Gil_EVOLUTIONARY ARCHITECTURE - A FORM OF ARTIFICIAL LIFE_2000

MSc Virtual Environments Bartlett School of Graduate Studies University College London Theoretical Analysis in Virtual Environments 29 / 5 / 2000 Jorge Alberto Lopes Gil EVOLUTIONARY ARCHITECTURE: A FORM OF ARTIFICIAL LIFE Abstract This paper sets John Frazer’s work presented in An Evolutionary Architecture in the framework of Artificial Life by establishing links between both subjects at the theoretical as well at the empirical level. In his book John Frazer proposes a new kind of architecture created with an evolutionary design process, based on a Worldview of natural phenomena related to complex dynamical systems. This theoretical background is shared with the scientific field of artificial life. The key concepts of emergence, self-organisation, adaptation and evolution through selection and genetics found in natural systems are developed in the different projects and experiments. They are closely related to relevant projects of artificial life, using their techniques and tools like cellular automata and genetic algorithms. As a result, this new architecture produces models that share the properties of living organisms, becoming a form of artificial life. Furthermore, evolutionary architecture is shown as having in cyberspace its natural environment, one of information. As a payback to the scientific field John Frazer leaves a series of creative experiments, where the combination of various concepts and the creation of “tools for understanding” broadens the horizon of speculation in both artificial life and architecture. “Artificial life, or a-life, is devoted to the creation and study of lifelike organisms and systems built by humans. The stuff of this life is non-organic matter, and its essence is information: computers are the kilns from which these new organisms emerge.” [Levy, 1993, p.5] Introduction In his book An Evolutionary Architecture John Frazer reports on projects developed by himself, several collaborators and his students at the Architectural Association School of Architecture Diploma Unit 11. These projects are the body of a research on a new way of designing architecture, and ultimately of producing an architecture that while integrated in the physical environment behaves like a living organism in metabolic balance with the built environment [Frazer, 1995a, p.9]. The basis for the new evolutionary architecture that he proposes is the redefinition of the architectural design practice, by rethinking the concept of “blueprint” in architecture [Frazer, 1995a, p.11]. Initial inspiration to this change comes from the natural world and following a certain Worldview of our time it embraces new sciences that are concerned with cybernetics, complexity and chaos. John Frazer turns the design practice into a natural process of morphogenesis, considering the concept of “blueprint” used in natural sciences. The design practice leaves its deterministic role in the definition of the architectural object and instead the form is generated by the architectural concept. The architectural concept is information encoded in the “blueprint” as a set of instructions that don’t define the end result. Thus the object is not a result of a rigorous description, but it grows from the encoded set of instructions, similar to the DNA string in living beings. Information becomes the basis of this new architecture, the same as life, according to a-life scientists [1] . As a theoretical background we find the same references as a-life’s background: Alan Turing and John von Neumann. These two scientists shared the belief that information is the basis of natural phenomena and created conceptual computers, before the computer had been invented, that would be able to perform lifelike behaviour. Turing imagined the Universal Turing Machine capable of any kind of computation, thus expected to acquire intelligence [Woolley, 1992, p.66]. Being able to emulate any other machine whose behaviour could be determined by its state, and if we consider that the brain has a finite number of states [2] , this achievement seems plausible. The field of Artificial Intelligence was created. Von Neumann dedicated himself to the subject of life and the conception of the Self-reproducing Automaton. He would become the founding father of artificial life [Levy, 1993, p.17] based on this mental construct, which he designed first as a mechanical robot and then as a cellular model based on pure logic. The first cellular automaton had been created. Theoretical framework John Frazer’s research has been conducted during more than thirty years, since the end of the 60’s decade, in parallel with the discoveries and breakthroughs in the a-life field. John Frazer progressively integrates these advances in his work borrowing tools and concepts, which are applied to his new architectural design process. A-life is a broad field with different approaches and objectives, that covers several phenomena such as self-organisation, self-replication, emergent behaviour, evolution, learning, co-operation and ultimately the creation of life. The computer has the key role in projects that involve the emulation of natural phenomena. Its use by John Frazer is extensive, from modelling tools responsible for the representation of the forms, to generative tools based on the principles of cellular automata and evolutionary techniques [3] . The role of the computer is determinant in these because it allows the acceleration in space and time of the process of evolving different generations of an architectural concept [Frazer, 1995a, p.18]. These tools are directly imported from the field of alife to overcome the limitations of the current generative systems [Frazer, 1995a, p.14]. Unlike a-life that grows the complexity of systems from simple elements that go through different procedures of development, the generative systems are usually based on a vast and complex set of elements and rules used in a combinatorial way. Cellular automata and evolutionary techniques are based on the work of John Conway with cellular automata (CA), more specifically the Life game systems (CS) biomorphs [7] [6] [4] [5] , the development of genetic algorithms (GA) and classifier by John Holland and the use of morphogenesis by Richard Dawkins with the . CA bring emergence and self-organisation, GA bring evolution and optimisation and CS bring adaptation and learning. John Frazer covers with the projects a broad range of areas of the a-life field, making it difficult to place him in a specific area. His interest in the field is general, sharing most of the theoretical body behind a-life [Frazer, 1995a, p.55]. However he isn’t interested in bringing the discussion into the subject of Life itself. The a-life field is permanently related to that issue, being divided into two strands: the “weak claim” and the “strong claim” [Levy, 1993, p.290]. The first simulates natural phenomena, trying to provide an explanation or a working model of them. The second creates models that try to present some character of aliveness. Is a-life a form of life? The answer to this question is very problematic, since there is no clear definition for the concept of “Life” [Levy, p.6]. Some propositions have been made, like a crib sheet presented by Steen Rasmussen [8] at the second a-life conference, but it always stays in a delicate philosophical and even moral realm. This leaves Frazer in a particular position in respect to these two strands of a-life. He creates something new with the principles of a-life bringing him close to the creative impulse of the “strong claim”, but he isn’t concerned if his model responds to some criteria of life. On the other hand, his work isn’t just a model of natural phenomena like the major focus of the “weak claim”. The natural world is just a source of inspiration, not of aspiration. Anyway his model clearly is a form of a-life [Frazer, 1995a, p.103], the creation of an artificial organism that is originated from a process inspired in natural phenomena, may it be considered alive or not. Machines – tools for interaction Frazer has created a series of machines to implement the new design process. They are mainly new interfaces to make the development of an architectural concept more natural. Like the machines for modelling [9] , which make the workings of the computer software behind them tangible. Other machines are more than an interface for the design process: they can be considered forms of artificial life. The Universal Constructor [10] is a CA machine. Its structure is based on cells that communicate with their neighbours and with external users by means of light signals. The flow of information between all the participants in the process generates a result according to the rules programmed in the cells. This machine is an environment itself [Frazer, 1995a, p.49]. It’s a place where changes of state are negotiated between users and cells. The Universal Interactor [11] [12] is a machine with a series of inputs and outputs . They have the same function as the “detectors” and “effectors” in CS. These feedback elements are of various sorts, relating either to users or to the character of the physical environment. They establish the relationship between the machine and the surrounding system. Evolution comes from the combination of this characteristic with GA. The machine acquires the ability of adaptation to a living environment, like an organism operating in an ecosystem. It receives information from the environment and other organisms, processes this information, and acts accordingly. These two machines share the name “universal” with the Universal Turing Machine, and in fact they are a sort of universal computers, but for architecture alone. They are based on a simple process, to which some software with the rules is connected. By editing these rules you can obtain many different types of results. In this particular case architecture objects. These interactive machines are also relevant for enabling learning via positive feedback and a reward system. Depending on the responses to the actions, the software develops an understanding of the best reactions. They already represent a model of a-life, not just a tool for creation of form. These tools can be seen evolutionary architecture objects, due to their awareness of the environment and the user. The interactive machines behave much like “living” robots that in fact interact with the environment while performing a task. From an ecological metabolic point of view, John Frazer goes for the bottom-up approach that is the theoretical basis for robots and for developing ecosystems in a-life. Some examples come from the robots created by Rodney Brooks in the Mobot Lab [13] or from W. Grey Walter’s Machina Speculatrix [14] . Ultimately, the interactive machines are living architecture machines, with a character close to those imagined by Nicholas Negroponte [Negroponte, 1970, p.64]. But the machines created by John Frazer work on the basis of an evolutionary architectural design, not a traditional “top-down” design where the controlling mind of the architect oversees all factors and elements of the project. Design process – tool for evolution John Frazer’s work is in a first stage a search for form. With these projects he explores a new type of architectural design based on morphogenesis. As mentioned before, this design is a process that manipulates information, from which the forms emerge. The computer is the tool of evolution feeding on information from the described machines creating evolving architectural systems. Emergence is a key concept for John Frazer, based on Stuart Kaufmann’s phrase “life emerges on the edge of chaos”. This new architecture will also emerge in a natural way, just as order emerges from chaos [Frazer, 1995a, p.103]. In CA emergence occurs naturally after some generations, when a pattern becomes visible in the process of morphogenesis. Like Norman Packard’s snowflakes, Conway’s creatures or the patterns in the CAM-6 (cellular automata machine) of Toffoli and Margolus, originated from an initial random cluster of cells. Even in “elementary” 1D systems like the ones developed by