Arlati1 ,Valeria Bottelli1, Christian Fogh1, Maurizio Tirassa2.
1. Politecnico di Milano, Facoltà di Architettura, Dipartimento
di Programmazione Progettazione e Produzione Edilizia, Via Bonardi 3,
20133 Milano, Italy
tel. -39-2-2399-5125 fax. -39-2-2399-5150 e-mail: firstname.lastname@example.org
2. Università di Torino, Centro di Scienza Cognitiva, Via Lagrange,
3, 10123 Torino, Italy.
tel. -39-11-549475 fax. -39-11-549653 e-mail: email@example.com
paper presents on-going research aimed at the understanding and support
of process knowledge in architectural design, from early and not sufficiently
defined, to satisfactorily-defined phases. Today, technical, planning,
management and environmental issues have created a scenario of such
complexity that traditionally efficient control tools (e.g. technical
manuals) are inadequate and there is a demand for new, integrated instruments
to handle the decision process underlying architectural design. We assume
design as a recursive and incrementally specified intentional planning
activity, involving goals, constraints and their relationships. The
essence of architectural design is thus encapsulated in the continual
recursive transformation of the initial model, in order to map the desired
state onto the enacted one. On the basis of this concept of design we
describe the model of an environment aimed at progressively representing
the enlarging space of acquired knowledge, and at supporting the designer's
central role in the management of complexity.
knowledge-based design; case-based reasoning; design process control,
paper presents on-going joint research efforts in the areas of cognitive
science and architecture, aimed at the understanding and support of
Architectural design is a complex decision-making process, which is
generally referred to as an ill-defined problem area because of its
character: the influence of several related variables, most of which
vague and underspecified, and the continual and recursive evolution
of requirements and strategies.
are a number of factors which typically generate the complexity of the
architectural design and connected building process: the heterogeneous
nature of the knowledge domains and technical skills involved, the constraints
of building regulations, the simultaneous activity of an increasing
number of different subjects participating in the process, and finally
the density of interferences among decisions related to different domains
nature of this complexity can be summarised in the following features:
the poor and inadequate description of the object of the design process,
of which the intentional features expressed by the client are known,
but the nature of the actual technically articulated content of the
goals is unknown. A reasonable configuration of this object has to be
searched for by the design process, and its feasibility has to be verified
in front of a network of goals, the incidence of which is only partially
- the lack of a conventional representation system for architectural
values, able to declare without ambiguity the nature of knowledge-domains
assumed as the reference points of the design process; these values
are chosen cultural issues or technical and economical requirements
and are the expression of objective goals connected with resources optimisation.
- the lack of a conventional language able to describe and communicate
to a community the relationship between the assumptions of the different
knowledge domains composed in the design, i.e. the transfer of design
intentions into results -.
- the lack of a codified representation-system for the relationships
among the semantic, functional and topological components of a project
along its generation steps; for the interferences among the involved
- the lack of a conventional procedure to represent the decision making
- the widely prevalent specificity of design subjects, i.e. the unique
configuration of requirements, due to factors such as a very fragmented
market demand (e.g. Italy ), individuality of clients, unskilled in
their expression of a technologically qualified demand.
Design as a Decision Activity
is an intentional planning activity in which both the rules governing
the process and the required properties of the final solution are subject
to permanent refinement, substitution and review.
A decision taken in one area will necessarily impact on the configuration
of the process as a whole; therefore any sequential solution approach
seems to be necessarily inadequate.
In such a knowledge context, the decision-making activity requires the
adoption of an incrementally refinable model of reality, aimed at progressively
incorporating all possibly relevant issues.
In other words, in a context of "turbulent complexity" (Maldonado,
1992) - as specificity and variation are the main characters for architectural
design today- the decision-making process should abandon sectorial approaches,
to be founded upon the application of a methodology able to incorporate
the largest possible number of significant issues since the very first
heuristic phases of design, so as to define a preliminary solution model.
In this view, the design process may be seen as the planning of a path
among increasingly accomplished configurations of the preliminary model,
which develops as the involved issues win higher specification and integration
levels. The consequent increasing complexity should be validated by
incremental evaluation tools to ensure the truth-maintenance of the
network of relationships among constraints, rules and goals.
The observation of the skills of project managers has led to the conviction
that decision-making in design is an information intensive activity,
largely relying on the availability of quality-oriented and readily-accessible
information sources, supporting the experience and reasoning skills
of the decision-maker.
is to say that effective solutions to complex problems are seldom developed
from scratches and that decision makers build on knowledge and expertise
acquired in the solution of previous similar problems related to contexts,
or cases. Decision making, thus, requires refinement, mutual adaptation
of evaluation components and intuitive combination of solutions to previous
problems, leading to an incrementally configured hybrid solution model.
