SOFTWARE

MODEL
OPTI MI Z ATI ON
BUI LDI NG
PATTERN
PRECI SI ON
MATTER
CONTROL
MATERI AL
COMPONENTS
TI ME
GEOMETRY
DATA
APPLI CATI ON
ANALYSI S
PRACTI CE
DI AGRAM
STRUCTURAL
DEVELOPMENT
SI MUL ATI ON
I NFORMATI ON
DI FFERENT
PERFORMANCE
SPACE
COMPUTER
FABRI CATI ON
TECHNI CAL
TOOL
NEW
FORM
CONSTRUCTI ON
PRODUCTI ON
MODEL
ASSOCI ATI VE
SHEL L
ORDER
WORK
MANAGEMENT
COMPLEX
ELEMENTS
PROCESS
SYSTEM
PROJECT
STANDARD
DIGITAL WORKFLOWS IN ARCHITECTURE
DESIGNASSEMBLYINDUSTRY
SCOTT MARBLE, EDITOR
D
I
G
I
T
A
L

W
O
R
K
F
L
O
W
S

I
N

A
R
C
H
I
T
E
C
T
U
R
E
D
E
S
I
G
N

A
S
S
E
M
B
L
Y

I
N
D
U
S
T
R
Y
S
C
O
T
T

M
A
R
B
L
E
,

E
D
I
T
O
R
DIGITAL WORKFLOWS IN ARCHITECTURE
DESIGN ASSEMBLY INDUSTRY
THE ASSIMILATION AND SYNTHESIS OF DIGITAL COMMUNICATIONS AMONG ARCHITECTS,
ENGINEERS, FABRICATORS AND BUILDERS IS DRAMATICALLY ALTERING HOW WE WORK
AND OUR RELATIONSHIP TO THE TOOLS WE USE. NEW DIGITAL CAPACITIES ARE
RESTRUCTURING THE ORGANIZATION AND HIERARCHY OF DESIGN FROM AUTONOMOUS
PROCESSES TO COLLECTIVE WORKFLOWS. THE HISTORICAL ROLE OF THE DESIGNER
AS AN AUTHOR, A SOLE CREATOR, IS BEING REPLACED WITH SEMI-AUTONOMOUS,
ALGORITHMICALLY DRIVEN DESIGN WORKFLOWS DEEPLY EMBEDDED IN A COLLECTIVE
DIGITAL COMMUNICATION INFRASTRUCTURE. THIS IS CREATING A NUMBER OF
PRESSURES ON THE DISCIPLINE OF ARCHITECTURE TO REORGANIZE AROUND THE
OPPORTUNITIES, AND RISKS, OF THESE CHANGES.
EDITED BY SCOTT MARBLE, PARTNER AT MARBLE FAIRBANKS AND THE DIRECTOR
OF INTEGRATED DESIGN AT THE COLUMBIA UNIVERSITY SCHOOL OF ARCHITECTURE,
PLANNING AND PRESERVATION, THIS BOOK GATHERS SOME OF THE LEADING VOICES
ON THE ISSUES AND SOLUTIONS OF THE EMERGING WORLD OF DIGITAL WORKFLOWS.
CONTRIBUTIONS BY
FRANK BARKOW and REGINE LEIBINGER / Barkow Leibinger
DAVID BENJAMIN / The Living
BEN VAN BERKEL / UNStudio
PHIL BERNSTEIN / Autodesk Inc.
SHANE M. BURGER / Woods Bagot
NEIL DENARI / NDMA
MARTY DOSHER / SYNTHESIS
JAMES KOTRONIS / Gehry Technologies
SCOTT MARBLE / marble fairbanks
ADAM MARCUS / University of Minnesota
THOM MAYNE / Morphosis
JOHN NASTASI / Stevens Institute of Technology
JESSE REISER and NANAKO UMEMOTO / reiser + umemoto
FABIAN SCHEURER / designtoproduction
CRAIG SCHWITTER and IAN KEOUGH / Buro Happold
PAOLO TOMBESI / University of Melbourne
1 4 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

David Benjamin is Co-Founder of The Living and Assistant
Professor of Architecture at Columbia University GSAPP.
EFFICIENCY

