Multi-Agency as a Computational Strategy for Design Exploration
PhD Proposal | Kaustuv De Biswas | Massachusetts Institute of Technology
Advisor: Prof. George Stiny

With the onset of the ‘Giga-Hertz' generation, there emerges a tremendous possibility in the realms of computation - more complex tasks are being tackled and new paradigms of computation are evolving. Multi-Agency (MA) as a computational strategy is a particularly promising paradigm for conceptualizing, designing and implementing computational architectures. The capacity of a single agent is limited by its knowledge, its computing resources and its perspective. This ‘Bounded Rationality' (Simon 1957) is one of the underlying reasons for creating problem solving organizations. The domain of Design is particularly complex, large and unstructured. The strategy of solving complex problems in such a domain may be addressed by developing multiple functionally specific and modular components that are specialized at solving a problem aspect. Taking advantage of the advances in other domains like Computer Science, Artificial Intelligence and Operations Research, my proposal aims at relating MAs to design exploration and revisiting some of the difficult but interesting issues specific to the Design Computation. ‘Parallel' and ‘Layered' computational architectures, which can model key mechanisms in Design such as ‘Reflection' are of particular significance to the framework of this research and may potentially lead to the implementation of this theory within the design discipline. Generative systems augmented with such reflective layers would open up exciting new possibilities. The distributed, dynamic and open character of MA may also promote the possibility of engaging the designer into meaningful dialogues with computational tools, allowing a richer exploration of design spaces and making the framework for discovery broader and more value added.

The concept of ‘Parallelism', which is closely associated with the ideas of ‘Multiple' and ‘Distributed', is a powerful one. The concept is widely applicable to a wide range of systems – from Natural Evolution, to Electrical Circuits, to Supply Chain Management. The common feature of all these systems is the ‘Doctrine of Choice'. Choice relaxes dependence on any single pipeline. Computational Systems in the realms of design generally have a linear structure. I believe that this linearity is fundamentally a ‘weak' structure because of its dependence on a single pipeline. Multiplicity is the core thesis of my proposal. It models several layers of intelligence into the design regime and reduces the dependence on any single layer.

From a personal perspective, I have found my beliefs echoed in the works of several principals in MIT. I rest my arguments on the work of the two of my professors who deeply influenced the philosophical foundations, which are at the core of my current research – George Stiny and Marvin Minsky. Stiny's idea of a single shape having multiple structures or topologies debases the use of rigid singleton computational representation systems in the design domain and prompts a fresh look at basic systems of representation. Minsky's idea of distributed cognition (mind as a society) re-emphasizes the importance of multi-agency and multi-representation as a tool for understanding difficult and ambiguous problems. In this proposal, I have tried to build on this belief and show how some of the key issues in Computation today can be re-investigated in this light.

[Framework]

The divorce between the ‘ Real' and its ‘Representation ':

Knowledge representation and Multi-Agency

“To solve really hard problems, we'll have to use several different representations. This is because each particular kind of data structure has its own virtues and deficiencies, and none by itself would seem adequate for all the different functions involved with what we call common sense.” - Marvin Minsky

Before we compute, we must ‘represent' the knowledge and all ‘representations' are fundamentally a surrogate for the entity itself, not quite embodying all the principles of the real physical entity. Such fragmentation and incompleteness inherent in any representation scheme makes them almost synonymous to strategies of looking at the world. Such strategies are always closely linked with certain task domains, which are concerned with only certain aspects of the entity. Switching between strategies of ‘representation' can sharply change views about the entity. In the realm of design, the designer in his explorative mode flows from one representation to the other, switching from one strategy to the other. The rigidity of a single logical structure of representation is shattered.

Computational Design Strategies in use today do not quite allow the same fluid platform for the designer. There is always a lure of creating computationally manageable representation, which is unambiguous, rigid and monotonous. The consequence, however, is the divorce of many salient principles of the real construct from the representation. The reversal, i.e. the creation of a real construct from these abstractions, would result in a ‘deficiency of' or ‘insensitivity to' the initially ‘thrown away' principles. Human interventions/experience would become necessary to accommodate such deficiencies. Stiny mentions how the moment we represent a shape in any rigid computational structure, we lose opportunities and interesting possibilities.

To retain the richness and indeed the ambiguity of the ‘real' I propose a multi-agency, which will represent the world in several independent ways. This will allow ‘switching' between ‘ways' of looking or understanding the world. Encapsulating the essence of ‘Ambiguity' and ‘Context' are distinct interesting possibilities within this framework of knowledge representation.

