few traces. Software used in the architectural and
urban design industries are based on automation and technological support,
not design methodology. Further, that type of software is often not developed by
academics but by practitioners themselves.
Mediating Spatial Phenomena | panel

63

ACADIA 2010

life in:formation

| 412

Two areas traditionally identified by academia for design computation are layout automation and urban
simulation, as it appears to represent the ‘right problem to solve’ by computation, i.e. a combinatorial set of
statistical quantities and some abstractable geometry. But as Liggett concludes in his survey about layout
automation, no development has been successfully integrated by architectural practice (Liggett, 2000). That
still holds true no matter if the publication is already 10 years old. Equally, urban simulations have found
their application so far only at planning levels for land-use evaluation, demographics or transport, not at
urban morphology levels where designers search spatial solutions.
This appears to stem from two conditions: a lack of understanding ontological representations
made by designers and a technological ambition to create holistic simulations. The objects to be
represented in all layout simulations described by Liggett - for example HeGeL, LOOS or Wright - are
driven by computational techniques of close-packing and providing full convergence of solutions before
allowing the user to inspect the results. Although HeGeL already employs heuristics as evaluation
criteria, it still overlooks the nature of layout generation as a multi-stage and dynamic heuristics
process, meaning that no single system, no matter how complex, can satisfy the design search space
for layouts. Layout design happens through many representations for topologies, approximations of
shapes for rooms, grouping of activities, organizational hierarchies, operational arrangements, design
intentions and intuitive adaptation for under-constrained components like circulation through nongeneric heuristics. Even social dynamics for communication, micro-climate, and cognitive qualities
are starting to inform the constraint space. Technological ambition sidelines the designer into a seedwatch-evaluate role who feels his intentions and heuristics are not participating in the search. This
is particularly dangerous for urban design where current urban simulation developments bundle so
many scales and performances based on the ambition to create city information models (CIM) on the
template of building information models (BIM) that complete black-boxes result where the designer
adapts to the software, not the other way around.
Wicked problems (Rittel & Webber, 1973) like layout or urban design require the experience of designers
to negotiate the many explicit and implicit aspects that can be represented through computation.
Particularly, when design aspects are not discursive and the amount of data is large, the key organizing
principle of designers and design teams are their learned heuristics, not performance indicators and
data sets. While computation shouldn’t imitate analogue heuristics, it can express its own search
mechanism via visualization of processing steps. If a designer can interfere with computational
heuristics and observe the search struggle, the opportunity for identification between designers’
analogue and computational heuristics are given. This type of empathetic coupling (or ‘structural
coupling’ as Maturana calls it in his autopoietic theory) enables the validation for wicked problems
when no explicit goals are set. It constructs narratives communicated between designers that eventually
produce generalized abstract approaches to design tasks. For computational designers, such abstract
search mechanisms are meta-heuristics. Meta-heuristics can not be applied immediately but need to
be adapted through implementation into design contexts. In other words, meta-heuristic algorithms
are not to be taken literally and have no inherent automatic aesthetic but produce meaning when
situated.(2) Layout or urban design is composed of many design scales and aspects that designers have
developed many heuristics for. Each aspect is complex in itself and thus a multitude of computational
simulations need to be negotiated by a designer and evaluation spread across a series of applications.
One of the key problems with layout design represents the non-quantitative nature of circulation.
Designers don’t necessarily want close-packed topological diagrams as they know that the layout will
change many times over and no specific cost-function or geometry can be given to circulation.(3) Few explicit
performance criteria exist for circulation other than the regulatory maxima for area and fire routes, while
inherent performances to do with spatial, social, and cognitive properties rest within the forming intentions
of the designer. Circulation emerges from a series of negotiated implicit and explicit design searches.
Where lightweight computational simulations exist that can be assembled into integrated hybrid workflows,
design intentions for implicit architectural properties can gradually emerge through the empathetic
panel | Mediating Spatial Phenomena

ACADIA 2010

64

life in:formation

Figure 1. Four of the layout
applications developed
by CDR. From left: bubble
diagrams with graph
grammars, semi-automatic
agent & graph-based
hybrid, attract-repel
based volumetric and
evolutionary 3D packing.