In 1906, a bearded Italian academic named

Vilfredo Pareto published a thick book entitled
Manual of Political Economy.
1
It was filled with
small hand-drawn graphs and with hundreds of
equations, the algorithms of this era. It was the
authors fifth book, and it represented a retreat
into mathematics by a man who had once been
so fiery and political that his lectures had to be
advertised in secret or risked being raided and
shut down by government thugs. Yet Paretos
new math was just as fresh and revolutionary as
his old political critiques. And while his book was
underappreciated at the time, it is now cited as
initiating the study of modern microeconomics.
One of the concepts advanced in the
Manual of Political Economy was called Pareto
efficiency. It referred to a society where nobody
can be made better off without somebody else
being made worse off. A society in this state
of equilibrium was considered to be Pareto effi-
cient, or Pareto optimal, and a series of potential
societies could be neatly graphed so that all of
the Pareto efficient ones fell on a Pareto frontier,
the set of best-performing arrangements. As
an example, Pareto imagined a society with a
fixed amount of resources for producing bread
and wine. One option would be to make a large
amount of bread and a small amount of wine.
Another option would be the other way around.
For each option, if production was efficient,
then this version of society would become a
point on the Pareto frontier. And since all
of the points on the Pareto frontier were
mathematically equivalentthey were all
optimalthis frontier could be used to study
trade-offs between different options for dis-
tributing resources. [Figure 1]
While the principle of Pareto efficiency
was developed to describe economics and the
design of societies, it can also be applied to
design more generally. It can be used for many
design problems in many fields. Since engi-
neers often work on problems with two or more
clearly defined objectivessuch as the design
of an analog circuit with the objectives of least
footprint and greatest filtering of low frequen-
ciesthe Pareto frontier offers a very helpful
framework for mapping design permutations
and selecting the best ones. And in recent years,
architects have used a similar framework with
aspects of building designsuch as the design
of a structural frame with the objectives of least
displacement under load and least amount of
material required.
In this sense, perhaps Paretos Manual,
and the field of economics in general, have
provided a foundation for the current interest
in performance and optimization in architec-
tural design. Architects today, like economists
and engineers, are enchanted by efficient and
optimal designs because they can be clearly
identified. They are so unambiguous that they
seem inevitable. They are so computationally
pure that they seem desirable. But as Pareto
noted a century ago, efficiency is very narrow.
An efficient distribution of resources does not
necessarily equal a socially desirable distribu-
tion of resources. An optimal design does not
necessarily equal a good design. Along these
lines, critics of a performance-based approach
to complex architectural problems have argued
that good design cannot be reduced to mere
numbers. For them, judgment and intuition are
crucial components of architectural design
with nuances that cannot be quantified or
graphed on a Pareto frontierand they should
Figure 1. Pareto graphed bread
production and wine production to
derive a simple Pareto frontier. Each
point on the graph represents the
resource distribution in a different
potential society.
1 5 DESI GN ASSEMBLY I NDUSTRY BENJAMI N DESI GNI NG

not be ignored or diminished. In other words, to
rely too heavily on performance and optimiza-
tion as the drivers of design is to fall under the
spell of a dangerous myth of efficiency.
On the other hand, dismissing numbers and
optimization may be equally one-sidedespe-
cially in the current technological context where
almost every daily activity is conditioned by
numerical logics. To rely exclusively on judgment
and intuition to address complex architectural
problems is to fall under the spell of a different
dangerous mytha myth of creativity.
The myth of efficiency is countered by the
myth of creativity, and vice versa. But in our
current moment of data and digital processes,
the issue is not whether to choose one side or
the other, but rather how to structure a rela-
tionship between the two that balances their
different strengths.
EXPLOITATION VERSUS EXPLORATION

The field of statistics has developed several

robust approaches to achieving this kind of
balance between competing demands. One
approach offers a mathematical solution for a
situation called the multi-armed bandit prob-
lem. In this problem, there are two levers on a
slot machine and information is known about
the performance of one lever but not the other.
The question is which lever to pull, and how to
determine a sequence of steps that balances
two possible moves: utilizing existing infor-
mation (through pulling the known lever), or
searching for new information (through pulling
the unknown lever). More generally, this type of
problem involves what statisticians refer to as a
spectrum between exploitation and exploration.
For design problems, this spectrum may
be useful even if designers have different aims
than statisticians. In order to be more specific
about the balance between efficiency and
creativity in design, it may be helpful to outline
these terms on a spectrum between exploita-
tion and exploration. An example of exploitation
might be the optimization of the nose cone of
a high-speed train for aerodynamic perfor-
mance. Here, designers want to find the single
best-performing design according to only two
objectives: least drag and greatest crosswind
stability. They want to exploit small differences
in inputs to achieve the best output. They use a
narrow range of geometric variations with the
ultimate goal of achieving an efficient result.
[Figure 2] On the other hand, an example of
exploration might be the evolution of spanning
structures that are high-performing but also
unexpected. Here, designers want to find
novel designs above a minimum threshold of
structural performance. They want to explore
completely different sets of inputs that might
result in similar outputs (for example, spanning
structures with different forms but the same
performance under load). They use a wide
range of geometric variations with the goal of
achieving an innovative result. [Figure 3]
To identify whether a design problem lends
itself more to exploitation or to exploration, it
may be helpful to draw a map of its potential
design spacethe space of all possible design
permutations. When a design space is visualized
in three dimensions, it becomes a topological
surface, with the x-axis and the y-axis register-
ing the properties of potential solutions, and the
z-axis measuring the performance. By defini-
tion, specific design solutions must lie on this
topological surface, which then becomes an
extremely consequential design constraint.
Designers interested in exploitation prefer
a narrow, continuous design space, such as a
slanted plane or a topological surface with one
or two bumps. In this case, it is possible to quickly
hone in on the region of best performance and
to locate the single global maximum. The sim-
pler the design space is, the faster they can find
the optimal design.
Designers interested in exploration prefer
a wide, discontinuous design space, such as
a jagged mountain range with multiple peaks.
In this case, there are many distinct regions
Figure 2 (above left). Each design
permutation involves a different
geometric form and a different level
of aerodynamic performance for
the Zero train by Bombardier. Over
time, the automated process evolves
higher-performing designs.
Figure 3 (above right). Simple
building blocks are assembled in
different arrangements created with
EvoCAD by Jordan Pollack and the
Dynamic and Evolutionary Machine
Organization at Brandeis University,
in order to create a structure that
spans from right to left. Over time,
the automated process evolves novel
designs that meet minimum perfor-
mance requirements.
A pillar edge
atness of nose
undercarriage keel
1 6 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

of good performance, and it is often possible
to find multiple local maximums that are both
interesting and high-performing, even if they
are not the global maximum. The more com-
plex the design space is, the more likely it is
that they will make an unpredictable discovery.
[Figure 4]
As architects struggle to balance the
power of computation with the need to maintain
control, the exploitation/exploration spectrum
and the visualization of design space may offer
useful frameworks for studying and extending
current digital tools and methods. Five potential
strategies for designing design can be derived
from this context.
DESIGNING THE ALGORITHM