‘If you can't represent an idea in multiple ways, you don't have an idea' – Marvin Minsky

Stiny's ‘ Emergence' and Minsky's ‘B-Brain':

Reflection and Multi-Agency

Computation today lacks the idea of ‘reflection', which is fundamental to the mechanism of design. Stiny's Shape Grammars show how during the production of a design, there often ‘emerges' new possibilities that are not ‘intended' or ‘defined' in the system a priori. Without a layer of reflection, it becomes nearly impossible to figure out such emergence. The production system is concerned with the local rules, while the reflection system is more concerned with larger and more holistic goals and principles of the design. A Multi-Agent system has the potential to allow for such ‘Reflective Layers of Computation'.

“There is one way for a mind to watch itself and still keep track of what's happening. Divide the brain into two parts, A and B. Connect the A-Brain's inputs and outputs to the real world – so it can sense what happens there. But don't connect the B-brain to the outer world at all; instead, connect it so that the A-brain is the B-brain's world!” – Marvin Minsky

Design as a Complex Adaptive System:

Superposition of Principles and Multi-Agency

A multi-agency creates the possibility of a superposition of multiple principles or rule-sets. Superposition is an interesting phenomenon, which often results in intricate and complex wholes from much simpler parts. The bee really doesn't notice the holistic development that occurs simultaneous with the local operations it performs during ‘nest-building'. Each bee probably has limited local rules in its repertoire, and as an individual does ‘simple' things. However the aggregate work of the entire swarm amounts to a complex, intricate form, which no single agent could have possibly achieved. Over time, there seems to be perpetual flux, but somehow the overall pattern persists. Our cities are great examples of superimpositions. A wide array of agents from different layers of society follow their own sets of rules and perform local changes to the city form. However, mysteriously even though no single constituent remains in place, the city still persists. What enables cities to retain their coherence despite continual disruptions and a lack of central planning? Can the process of design be simulated over time as the work of a multi-agency? Can the process of design be established as a dialogue between the designer and multi-agency capturing multiple principles of design?

[Implementation]

A Multi Grammar-Set Generative System with the Principles of Context and Opportunism:

Though a deeper investigation of the theoretical model is necessary to be comprehensive, to ground the framework of the proposed research and highlight the possibilities of the paradigm I propose a prototype Multi-Agent System with a grammar-based production system that will drive a design exploration. Some of the interesting possibilities that I envision include:

1.Multiple Representation Schemes

-Allowing multiple topologies/structure for the same shape will capture notions of Ambiguity.

-Context may drive/select relevant representation schemes.

2. Reflective Layers

-Aligned with Donald Schon's notion of ‘reflection-in-action', local agencies will be focused on generation, while the reflective agents would look at higher level ‘intentions' or ‘interests' (such as ‘Linearity', ‘Symmetry', ‘Balance') and ‘opportunities' (such as ‘a courtyard space'). These reflective agents would trigger certain focuses and the selection of particular rule sets for the local agency to use.

3. Contexts

-The idea of context is particularly interesting and will play a pivotal role in resolving ambiguities.

-Contexts will be almost like platforms over which the agencies work. Contexts will have associated rule-sets.

4. Superposition of other Principles/Processes

- The interesting possibility of grammatical generation within certain environments (External physical or social forces).

- The possibility of mixing other prominent computational processes in the same environment. For example ‘Searches' might be employed to quickly figure out more holistic goals or targets and the rule-based generative system can be employed to reach that target.

- The possibility of objective Performance-Evaluation of interim states within a Grammatical Generation.

Establish Design as a ‘Process' rather than a ‘State':

During my graduate studies I have explored design as Search, as Rule Based Generation, as Parametric Constraint Propagation, as Mapping and in recently I have started working with Multi-Agent systems . From my experiences I have felt that even though I was working with ‘processes', they are purely ‘Computational' and the products were simply ‘Architectural States'. Please refer to my work on the ‘Generation of Ottoman Mosque Complexes' (Portfolio), where I have designed an elaborate search technique to find ‘best' solutions algorithmically. Though the metaphor ‘Search' embeds in it the element of time, the Generated Complex is a finally a state. It ignores the effect of time or the interaction of the city fabric with it. The actual Ottoman Mosque Complexes were designed five centuries back and in most cases pressure from the surrounding city fabric has changed the way the ‘internals' of the complex behave. My aim is to build a system, which can capture these different dialogues, and have multiple ways of interpreting the world, be it a hierarchical mosque complex or a joint system of a facade or a flat and abstract grammatical exploration. The power of computation is finally to capture processes and not products. I want to re-emphasize that.