On the basis of these distinctive characters of architectural design
activity, we analyse the underlying process knowledge with the aim of
obtaining a cognitive model of the space of intention and decision-making
in architecture. Then this cognitive phase of the work leads to the
modelling of a set of tools aimed at supporting the design process from
the very beginning, aiding designers in the specification of the problem,
in the evaluation of alternative solutions, in the retrieval and adaptation
of relevant cases and in the overall representation of the decision-making
To accomplish these tasks is the goal of 'Patri-Arch', a system conceived
as a platform - better, as a wide comprehensive environment - of integrated
and evolving tools and agents, aimed at the description of the design
work in its process nature.
architectural design process
authors converge on a definition of the design process as a "global
activity occurring in a contemporary and non-linear evolution of the
requirements and design state through continuous reciprocal adaptation
" (Smithers, 1991)
This conception configures an architectural project as the progressive
definition of a model apt to solve the design subject's specific requirements,
based on the resources of the designer', producing a synthesis - the
final solution - of which the designer is the one responsible 'agent
of proposals', together with the cooperative coordination of the other
concurrent specialised domains.
It is during the evolution toward the intended level of quality - satisfaction
of requirements - that the designer carries out the full application
of his intelligence of correlations, which may be defined as his ability
to know, to describe and solve in a model the contents - and the possible
conflicts - of the interfering decision factors.
Building on the conception of design we have just outlined, we analyse
architectural design as a process path leading from an initial configuration
toward a desired situation by means of an incremental specification
of the goals, constraints, rules of evaluation and verification and
involved variables. In this view, design becomes a continuously recursive
- and revisable - decision path among increasingly detailed and developed
versions of the initial model.
The model represents and tackles the decision path both in its globality
and in its constitutive elements, i.e. steps/sequences of steps at different
abstraction levels, treated as unitary autonomous sub-processes.
We represent the process on three levels of abstraction:
- the first level comprises all the potentially acceptable solutions
("chosen", in an intentional model such as that of Cohen &
Levesque, 1990). This level regards the mapping of goals and constraints
(where goals are viewed as a subjective subset of constraints, i.e.
they are not environment-driven but intention-driven constraints). At
the outset, a certain number of goals and constraints are specified
by the designer. These allow the description of the initial configuration
of the model, which is necessarily incomplete and underdefined but already
presents the specific character of the 'creative' design activity: produces
a hybrid between the nature of abstract intentions and an object;
- the last level comprises the set of decisions/objects actually adopted
("intended", again in terms of an intentional model) which,
once taken together, make up the actual solution;
- the intermediate level comprises the space of all decision nodes explored
along the process making up the path leading from the space of "possible"
solutions to the final "intended" solution. Decision nodes
are viewed as the building blocks of the process. They are elementary
units flexible enough to govern the complexity of each single step,
at whatever level of abstraction, in the process (see description of
Galathea, Bottelli and Fogh, 1995).
This level is constituted by the decisions/objects needed to combine
the elementary units forming the adopted solution (second level) coherently
with the evolving map of goals and constraints (first level).
The executed solution represents the actual intention taken out of all
chosen possibilities. The intermediate levels are based upon an evaluation
system, in its turn. When reasoning on any sub-level, its higher and
lower boundaries are kept stable, while the space in between is modified.
The results achieved may eventually reflect on either boundary, and
the whole process may iterate on both the hierarchically superordinate
and subordinate levels. Thus, the whole process is recursively analysed.
This three-level model proves very useful because it may be equally
applied to the process as a whole, when designers need to look at a
project from afar, considering just the landmarks; and to any specific
phase or sub-phase when it is necessary to descend in the process details.
Consider the following example:
an architect, setting out on a new project assignment, is searching
for knowledge deriving from previous projects apt to help him/her in
the definition of a general layout. He/she may thus look back on the
knowledge acquired from a previous project, just considering general
basic decision nodes, i.e. viewing it in its globality. At this stage,
in fact, it is useless and uninteresting to retrieve knowledge pertaining
to specific decisions, e.g. the choice of floorings. At later stages
of the same project, instead, the architect might search for knowledge
suitable to aid him/her in the decision-making process of a specific
domain area (e.g. choice of structure or windows or floorings etc.)
and look back on the same previous project entering the desired level
of detail of specific lower order decision nodes.
The decomposition of decisions, based on the chosen evaluation system,
in such a model is therefore not pre-defined but leaves designers free
to explore different abstraction layers, according to the task in question.
Constraints are heterogeneous in their nature, origin and weight:
- nature: they may refer to extremely different domains, all of which
interact in the decision process. They may have quantitative or qualitative
nature, or both. They are an expression of the several specialist areas
contributing to a successful design solution. For instance, they may
refer to the areas of costs, of environmental issues, of aesthetical
values, of technical performances etc.