It is obvious but worth stating that algorithms

have a great influence on the outcome of archi-
tectural design processes. Algorithms specify
the way a double-curved surface takes its
shape from two wavy lines; they define the way
complex structural forces are broken down into
finite elements; and they perform more complex
tasks that may be closer to artificial intelli-
gence. Algorithms written by corporations such
as IBM are now able to compete in (and win) the
game show Jeopardy, or quickly find pertinent
information in mountains of legal documents (a
task that would take a human weeks to accom-
plish). Algorithms written by academics such
as Hod Lipson at Cornell University can invent
new machines without direct human input, or
take any data set and produce a mathematical
formula to explain it. [Figures 5a, 5b]
Yet algorithms are not neutral or inevitable.
They are designed with assumptions and biases
that condition what they produce. And if these
assumptions were different, the designs produced
through them would be different. In architecture,
digital workflows and applicationssuch as
software for parametric modeling, BIM, build-
ing simulation and building optimizationpush
Figure 4 (above). Each graph rep-
resents a different design problem.
As the problem becomes more
complex, the surface describing
the design space becomes more
complex. Finding the maximum
point is relatively straightforward
for the problem on the top, but more
challenging for the problem on the
bottom.
Figures 5a (top), 5b (bottom). With
almost no predetermined rules,
the algorithm in the Golem Project
by Hod Lipson and Jordan Pollack
evolves successful but unexpected
designs for congurations of robots
that move rapidly in a specied
direction.
e
x
p
l
o
i
t
a
t
i
o
n
/
e
x
p
l
o
r
a
t
i
o
n
1 7 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

Figures 6a (top left), 6b (top right),
6c (bottom left), 6d (bottom right).
Casey Reas has designed both the
algorithms of the core application
and the algorithms of his individual
projects, using Processing to create
innovative drawings.
BENJAMI N
1 8 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

us in certain design directions. They play a
significant role in architectural design, defin-
ing the design space of any project that uses
them. Software has an enormous influence
on the topological surface that describes any
design problem. While end user input affects
the features of this topological surface, soft-
ware defines the type of topological surface.
A change in software would mean a change in
type. But architectural software, and the algo-
rithms behind it, are usually authored by teams
of computer scientists, and they are usually
opaque to architects. The assumptions and con-
sequences of the algorithms are often difficult
for architects to ascertain. As computer scientist
Eitan Grinspun observes: I think architects
have been hijacked! The tools they are working
with are written by programmers whose training,
by and large, comes from a very scientific,
engineering-based mindset. These tools provide
the language that architects have to use, and as
architects start becoming more proficient with
the tools, they start adapting to the engineers
language.
2
In other words, the sheer fact of
using architectural software means already to
operate like an engineer.
Perhaps this could change. It might be pos-
sible for architects to become more proactive in
understanding, and even authoring, some of the
algorithms that influence their designs. Instead
of going through a typical sequence of architects
demonstrating creative control through analog
processes, and then struggling to maintain
control as digital processes are applied, it may
be possible for architects to lead these digital
processes. In other words, one option for
designing design is designing the algorithm.
Of course, many young architects are
already fluent in developing digital workflows
and writing custom scripts within software
applications. But the algorithms in these custom
scripts may have a limited influence on the
design results in comparison to the algorithms
in the core software application. In order to
avoid being hijacked, as Grinspun described,
architects may need to play a more active role
in designing the core software. This is no easy
feat, but there are a few precedents.
In the film animation industry, software
companies often provide early versions of
applications to animators and deliberately seek
input on core features and uses of the soft-
ware before it is fixed and frozen. This kind of
open-ended input is different from input during
typical beta testing of software that simply aims
to fix bugs and refine features that have already
been finalized by the programmers.
A more profound example involves designers
writing entirely new software applications.
In 2001, designers Casey Reas and Ben Fry
released the first version of a software applica-
tion called Processing. This programming
language was created by designers who also
had deep experience with computer science.
Processing emphasized interactivity and visu-
alization, and it allowed Reas and Fry to create
their own designs that would not have been
possible otherwise. [Figures 6a, 6b, 6c, 6d] But
it also offered other designers a new, open-
ended platform. It was software for designers by
designers.
3
Ten years later, there are thousands
of diverse design projects created through
Processing, and there is also an incredible
online community around this tool and these
projects. [Figure 7] Yet Processing is just the
beginning of what might happen if designers
authored their own software and made deliberate
decisions about the assumptions embedded in
it. For the field of architecture to take control of
digital processes by designing the algorithm, it
may benefit from a few architects designing a
few new software applications like Processing.
COMPUTING THE SOCIAL