Multi-Agency as a Computational Strategy for Design Exploration PhD Proposal \| Kaustuv De Biswas \| Massachusetts Institute of Technology Advisor: Prof. George Stiny	.back
	[masters thesis]
\| The elephant tale \| MA's \| Tools \| Philosophy \| Minsky \| Stiny \|

With the onset of the ‘Giga-Hertz' generation, there emerges a tremendous possibility in the realms of computation - more complex tasks are being tackled and new paradigms of computation are evolving. Multi-Agency (MA) as a computational strategy is a particularly promising paradigm for conceptualizing, designing and implementing computational architectures. The capacity of a single agent is limited by its knowledge, its computing resources and its perspective. This ‘Bounded Rationality' (Simon 1957) is one of the underlying reasons for creating problem solving organizations. The domain of Design is particularly complex, large and unstructured. The strategy of solving complex problems in such a domain may be addressed by developing multiple functionally specific and modular components that are specialized at solving a problem aspect. Taking advantage of the advances in other domains like Computer Science, Artificial Intelligence and Operations Research, my proposal aims at relating MAs to design exploration and revisiting some of the difficult but interesting issues specific to the Design Computation. ‘Parallel' and ‘Layered' computational architectures, which can model key mechanisms in Design such as ‘Reflection' are of particular significance to the framework of this research and may potentially lead to the implementation of this theory within the design discipline. Generative systems augmented with such reflective layers would open up exciting new possibilities. The distributed, dynamic and open character of MA may also promote the possibility of engaging the designer into meaningful dialogues with computational tools, allowing a richer exploration of design spaces and making the framework for discovery broader and more value added. The concept of ‘Parallelism', which is closely associated with the ideas of ‘Multiple' and ‘Distributed', is a powerful one. The concept is widely applicable to a wide range of systems – from Natural Evolution, to Electrical Circuits, to Supply Chain Management. The common feature of all these systems is the ‘Doctrine of Choice'. Choice relaxes dependence on any single pipeline. Computational Systems in the realms of design generally have a linear structure. I believe that this linearity is fundamentally a ‘weak' structure because of its dependence on a single pipeline. Multiplicity is the core thesis of my proposal. It models several layers of intelligence into the design regime and reduces the dependence on any single layer. From a personal perspective, I have found my beliefs echoed in the works of several principals in MIT. I rest my arguments on the work of the two of my professors who deeply influenced the philosophical foundations, which are at the core of my current research – George Stiny and Marvin Minsky. Stiny's idea of a single shape having multiple structures or topologies debases the use of rigid singleton computational representation systems in the design domain and prompts a fresh look at basic systems of representation. Minsky's idea of distributed cognition (mind as a society) re-emphasizes the importance of multi-agency and multi-representation as a tool for understanding difficult and ambiguous problems. In this proposal, I have tried to build on this belief and show how some of the key issues in Computation today can be re-investigated in this light. [Framework] The divorce between the ‘ Real' and its ‘Representation ': Knowledge representation and Multi-Agency “To solve really hard problems, we'll have to use several different representations. This is because each particular kind of data structure has its own virtues and deficiencies, and none by itself would seem adequate for all the different functions involved with what we call common sense.” - Marvin Minsky Before we compute, we must ‘represent' the knowledge and all ‘representations' are fundamentally a surrogate for the entity itself, not quite embodying all the principles of the real physical entity. Such fragmentation and incompleteness inherent in any representation scheme makes them almost synonymous to strategies of looking at the world. Such strategies are always closely linked with certain task domains, which are concerned with only certain aspects of the entity. Switching between strategies of ‘representation' can sharply change views about the entity. In the realm of design, the designer in his explorative mode flows from one representation to the other, switching from one strategy to the other. The rigidity of a single logical structure of representation is shattered. Computational Design Strategies in use today do not quite allow the same fluid platform for the designer. There is always a lure of creating computationally manageable representation, which is unambiguous, rigid and monotonous. The consequence, however, is the divorce of many salient principles of the real construct from the representation. The reversal, i.e. the creation of a real construct from these abstractions, would result in a ‘deficiency of' or ‘insensitivity to' the initially ‘thrown away' principles. Human interventions/experience would become necessary to accommodate such deficiencies. Stiny mentions how the moment we represent a shape in any rigid computational structure, we lose opportunities and interesting possibilities. To retain the richness and indeed the ambiguity of the ‘real' I propose a multi-agency, which will represent the world in several independent ways. This will allow ‘switching' between ‘ways' of looking or understanding the world. Encapsulating