coupling of heuristics. The lightweight applications must be limited in functionality
(parsimonious) and visualize in real-time the ‘intentions’ of the simulation as
it searches, to render its heuristics transparent. Like many analogue precedents,
models need to be interactive and responsive during run-time. Frei Otto’s models
as well as interactive workshops in urban design (‘charrettes’) show how to learn
from autonomous processes: the models represent independent organizations and
can not be linearly affected but only ‘nudged’. It’s a kind of dynamic ‘stygmergic’ (4)
system where designers’ interferences are absorbed visually, allowing all participating
stakeholders to understand the organizing principles of the autonomous system.
Knowledge can be extracted via the interaction between designer and model, or
the model and its context. Models generate diagrammatic ‘states’ as outputs, not
optimized results.
As the organizing logic is of simulations distributed across the constituent
parts that produce consensual states as outputs, it becomes difficult to attach the
often desired explicit ‘performances criteria’. While some basic measures can be
extracted, many desirable states must emerge as a validation of the assumptions
that a computational representation and its processes is built around. For most
architectural design aspects, many desired performances must emerge, not be
hard-coded as sum measures of some parts.
A common reaction to the above arguments is that this is ‘old hat’ by now.
But one can hardly see this approach practiced as it requires constant learning
of dedicated teams that can construct frameworks of computational heuristics
within live design contexts (hence industry has an edge over academia in this
field of research). Only when these new design heuristics based on computation
have been learned and generalized into frameworks through many repetitions
and on complex building types (another issue with computational design
education are overly simple building types) can new workflows emerge that noncomputational designers trust and couple with.

4 Hybrid Paradigm: Value of Spatial Phenomena
Clearly, technology advances play a crucial role in the realization of such
computational approaches. Real-time processing, run-time interaction, and
accessible APIs and SDKs, as well as publicly available GPL libraries support this
computational design development. But it remains in the hands of experienced
designers to responsibly use this technological power and resist the common
temptation to increase the complexity of models.
After the experience with academic and bespoke parametric models, the CDR
used two projects in 2007 – the Ground Zero Memorial Museum and Smart

65

ACADIA 2010

life in:formation

| 412

Figure 2. Four spatial analysis
applications from the Ground
Zero Memorial Museum project.
From left: dynamic path
based view-sheds, sections
from Visual Graph Analysis,
polyhedral isovist geometry
and visual integration with
gravitational mapping.

Solutions for Spatial Planning – to start building an open framework for heuristic design simulations in
Java and OpenGL. The projects aimed at the development of an assembly of applets that would investigate
the design and representation of various spatial phenomena for architectural and urban design and the
alignment with designers’ heuristics. It allowed for example the long overdue extension of Space Syntax’s
2-dimensional analytical methods into interactive real-time and dynamic – i.e. time-based - spatial
simulations that architects and planners can choose to simply analyze designs with or create generative
design procedures from. Currently, the framework consists of a long list of algorithms, meta-heuristics, and
data structures that allow the group to design intuitively through computation and create new heuristics.
Having arrived at a minimal maturity and being able to demonstrate to clients this new design paradigm
of artificial heuristics through computation, a careful return to the initial debate about the representation
and generation of spatial phenomena might be possible. Particularly, the debate about the ‘value of design’
being led in the industry presently allows an entry for computational approximation of spatial phenomena.
In a negative economic climate, companies struggle to create a competitive edge through ‘unique selling
propositions’ (USP) paired with a wariness of complicated building structures as the only visible outcome
from computation so far. A shift to hitherto implicit design values is demanded that are located within
social, cultural, environmental, behavioral and cognitive conditions.
Dusting off the old arguments with new methodological insights appears to gain momentum. Theories
and observations of spatial organization and perception from George Perec (Perec, 1974) to Rudolf Arnheim
(Arnheim, 1977) can support the design of computational heuristics and their emerging evaluation criteria.
It requires also the inclusion of previously ignored design partners like cost-consultants or developer
agents who can provide insights into tacit evaluation mechanisms and who must adjust their models to
more contextual conditions. The value of objects (solids) in a design is ever more moving towards the
original notion of ecology (Maruyama, 1963) that can now be simulated in design searches via computation
as the behavior – organizationally not statistically – of the system can be visualized and interacted with.
Visualized relationships and organizational dynamics that objects (dynamically) belong to determine
their value as much as performance values and material cost. Spatial value can be approximated
in relation to non-discursive dynamics like occupational potential and cognitive performance. These
relations can partially be made explicit through design intentions that emerge through the coupling of
designers’ and computational heuristics. Circulation can be understood to have more value than just
egress times, flow capacities and area. It emerges from the mediation of multiple simulated conditions.
Naturally, it is very difficult to avoid tautologies in the set up of the models and instill truly emergent
properties. Many researchers have attempted to create ‘objective’ representations like the members of
the Cambridge school - Haggert & Chorley, March & Steadman, etc. (Keller, 2006) - and some artists have
shown how slippery the slope can be for assumed values of spatial properties - Bense (Bense, 1960) or
Klein (Noever, 2004). Now, with recent developments, architects are finally in the position to allow for better
arguments to use computation that sits as a spatial design paradigm between the scientific(5), problemsolving and the purely artistic. ‘Good design skills’ are not dependent on computing knowledge. But they
facilitate deeper understanding of spatial conditions and how to form new design search strategies. The
negotiating of artificial and analogue heuristics enables designers to find new expressions for sensitive
spatial designs that will probably embody very simple geometries but subtle perceptive intricacies.