One aspect of computation that has become

increasingly important in architectural design is
software for simulating building performance.
The design of most complex architectural projects
now involves several types of digital simula-
tion, including finite element analysis (FEA) for
structure, computational fluid dynamics (CFD)
Figure 7. The Processing website
offers tutorials, example projects, a
source code, a community of users
and free downloads of the software
application.
1 9 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

for airflow and environmental/energy analy-
sis for building systems. Simulation software
is especially useful when the design problem
is complex (when the topological surface of
design space is like a jagged mountain range)
and when it is difficult to predict the best design
based on standard rules of thumb. The process
of digital simulationoften combined with digital
optimizationhas enabled design workflows
that produce convincing quantitative results,
which in turn have contributed to the trend of
performance-based architectural design. In
other words, if Paretos principles provided
the foundation for performance-based design,
simulation software provides the fuel.
But clearly there are aspects of design that
are difficult to digitally simulate and optimize.
There are qualitative features of architecture that
seem beyond quantification. While quantitative
featuressuch as structural and environmental
performanceare typically driven by simula-
tion software, qualitative featuressuch as
aesthetics and atmosphereare driven by less
computational, more open-ended processes.
Often this leads to a duality in the design
process, a duality which may simply be another
version of the efficiency/creativity division or
the exploitation/exploration spectrum. Yet this
duality may be too simplistic, and it has been
challenged by some recent projects that apply
simulation software to new domains.
AnyBody is a software application that
simulates human well-being and ergonomic
comfort. It has been used in the design of auto-
mobile foot pedals and wheelchairs. [Figures 8a,
8b, 8c] From the perspective of engineering
design, the software is useful since the com-
plexity of human anatomy makes it difficult to
compute forces and stresses by hand.
4
From
the perspective of architectural design, the
software is interesting because it involves
a system of metrics for quantifying human
comfort, which is normally considered to be a
qualitative feature outside the domain of compu-
tational analysis.
In a related area, a growing number of
software applications now simulate human
crowd behavior, including SMART Move by Buro
Happold. [Figure 9] These simulations consider
the interactive behavior of very large numbers
of people, which would be difficult to calculate
without computation. In fact, most crowd behav-
ior software simulates the flow of people much
like CFD simulates the flow of air, by means of
simple rules and mathematical vectors.
Vacate is a crowd behavior application that
adds another level of complexity. It simulates
human behavior during the emergency evacu-
ation of an airplane, and it computes social
interaction and factors of human psychology
that include situations like the presence of a
group leader, the act of helping a disabled
person, panicking, seat jumping and competi-
tive vs. cooperative behavior. [Figures 10a, 10b,
10c] With Vacate, critical human behaviors are
identified from a computer database known
as Aircraft Accident Statistics and Knowledge
(AASK)developed from survivors accounts
of aviation accidentsand then programmed
into the software. Since many social and
psychological behaviors have been identified
Figures 8a (top), 8b (above left),
8c (above right). In this simulation,
AnyBody starts with the geometry
of human bones and muscles, adds
multiple forces of stress and strain
that continually change as the body
moves, and identies a numerical
level of comfort for each scenario.
Figure 9 (above). With SMART Move,
the crowd behavior simulation pro-
duces both performance metrics
such as duration of circulationand
animations for visual representation.
BENJAMI N
20 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

and cataloged in this database, it is possible to
integrate them into a computational model and
establish an environment for digital simulation.
Vacates simulations provide a unique resource
for design, as it is difficult to obtain reliable
data from real accidents and also because there
are ethical issues involved with conducting
physical experiments to generate these data.
5

Vacate offers an example of computing the
social, and as the category of simulation soft-
ware expands to include features for predicting
soft reactionssuch as human comfort and
human movement in emergenciesthe territory
of computation in architecture may shift.
DESIGNING THE PROBLEM

Airports
6
are complex, and they are also data-
driven, so it makes sense that they would be a
testing ground for some of todays most com-
plex digital workflows, and an important site for
balancing efficiency and creativity. The design
of airports already incorporates a wide variety of
simulation applications, both for far-reaching
topics such as financial models of GDP and simu-
lation of urban growth, and for building-specific
topics such as structural and environmental
analysis or the flow of planes, cars, baggage
and passengers (though the crowd behavior
simulations utilized do not typically involve social
or psychological behaviors as Vacate does).
Each simulation must be performed in a logical
sequence. Each expert must be consulted at the
right stage of design. Non-quantitative goals and
factors must be articulated and balanced with
quantitative goals and factors. Generally, each
simulation application is run by a different expert,
and this brings to the table a large group of
diverse players. It is generally understood that
architects play the role of director. Architects
work with all of the different experts; they syn-
thesize the different simulations into a cohesive
design process; they direct multiple digital pro-
cesses and combine them with the architectural
vision for the building. This involves judgment
as well as integration of data. The architect must
check the setup, constraints and assumptions
of each simulation and then negotiate trade-offs
between competing demands.
For example, as veteran airport architect
Derek Moore explains, in designing passenger
flows, one of the first steps is to conduct a
low-tech, analog survey in order to determine a
show-up profile for the expected passengers.
This show-up profile involves going to existing
airports and observing people as they arrive at
the airport and go through the check-in process.
Figures 10a (top), 10b (middle), 10c
(bottom). With Vacate, the crowd
behavior simulation involves multiple
types of passengers, each with dif-
ferent rules, to account for different
social and psychological behaviors.
Time step n
Fire emergency scenario at t=0.0 s.
RI exit
Fire pool
LI exit
Ramping re
Group leader
Disabled being
helped
Backing to the
seat to give way
to helper
Passenger has
jumped over seat
Two females are
following the leader
as a group
Two disabled passengers are
helped by helpers
+ sign means the passenger
has made exit selection
based on re hazard
Male
Female Disabled Group Passenger may jump over seat
Passenger Boarding Door
Fire emergency scenario at t=2.65 s.
Fire emergency scenario at t=9.85 s.
Fire emergency scenario at t=38.70 s.
FD
Exit
C
B
Fire
Path
FD
GB
D
Total
RD
Fire
RD
Occ
RD
Obs
A
Old
A
New
For each Particle, i,
use Objective Function
to select exit
Calculate Forward
and Repulsive Drives
Update position of
Particle i
Update Personal Best
Update Global Best
once all Particles move
Increment n and repeat
until Particle exits
Flow Split
Flow Merge
Queues
Accumulation
21 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