the essence of ‘Ambiguity' and ‘Context' are distinct interesting possibilities within this framework of knowledge representation. ‘If you can't represent an idea in multiple ways, you don't have an idea' – Marvin Minsky Stiny's ‘ Emergence' and Minsky's ‘B-Brain': Reflection and Multi-Agency Computation today lacks the idea of ‘reflection', which is fundamental to the mechanism of design. Stiny's Shape Grammars show how during the production of a design, there often ‘emerges' new possibilities that are not ‘intended' or ‘defined' in the system a priori. Without a layer of reflection, it becomes nearly impossible to figure out such emergence. The production system is concerned with the local rules, while the reflection system is more concerned with larger and more holistic goals and principles of the design. A Multi-Agent system has the potential to allow for such ‘Reflective Layers of Computation'. “There is one way for a mind to watch itself and still keep track of what's happening. Divide the brain into two parts, A and B. Connect the A-Brain's inputs and outputs to the real world – so it can sense what happens there. But don't connect the B-brain to the outer world at all; instead, connect it so that the A-brain is the B-brain's world!” – Marvin Minsky Design as a Complex Adaptive System: Superposition of Principles and Multi-Agency A multi-agency creates the possibility of a superposition of multiple principles or rule-sets. Superposition is an interesting phenomenon, which often results in intricate and complex wholes from much simpler parts. The bee really doesn't notice the holistic development that occurs simultaneous with the local operations it performs during ‘nest-building'. Each bee probably has limited local rules in its repertoire, and as an individual does ‘simple' things. However the aggregate work of the entire swarm amounts to a complex, intricate form, which no single agent could have possibly achieved. Over time, there seems to be perpetual flux, but somehow the overall pattern persists. Our cities are great examples of superimpositions. A wide array of agents from different layers of society follow their own sets of rules and perform local changes to the city form. However, mysteriously even though no single constituent remains in place, the city still persists. What enables cities to retain their coherence despite continual disruptions and a lack of central planning? Can the process of design be simulated over time as the work of a multi-agency? Can the process of design be established as a dialogue between the designer and multi-agency capturing multiple principles of design? *[Implementation]* A Multi Grammar-Set Generative System with the Principles of Context and Opportunism: Though a deeper investigation of the theoretical model is necessary to be comprehensive, to ground the framework of the proposed research and highlight the possibilities of the paradigm I propose a prototype Multi-Agent System with a grammar-based production system that will drive a design exploration. Some of the interesting possibilities that I envision include: 1.Multiple Representation Schemes -Allowing multiple topologies/structure for the same shape will capture notions of Ambiguity. -Context may drive/select relevant representation schemes. 2. Reflective Layers -Aligned with Donald Schon's notion of ‘reflection-in-action', local agencies will be focused on generation, while the reflective agents would look at higher level ‘intentions' or ‘interests' (such as ‘Linearity', ‘Symmetry', ‘Balance') and ‘opportunities' (such as ‘a courtyard space'). These reflective agents would trigger certain focuses and the selection of particular rule sets for the local agency to use. 3. Contexts -The idea of context is particularly interesting and will play a pivotal role in resolving ambiguities. -Contexts will be almost like platforms over which the agencies work. Contexts will have associated rule-sets. 4. Superposition of other Principles/Processes - The interesting possibility of grammatical generation within certain environments (External physical or social forces). - The possibility of mixing other prominent computational processes in the same environment. For example ‘Searches' might be employed to quickly figure out more holistic goals or targets and the rule-based generative system can be employed to reach that target. - The possibility of objective Performance-Evaluation of interim states within a Grammatical Generation. Establish Design as a ‘Process' rather than a ‘State': During my graduate studies I have explored design as Search, as Rule Based Generation, as Parametric Constraint Propagation, as Mapping and in recently I have started working with Multi-Agent systems . From my experiences I have felt that even though I was working with ‘processes', they are purely ‘Computational' and the products were simply ‘Architectural States'. Please refer to my work on the ‘Generation of Ottoman Mosque Complexes' (Portfolio), where I have designed an elaborate search technique to find ‘best' solutions algorithmically. Though the metaphor ‘Search' embeds in it the element of time, the Generated Complex is a finally a state. It ignores the effect of time or the interaction of the city fabric with it. The actual Ottoman Mosque Complexes were designed five centuries back and in most cases pressure from the surrounding city fabric has changed the way the ‘internals' of the complex behave. My aim is to build a system, which can capture these different dialogues, and have multiple ways of interpreting the world, be it a hierarchical mosque complex or a joint system of a facade or a flat and abstract grammatical exploration. The power of computation is finally to capture processes and not products. I want to re-emphasize that.