- origin: some constraints are context-dependent, i.e. they are variables
on which designers have no power of choice (legal domain, functional
issues, client requirements, site features etc.); others are subjectively
driven, i.e. they are a function of the culture, experience and values
of the designer; others again are an expression of both;
- weight: not all constraints have comparable importance in the determination
of the design process. Moreover, the weight of constraints may vary
and change considerably during a project.
Typically, some constraints are wholly unknown or disregarded at the
outset, but acquire importance during the process, or vice-versa. For
instance, incomplete information regarding the noise level of a building
site at the beginning of a project may induce an architect to give an
inadequately small weight to noise reduction techniques. Information
acquired during the process will modify the relative weight of this
variable in the mapping of constraints and consequently the project
will progressively incorporate noise reduction techniques such as insulation,
choice of windows etc.
: a key for design - decision activity
view design as a form of intentional planning. Schematically, an agent
can be described as an intentional system capable of analysing and evaluating
the current world situation in terms of a number of characters, assigning
each of them a causal role in the overall likeability of the situation
itself, and deliberating and acting in order to change one or more characteristics,
so as to make the situation more to his/her likings. (Pollock, 1992,
1993) By the likeability of a situation, we simply mean the extent to
which it is judged valuable by the agent.
this point of view, an agent exhibits the following basic set of cognitive
- the ability to analyse the current world situation; each agent will
subjectively analyse the situation according to his/her previous structure
of knowledge and goals; the analysis itself will in turn modify such
- the ability to evaluate the causal role of the characteristics, allowing
the agent to decide which of them can and should be modified in order
to bring about the desired changes in the overall situation;
- the ability to deliberate about the actions which are likely to bring
about these changes; this requires that the agent knows what its possibilities
for action are, how the world constrains such possibilities, etc., which
takes us back to the analysis of the situation.
Given these abilities, to make a plan means to decide how to transform
a certain world situation, by modifying one or more of its defining
characteristics, so as to make it expectedly more to the agent's likings.
To plan means to select, among the possible world of configurations
the expected likeability of which is higher than the current one, i.e.
the configuration exhibiting the most satisfactory expected value in
terms of costs/benefits. A distinction needs thus to be drawn between
what, in principle, an agent thinks he/she would like and what he/she
actually decides to (tries to) bring about, the latter obviously being
a subset (as large as possible) of the former.
The difference is that an agent will commit only to the latter subset;
we will use the term choice for any potentially likeable situation,
and the term intention for the one to which the agent is actually committed.
If an agent cannot achieve his/her goals (i.e., if he/she cannot satisfy
his/her intentions), he/she will possibly fall back to the larger set
of what would be likeable (i.e., of the previously discarded choices),
thus committing to a different course of actions.
agents cannot decide in advance all the details of their intended courses
of action, because they simply cannot foresee all the tiniest consequences,
side-effects, etc., of their plans. Furthermore, the world is dynamic,
i.e., it is likely to change before an agent has completed the execution
of a plan; this would make an excessively careful deliberation a waste
At each moment in time, therefore, a plan should be neither more nor
less detailed than is needed at the current stage of development. This
idea is known in the planning community as least commitment. (Weld,
1994) A plan can be viewed as a complex network of interdependent nodes
of knowledge and goals, large areas of which are initially blank and
will be progressively detailed as the agent proceeds in deliberation.
Parts of the network might be still underspecified even at the time
when other parts are already being executed.
The links of the network are the constraints (of space/time, knowledge,
priority, etc.) which the nodes impose to one another. A change in a
node may consist in a further development of its details, or in a real
revision; revisions may be due e.g. to information feedback from other
areas of the plan which have undergone further development or execution.
The consequences of a change in a node may propagate, through the different
constraint links, to other nodes, so that they may be accordingly modified
in their turn. The plan is therefore a continually evolving structure
of knowledge and intentions, which undergoes dynamical revision under
the action of internal as well as external pressure.
This model of planning applies straightforwardly to design. A project
may be viewed as a complex network of knowledge and decisions (although
these will not be actions in the usual, "behavioural" sense,
they will be decisions anyway), each of which has, or may have, an influence
on the others. To fully specify a project means to develop all the nodes
down to the desired level of detail. In this process, each piece of
knowledge and intentions may reverberate its effects to the others.
Starting from scratch, and from an analysis of the initial goals and
constraints, the designer's task is to transform an initial situation
into a more likeable one, by recursively modifying the appropriate characteristics
of the progressively evolving situation.
This conception of planning is crucial in our analysis of design and,
coupled with our conception of case-based reasoning, dealt with in the
following paragraph, constitutes the cognitive framework for the design
process model we propose.
an integrated environment for architectural design support
main paradigm for an integrated computer-based support environment for
architectural design consists in the opportunity of conceiving and representing
projects as networks of relationships which progressively assume their
identity, composing in a well-structured framework the designer's research
for an optimised solution.