Information from the show-up profile is then fed
into a spreadsheet in order to study the flow of
passengers. Here, the spreadsheet becomes a
simple type of building simulation. Then, once a
schematic design is created with a correspond-
ing digital model, more sophisticated simulations
of passenger flow are conducted. The design
process involves multiple steps, simulations and
professionals, which the architect must direct.
[Figure 11]
In another example, according to Moore,
the process of designing airside movement
starts with creating the schedule for a sample
day of airport operations called the design day
schedule. [Figure 12] This process involves
multiple specialists and types of analysis, and
is sometimes conducted before the architect
is hired. Utilizing the design day schedule, the
architect produces a sketch layout based on
human experience and rules of thumb. A digital
model is created of the sketch layout and then
a sophisticated simulation application, such as
PathPlanner, analyzes local movements at typical
gates to identify points of congestion. [Figure 13]
After the gate sizes and locations are refined,
even more sophisticated and complex simulation
applications, such as SIMMOD or TAAM, analyze
movements across the entire airfield during
crucial periods of the design day schedule.
A third example involves baggage flows.
Here, in Moores summary, there are many
factors for the flow and each system has sub-
systems. The design must account for check-in,
security screening, human baggage handling,
automated baggage sorting, airplane loading and
unloading as well as carousel delivery. One inter-
esting issue is that it is necessary to distinguish
between the simulation of machines handling
baggage and the simulation of humans handling
baggage. Machines and humans both make
errors, but the errors of machines are easy to
compute, while the errors of humans are difficult
Figure 11 (top left). In this diagram
of passenger ow, each step requires
input from different specialists, and
multiple digital simulations are run
to evaluate performance.
Figure 12 (top right). The design day
schedule incorporates changes in
the number of passengers over time.
Figure 13 (bottom left). Based on
ight arrival and departure informa-
tion, along with data about passenger
movement and processing, airplane
movement at the gate is simulated
and visualized.
Figure 14 (bottom right). The design
of baggage ows, like most aspects
of airport design, involves numerical
inputs, numerical outputs, digital
simulations and models such as this
graph of passenger ow and baggage
ow over time.
BENJAMI N
Baggage Claim Demand Model
AA and AA Eagle Gate Requirements
with 20 minute inter-gate use times
Time on Claim (minutes)
Time of Day
Flow Split
Flow Merge
Queues
Accumulation
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
AA no RON
AX no RON
AA & AX w/RON
Current Gates
200.0
180.0
160.0
140.0
120.0
100.0
80.0
60.0
40.0
20.0
0.0
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
Passengers at Claim Bags on Claim
P
a
s
s
e
n
g
e
r
s
/
B
a
g
s

a
t

C
l
a
i
m
N
u
m
b
e
r

o
f

G
a
t
e
s
22 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

to encapsulate in a formula. As with the other
processes and examples, it is the role of the
architect to synthesize the different simulations
and then integrate them into a coherent building
design. [Figure 14]
In the architect-as-director model, there
is an implicit division of poweror a system of
checks and balancesbetween efficiency and
creativity. There is an assumption that human
intuition and creativity is separate from compu-
tation. The director must monitor and tame the
computer, while the computer must verify the
choices made by the director.
But perhaps the separate processes could
be integrated. Perhaps efficiency and creativity
could be part of the same computational model.
Perhaps subjective criteria could be introduced
into the optimization routines themselves. Instead
of optimizing different criteria independently, a
single routine could solve for multiple criteria at
the same time.
This would involve a well-known approach
called multi-objective optimization, common in
fields like product design and aircraft design.
While this approach seems well-suited for the
multiple and diverse demands of architectural
design, it is currently under-utilized. Even less
explored are digital processes that combine sub-
jective criteria with objective technical criteria
in the same optimization routine. In this type of
process, multi-objective optimization could be
used to combine objectives such as atmosphere,
aesthetics and program with objectives such as
structural performance, circulation and baggage-
handling efficiency. Here, the algorithm would
integrate many of the desired features of the final
design, and the creativity of the design process
would involve designing objectives and design-
ing experiments rather than simply designing
solutions. In other words, the role of the architect
would involve designing the problem. Rather than
focusing on form and performance in an alter-
nating sequence, the architect could focus on
creating the potential design spacethe complex
topological surfacefor the overall project.
OPTIMIZING CREATIVELY

Anthony Radford and John Gero, in a book titled
Design Optimization in Architecture, Building,
and Construction, study the application of
Figures 15a (top), 15b (bottom). In
this airport roofscape experiment by
GSAPP student John Locke, design
permutations are based on precise
design objectives, but the results are
more open-ended and exploratory
than those of typical optimizations.
23 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