“Space between things turns out not to look simply empty” (Arnheim, 1977)
panel | Mediating Spatial Phenomena

66

ACADIA 2010

life in:formation

References
Alexander, C. (1967). Notes on the synthesis of form. Cambridge, Mass: Harvard University Press.
Arnheim, R. (1977). Dynamics of architectural form. Berkley and Los Angeles: University of California Press.
Bense, M. (1969). Programmierung des Schoenen. Aesthetica IV. Baden Baden: Agis Verlag.
Bonabeau, E., Dorigo, M., & Theraulaz, G. (1999). Swarm intelligence: From natural to artificial systems. New
York and Oxford: Oxford University Press.
Coates, P., Jackson, H., & Broughton, T. (1999). Exploring three-dimensional design worlds using Lindenmeyer
systems and genetic programming. In Evolutionary design by computers, ed. Bentley, P., 323-341. San Francisco,
Morgan Kaufmann.
Derix, C., Coates, P., Simon, C., & Appels, T. (2001). Dust, plates and blobs. In Proceedings of Generative Arts
conference 4. Milan: Politecnico di Milano.
Derix, C. (2006). Approximating phenomenological space. In Intelligent computing in engineering and
architecture, 13th EG-ICE workshop, ed. Smith, I., 136-146. Berlin Heidelberg: Springer Verlag.
Elezkurtaj, T. & Frank, G.(2002). Algorithmic support of creative architectural design. In Umbau vol. 19, 129-37.
Dubai: SPMS.
Keller, S. (2006). Fenland tech: Architectural science in post-war Cambridge. In Grey Room 23, 44-65. Cambridge,
Mass: MIT Press Journals.
Liggett, R. (2000). Automated facility layouts: past, present and future. In Automation in Construction, vol. 9,
197-215. Amsterdam: Elsevier Scientific Publishing, Inc.
Maturana, H. & Varela, F. (1987). The tree of knowledge. Munich: Goldmann.
Maruyama, M. (1963). The second cybernetics: Deviation-amplifying Mutual Causal Processes. In American
Scientist 5 (2), ,164-179.
Miranda, P. & Coates. P (2000). Swarm modelling: the use of Swarm Intelligence to generate architectural form.
In Proceedings of Generative Arts conference 3. Milan: Politecnico di Milano.
Miranda, P. & Derix, C (2009). Global practices of computational design. In Proceedings of ACADIA 29: 242-244.
Chicago: School of the Art Institute of Chicago.
Noever, P. (2004). Yves Klein: Air architecture. Ostfildern-Ruit: Hatje Cantz Publishers
Perec, G. (1974): Species of spaces and other pieces. Paris: Editions Galilee.
Rittel, H. & Webber, M. (1973). Dilemmas in general theory of planning. In Policy Sciences, vol. 4, 155 – 169.
Amsterdam: Elsevier Scientific Publishing, Inc.
Simon, H. (1969). The Sciences of the Artificial. Cambridge Mass: MIT Press.

Endnotes
(1) The Computational Design and Research group consists of Pablo Miranda, Åsmund Gamlesæter, Lucy
Helme and Christian Derix.
(2) There is a mis-perception that algorithms, especially heuristic algorithms like evolutionary algorithms,
should produce complex-looking or organic outputs. After a lecture at the Hyperbody at TU Delft, a student
asked the question: “Why don’t your designs look more funky when you are using so much computation?”
(3) And they feel that their experience – heuristics and intentions – is overridden if they cannot identify with
the visualized process and result. A demonstration of Eleczkurtaj’s excellent layout simulation (Elezkurtaj
and Frank, 2002) at Aedas architects in 2005 resulted in a categorical rejection as architects voiced the here
mentioned reasons. The CDR has as a consequence built series of simulations dealing with distinct aspects of
layout generation that the architects can assemble themselves according to brief and intentions.
(4) Stigmergy represents a kind of indirect communication system via an external environment
(Bonabeau et al, 1999)
(5) Christopher Alexander’s skepticism is representative that informed Notes on the Synthesis of Form where
he states: “Though design is by nature imaginative and intuitive, and we could easily trust it if the designers’
intuition were reliable, as it is it inspires very little confidence.” (Alexander, 1967).

Derix 2010: mediating spatial phenomena through computational heuristics

Recommended

More Related Content

What's hot (18)

Viewers also liked (16)

Similar to Derix 2010: mediating spatial phenomena through computational heuristics (20)

More from ArchiLab 7 (20)

Recently uploaded (20)

Derix 2010: mediating spatial phenomena through computational heuristics