In this conception the design activity is a process guided through the
following steps of evolution of knowledge contents:
designer's interpretation of the theme and requirements submitted by
- building project's feasibility verification:
- progressive identification of the fragments of solution that the designer
evaluates to be of mature formulation
- proposal for assembling solutions that configure a satisfactory synthesis
for the set of requirements,
- parallel activation of a verification procedure of their validity
and respect of regulations, or of constraints and goals given by the
- adaptation procedures allowing the solution to comply with the newly
formulated relationships between rules and goals, the description of
which has been increased;
- reiteration of the validation procedures of the solution in the assembling
of the different members in a general space configuration;
- control of the client's satisfaction degree.
further paradigm is to conceive the design solution as a network of
relationships among constraints and goals that takes shape in the dimension
of space and functions, the validity of which can be verified along
its same drawing up, although this form is only partially defined.
On the contrary, exactly because of its provisional and partial definition,
the validation of an in-progress solution can point out a new more satisfactory
direction of development which can save time and energies, requiring
a smaller and more acceptable sacrifice in abandoning a not yet accomplished
solution for more promising ones.
Another important point consists in enabling the practise of the designer's
intelligence to be supported in its ability to operate and compare a
number of solution models competitive among themselves, to choose the
most satisfactory technical and architectural language potentialities.
A final topic for the "designer's intelligence" is to see
a project as a phase in the definition of a process, the nature of which
is intimately connected with its material feasibility.
The essential requirements on the basis of which PatriArch is being
conceived are the following:
the ability to represent, operate, validate design solution models during
their progress, at the reached level of definition;
- to allow models' elaboration by different tools and agents at their
state of definition, and to add new acquired quanta of technical information
that can be assumed in the definition of the model;
- to describe the design models by the technical languages and reasoning
schemes currently assumed by human designers;
- to support by continuity the representation of the increasing levels
of definition of the knowledge contents implemented in the solution
- to support a well structured set of integrated evaluation tools;
- to store and retrieve relevant cases of process knowledge, i.e. fragments
of relationship among the design decision factors that have determined
the drawing out of previous successful design experiences;
- to activate the agents able to operate the " building regulations
application control", the "truth maintenance systems",
the "design procedures planning agents": that is to say the
entire population of tools and agents that, by successive generations,
enrich the design supporting environment.
have outlined a cognitive approach to architectural design based on
four main knowledge areas:
- the general complex, ill-defined context characteristic of design
- our approach to design as a recursive, incrementally defined process;
- our approach to design as an intentional planning activity;
We have thus proposed a model for design activity based on three levels
of abstraction applied to the process.
These three levels of abstraction develop into the model of an environment
aimed at supporting the design process.
E. (1992) L'informazione tecnica: supporto indispensabile alla qualità
del progetto. La "qualità globale" negli interventi
di manutenzione. Milano: Ediarch.Bottelli,
V., Drago, P., Fogh, C. (1994) Can preconditions of architectural projects
be considered as cases? Development of a case-based tool in: Advances
in knowledge-based building design systems, ed. by J. Pohl, Windsor
School of Computer ScienceBottelli,
V. Fogh, C. (1995) Galathea: a case-based knowledge navigator for the
support of the architectural design process, accepted paper at Intersymp
95 conference, Baden Baden, August 1995Cohen,
P.R., Levesque, H.J. (1990) Intention is choice with commitment. Artificial
Intelligence 42: 213-261Colajanni
B., De Grassi M., Di Manzo M., Naticchia B. (1991) Can Planning be a
Research Paradigm in Architectural design?, in: Proceedings of the 1st
International Conference on Artificial Intelligence in Design, EdinburghDe
Grassi M.,Di Manzo M. et al., ASA - Achitectural Symbolic Assistant,
CNR Research Project, Ancona, 1992Flemming
Ulrich, Case-Based Design in SEED System, in: G. Carrara and Y.E. Kalay,
"Knowledge-based Computer aided Architectural design", Elsevier,
T., Reale e Virtuale, Feltrinelli, Milano 1992.Pohl
Jens, A Distributed Cooperative Model for Architectural Design, in:
G. Carrara and Y.E. Kalay, "Knowledge-based Computer aided Architectural
design", Elsevier, 1994Pollock,
J.L. (1992) New foundations for practical reasoning. Minds and Machines
J.L. (1993) The phylogeny of rationality. Cognitive Science 17: 563-588.
Weld, D.S. (1994) An introduction to least commitment planning. AI Magazine
Winter 1994: 27-61.Smithers
T. et al., Design is intelligent behaviour, but what's the formalism?,
in (AI-EDAM) 4(2), 89-98Winograd,
T., Flores, F. (1986) Understanding computers and cognition. Norwood,