Pareto efficiency for architectural design.
Compared to other optimization techniques,
Pareto optimization is more realistic and useful
for design, the authors conclude, because it
allows subjective criteria to be taken into ac-
count.
7
Radford and Gero are mainly referring
to the way that judgment can be applied to select
a single design from the set of optimal designs
on the Pareto frontier.
An extension of the argument might involve
using optimization to explore rather than to
exploit: to search creatively within wide potential
design spaces rather than hone in efficiently
within narrow potential design spaces. In other
words, perhaps we could use algorithms for
invention, and not simply for cold-blooded
efficiency. We could optimize creatively. While
optimization is most often used to locate a
single global maximum, its fundamental algo-
rithms could be equally proficient at identifying
multiple local maximums. So in a workflow for
exploration rather than exploitation, the process
could search for several novel, high-performing
designs rather than searching for a single best-
performing design.
For example, in the initial massing design
for an airport, an automated optimization
routine could be set up to explore different
configurations of program, circulation and
an all-encompassing space-frame roof. The
geometry and the objectives of the optimization
process could be calibrated to generate many
interesting massing arrangements that meet the
design objectives, including a few that are very
unusual. [Figures 15a, 15b] Or in the schematic
design for mixed-use towers, an automated opti-
mization routine could be created to investigate
a very large number of connection configura-
tions. Here, the geometry and objectives could
be designed to search for options that are
structurally stable, provide striking public sky
lobbies, link to one another through bold angles
and lack redundancy. Beyond computing all of
the complex relationships, the digital process
could be helpful to generate a wide variety of
geometric options and precisely indicate the
benefits of each option. [Figures 16a, 16b]
In both examples, the computational
process expands the design options rather than
narrows them. Optimization yields results that
can be understood as creative solutions to the
design problem. More generally, this design
process could be valuable in order to navigate
design spaces that are not immediately legible,
or to step outside of the limitations of linear
thinking, or to identify new regions of perfor-
mance worthy of further study. To put this in
terms of the multi-armed bandit problem used
by statisticians: pulling the unknown lever may
have value in itself, beyond winning the jackpot,
because it will provide new information, and
information is intrinsically valuable.
DEMOCRATIZING DESIGN

Beyond efficiency, and even beyond creativity,

we might use algorithms and digital processes
to debate values. Any version of multi-objective
optimization and any process of evaluating a
parametric model reinforces the need for clearly
defining design objectives (also called fitness
criteria). And design objectives are values: they
are the goals and desires of a project that in-
volve judgment and beliefs, outside of efficiency
and computation. All algorithms contain values,
but this is especially true for the algorithms of
design objectives.
The relationships between different
values, and the prioritization of these values,
are explicitly defined in the digital model, and
although they might be buried and hidden, they
are there. It is difficult to set up a parametric
model without values. And it is impossible to run
an optimization process without values. If the
values of a project could be exposed and stated
in plain language, then the digital model could
become a platform for debate and discussion.
In many architectural projects, interdisciplinary
members of the design team are now finding
ways to structure discussions about values
around digital parametric models. Perhaps this
process could be extended. These discussions
could easily encompass a wider audience and
a larger collective beyond design professionals.
They could involve input from a broader public,
in a structured and productive format. Artists,
philosophers, residents and citizens of all back-
grounds could join the discussion and debate.
In many ways, current digital design
processes already lend themselves to this
involvement. The process of setting up the
digital model could be considered a location
for discussion. For example, in the design of a
new tower, should the number of floors be fixed,
or dependent on the amount of public space
at ground level? This question clearly involves
values that could be discussed and debated
by a wide public, and its answer would clearly
affect the digital model. Similarly, the process
of tuning the digital model could be consid-
ered a location for discussion. For example,
through a parametric model, one constitu-
ency could argue for increasing environmental
performance even if cost also rises. Another
BENJAMI N
24 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

constituency could argue for an iconic building
entrance even if it decreases environmental
performance. The model could be tuned in
different ways according to different values.
Different design permutations would reflect
these different values. The sliders of the
parametric model could be tools to negotiate
between competing demands and competing
constituencies. [Figure 17]
In this sense, the digital model becomes a
platform for an open and inclusive process of
design. It widens participation rather than nar-
rows it. It democratizes design.
Pareto may have foreshadowed this in
his original example of bread and wine. In this
case of a societys distribution of resources,
a Pareto frontier could be a site for debate
about values. It could be a starting point for
discussing design permutations and making
trade-offs. Since all design permutations on the
Pareto frontier are optimal, values and judg-
ment are required to choose between them. The
act of selecting a single design from the set of
all Pareto-efficient designs could be understood
as the exact location where computation meets
designas well as where computation meets
society, politics and even culture.
If the objectives of greatest bread and
greatest wine were supplemented by a third
objective of greatest number of books, then a
new Pareto frontier would be established. Thus
alternate Pareto frontiers could be generated
through different values. The computational
model and the multiple possible Pareto frontiers
could be a very information-rich site for debate
about prioritizing values.
BEYOND

In 1916, ten years after his groundbreaking

work on economics, Vilfredo Pareto published
a completely different book entitled The Mind
and Society: A Treatise on General Sociology.
8

Pareto had resigned as Chair in Political
Economy at the University of Lausanne. He
had tired of the over-simplification involved in
economic theory and had become disillusioned
with the terrible track record of economic
1
2
3
4
5
maximize connections
maximize stability
minimize circulation
maximize connections
minimize redundancy
minimize public path
maximize stability
Typical Tower Simple, Stable
Tower
Highly Networked
Unstable Tower
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
10
8
0
0m
384m
960m
16/9/16/9/16
9/16/9/16/9
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
5
2
0
50m
96m
720m
16/4/16/4/16
3/16/7/16/7
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
8
0
0
224m
367m
1,104m
16/0/16/15/16
14/16/3/16/9
maximize connections
maximize stability
minimize circulation
maximize connections
minimize redundancy
minimize public path
maximize stability
Typical Tower Simple, Stable
Tower
Highly Networked
Unstable Tower
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
10
8
0
0m
384m
960m
16/9/16/9/16
9/16/9/16/9
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
5
2
0
50m
96m
720m
16/4/16/4/16
3/16/7/16/7
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
8
0
0
224m
367m
1,104m
16/0/16/15/16
14/16/3/16/9
maximize connections
maximize stability
minimize circulation
maximize connections
minimize redundancy
minimize public path
maximize stability
Typical Tower Simple, Stable
Tower
Highly Networked
Unstable Tower
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
10
8
0
0m
384m
960m
16/9/16/9/16
9/16/9/16/9
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
5
2
0
50m
96m
720m
16/4/16/4/16
3/16/7/16/7
Inputs
Outputs
connections
redundancy
short members
stability
public circ.
total height
8
0
0
224m
367m
1,104m
16/0/16/15/16
14/16/3/16/9
Figures 16a (above left), 16b (above
right). In this mixed-use towers
experiment by GSAPP student Danil
Nagy, the digital process offers a tool
to creatively explore design options:
it helps lter an overwhelming
number of possible congurations,
identies multiple local maximums
and provides geometric and analyti-
cal data about each one.
Typical Tower
maximize connections
maximize stability
minimize circulation
Inputs 16/9/16/9/16
9/16/9/16/9
Outputs
connections 10
redundancy 8
short members 0
stability 0m
public circ. 384m
total height 960m
Simple, Stable Tower
minimize public path
maximize stability
Inputs 16/4/16/4/16
3/16/7/16/7
Outputs
connections 5
redundancy 2
short members 0
stability 50m
public circ. 96m
total height 720m
Highly Networked
Unstable Tower
maximize connections
minimize redundancy
Inputs 16/0/16/15/16
14/16/3/16/9
Outputs
connections 8
redundancy 0
short members 0
stability 224m
public circ. 367m
total height 1,104m
possible designs
1 17 = 17
2 17
2
= 289
3 17
3
= 4,913
4 17
4
= 83,521
5 17
5
= 1,419,857
. . .
10 17
10
= 2,015,993,900,449
connection congurations
inputs
0 7 14
1 8 15
2 9 16
3 10
4 11
5 12
6 13
#5
#16
#16
#16
#16
#16
#12
#6
#7
#8
25 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

predictions. He had abandoned the fundamen-
tal economic tenet that people act logically.
Instead, Pareto now believed that people are
non-logical but believe they are logical. In his
new book, Pareto turned from economics and
equations to more complex theories of behavior,
and to the field of sociology. It was as if he was
saying that his earlier charts and his Pareto
frontier needed to be re-evaluated. There
needed to be more. Paretos simple mathemati-
cal models were no longer sufficient to bear the
weight of a multi-dimensional world. It was clear
that the models would have to be transformed.
Perhaps the same is true for our early
versions of digital processes in architecture.
Perhaps they need to be re-evaluated and
transformed. Perhaps there needs to be more.
The question is how much of the original model
will remain, and whether the new framework will
look anything like the one which has been so
influential and has taken us so far.
ENDNOTES

1. Pareto, V. (1906) Manuale di economia politica, Milan,

Societ Editrice.
2. Eitan Grinspun, from a transcript of the Columbia
Building Intelligence Project Think Tank, New York,
February 18, 2011.
3. Maeda, J. (2007) Forword, in Reas, C.; Fry, B. (eds)
Processing: A Programming Handbook for Visual
Designers and Artists, Cambridge, Mass., MIT Press,
p. xix.
4. Design Optimization for Human Well-Being and Overall
System Performance, Gino Duffett and Hiram Badillo
(APERIO Tecnologa en Ingeniera, Barcelona) and Sylvain
Carbe and Arne Kiis (AnyBody Technology, Aalborg East,
Denmark).
5. Xue, Z.; Bloebaum, C. L. (2009) Human Behavior
Effects on Cabin Conguration Design Using VacateAir,
47th AIAA Aerospace Sciences Meeting and Exhibit,
Orlando, Florida, AIAA-2009-0042. Also, Xue, Z. (2006)
A particle swarm optimization based behavioral and
probabilistic re evacuation model incorporating re
hazards and human behaviors, Master Thesis, Mechanical
and Aerospace Engineering, State University of New York
at Buffalo.
6. Material for this section is based on discussions with
and writing by Derek Moore, Associate at Skidmore,
Owings & Merrill, and an expert on the design of airports.
7. Radford, A.; Gero, J. (1988) Design Optimization in
Architecture, Building, and Construction, New York, Van
Nostrand Reinhold.
8. Pareto, V. (1916) Trattato di sociologia generale,
Florence, G. Barbra.
Figure 17. In this hypothetical re-
design of the Javits Center in New
York by GSAPP students Muchan
Park, Patrick Cobb and Miranda
Romer, the parametric model
structures a trade-off between two
competing values: new public space
on the roof versus lightwells that
provide natural light for the interior.
BENJAMI N
26 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

The identity of the architect is

largely built upon her or his ability
to author design solutions that
satisfy pragmatic concerns while
also capturing the imagination by
producing unique visual and spatial
experiences. Pragmatic concerns
are usually well dened and there-
fore can be solved with a high
degree of certaintythe respec-
tive design space is relatively
narrow, more quantitative, and pos-
sible to dene algorithmically. The
creation of unique experiences, by
contrast, is ambiguous, it relies on
inference and an indirect connection
between an architects knowl-
edge and own lived experiences
and how well they can anticipate
responses from users. The result
is never fully known or predictable.
In this case, the design space is
difcult, if not impossible to dene
algorithmically; it is highly qualita-
tive and therefore better suited to
human judgment.
While this portrayal of design
space as either quantitative or
qualitative is perhaps oversimpli-
ed, as these two threads are
more typically intertwined, it
serves to highlight the challenge
of capturing the full range of archi-
tectural design intent within digital
workows. With more and more
steps in these workows being
driven by design, analysis and
performance algorithms authored
by anonymous programmers,
the identity and authorship of the
architect comes into question. If
the architect has been increas-
ingly displaced by technologically
mediated processes over a long
time, the expanded realm of
digital workows transforms this
in historically new ways.
David Benjamin embraces the full
use of such techniques in his prop-
osition that both qualitative and
quantitative criteria be contained
within design algorithms. In his ve
scenarios for designing design,
automating the process of design
through discrete rules poses no
threat to human creativity nor to
professional identity and instead
offers ways to redirect digital
techniques in architecture. Beyond
the well-known applications of
simulation, analysis and optimiza-
tion techniques that target the
quantitative performance of select
parts of buildings like structure
and environmental behavior, the
examples Benjamin cites cross
over to human behavior and
comfort. They offer a glimpse into
new digital workows for architec-
ture that integrate qualitative and
quantitative design criteria.
Buildings are physically discrete
assemblies of parts in which the
ultimate decisions that deter-
mine which parts to use and
the relationships of these parts
are unambiguous. As such, the
process of decision- makingthe
design spacecan theoretically
be dened exclusively through
algorithms that quantify and orga-
nize what is assumed to be the
messy act of creativity.
1
Together
with the multi-objective optimiza-
tion process required to explore
the myriad relationships and
trade- offs between the outputs
of these algorithms, this results
in a workow where all human
decisions are ltered through
computation. For Benjamin, human
intuition and judgment occur when
designing the design space of a
problem, by choosing the inputs
and evaluating the outputs to an
algorithm but also by designing
the algorithm itself. This, then, is
not seen as a reduction of author-
ship; by focusing exclusively on
the design space as the locus for
decision-making, algorithms are
positioned as creative tools that
expand the design capabilities of
architects.
By designing the algorithm, the
relationship between constraints
(to control possible design
options) and variables (to explore
possible design options) can
become an integral part of the
architects overall design intent.
Several case studies presented
in this book demonstrate the
emergence and use of architect-
authored algorithms. These range
from small-scale projects like the
the net sculptures presented in
the contribution by Craig Schwitter
and Ian Keough of Buro Happold,
where custom algorithms were
AUTHORSHI P

EDITOR S NOTES
27 DESI GN ASSEMBLY I NDUSTRY DESI GNI NG

used for all aspects of the project

development, to large-scale
projects like the Museo de Acero
presented by Shane Burger of
Grimshaw Architects, where cus-
tom algorithms were created to
rene the design of the faade. By
computing the social, the complex
actions and responses of future
users to building program condi -
tions can be assessed through
historical data and statistical prob-
ability, in order to better inform
the heuristic processing skills of
human experts. By designing the
problem, multiple quantitative and
qualitative objectives in a complex
building can be set up in relation
to each other in a single algo-
rithmic process, such as in the
above-mentioned multi-objective
optimization. This is a process
that Benjamin has been exploring
in his design research studios at
Columbia University over the past
several years. As an extension of
this work, optimizing creatively
repurposes optimization proce-
dures. Typically used to rene
very specic attributes of a design
with regard to narrowly focused,
closely related objectives, these
procedures can be adapted to
explore more open-ended design
possibilities, by juxtaposing seem-
ingly unrelated objectives within
a single optimization routine. Far
from being random or arbitrary,
these juxtapositions express the
creative choices of the architect,
introducing an aspect of playful-
ness into an otherwise extremely
logical process. They could be
seen as part of a lineage of proce-
dural experiments extending from
the musical chance operations
of John Cage to the drip painting
techniques of Jackson Pollock.
Perhaps the most far-reaching
of Benjamins ve strategies
for designing design is the use
of algorithms to democratize
designto shift the culture of
architecture from prioritizing
the value of a single author to
placing greater value on the team
of authors, who each contribute
to the design of any building. As
Benjamin notes, design processes
that are structured through algo-
rithms require clearly articulated
objectives and parameters that
can be understood, discussed and
debated by many. These param-
eters, he suggests, can play the
role of agreed-upon design prin-
ciples that drive possible design
solutions. By turning the black
box of computation inside out and
exposing the myth of creativity
often associated with the most
successful architects, the values
and biases of both programmers
and designers can nd a common
platform to rene authorship and
subsequently evolve the next gen-
eration of digital workows.

1. William J. Mitchell, elaborating on

the articulation of a specic language
of terms and explicit formal grammars,
created a theoretical framework for
early versions of computer-aided
design in anticipation of the impact that
computer technology would have on
architectural design. See Mitchell, W. J.
(1990) The Logic of Architecture: Design,
Computation, and Cognition, Cambridge,
Mass., MIT Press.