Stephen Wolfram complexity emerges from very simple initial states [Wolfram, 1982]. In the design process, the initially emergent form (seed) goes through a process of evolution. What evolves is not the form itself by direct changes on its geometry, but the data or the rules behind the generation of its geometry. The seed can also be replaced by another with a different genotype growing something new [Frazer, 1995a, p.99]. Genetics principles in GA and selection are the tools for this evolution. The definition of the criteria for selection when using the GA is always a delicate issue and John Frazer and his students used different techniques, sometimes in combination [Frazer, 1995a, p.59]. A method that in his opinion proved efficient was artificial selection. When the criteria are non-quantifiable the architect applies his personal evaluation to select a successful result [Frazer, 1995a, p.75]. Even purely aesthetic concerns can be considered whenever the selection in this design process is a visual task. Some projects developed with this approach are an Evolving Sequence that is generated by a GA using data from the CA of the Universal Constructor, and other GA based for evolving and growing forms [15] . They clearly relate to the experiment of Richard Dawkins with the biomorphs, where the selecting agent of the growing organisms is human. The main advantage pointed to this mechanism is the ability of searching through the genetic space [Levy, 1993, p.175]. Of all possible mutations contained in this space the user directs evolution to the result that interests him most, skipping many evolutionary dead-ends or deviations. We find further advances in a visual and aesthetical fitness as form of selection in computer generated art. One example is the work of Karl Simms, who was also inspired by the biomorphs. But he added the power of GA to morphogenesis, searching through a genetic space of algorithms instead of forms. So the focus turns to the rules behind the creation form, not its description. It’s a procedural method for creating complexity [Holtzmann, 1997, p.85]. Oppenheimer uses morphogenesis by directly manipulating the algorithms, and also reached the conclusion that life is not aggregations of matter but processes that organise the matter [Levy, 1993, p.239]. These two examples are closer to the evolutionary method proposed by John Frazer. This controlled evolution based on artificial selection is important if the architect has specific non-quantifiable goals in mind. But Frazer supports an architecture model that results from a “blind” evolution, leaving the designer without knowledge of its outcome [Frazer, 1995a, p.12]. It only respects the encoded rules of generation and the external forces from the environment. This new process takes the control out of the hands of the architect. The different elements organise themselves in an unpredictable, often unexpected way. This approach is the openended strategy where the criteria emerge from the interaction between the organism and the environment. The form adapts itself in unimagined ways, because those requirements weren’t identified from the start [Levy, 1993, p.215]. The ultimate creation of John Frazer’s evolutionary design systems is the specific model, where most of the advanced designs are created. It gets close to the concept of universal computer. This model uses concepts of 3D CA for the definition of its space and for the information flow building what is called the Universal State Space Modeller. It’s an abstract virtual environment, existing solely in the form of software that receives data from the before mentioned Universal Interactor. This way it uses the concepts of CS to deal with the input/output operation of physical environmental information into the system. Inside the CS resides the GA, which takes care of evolution of the genotype and the rules, through a learning process. The forms are generated in relation to the abstract environment, establishing a dialogue between the elements of both systems – an ecosystem is created. Examples of this symbiosis can be found in several projects [16] . The possibility of co-evolution of two objects also exists in this environment, with the consequent advantages. It enables an open-ended evolution due to competition between the two systems. They constantly improve their performance in relation to each other without the need of an external stimulation. A scientific record of this phenomenon occurs in William Daniel Hillis’ work and he concludes that this way the evolved algorithms can achieve better results than human designed algorithms [Levy, 1993, pp.201-203]. In fact, awkward algorithms emerge, defying the usual human logic, but reveal an efficient response to the environment. Thus co-evolution reveals itself as a technique relevant to the “blind” uncontrolled evolution defended by Frazer. The evolution as part of the design process is the change between different generations of an object until a certain prototype is reached. Adaptation through response and interaction of the organism during its lifecycle leads to behaviour, the second stage of John Frazer’s evolutionary architecture. Architecture – adaptive system The evolutionary architecture of John Frazer has only evolved until now in the virtual space of the computer. It is like a prototype of the architectural object. But it doesn’t mean that at the final stage of the project it becomes a static materialisation of the design process, a frozen moment of evolution. Frazer envisions an architecture that works as an ecosystem in interaction with the environment [Frazer, 1995a, p.16] and through this interaction proceeds being an evolutionary architecture. The development of interfaces in other machines can become a useful experience for integrating these in real buildings. With this direct relation to the previous stage of design, the building would continue to evolve in the same data setting that originated it during the design process. The project that got closer to physical manifestation was the Generator [17] . Unfortunately it remained only a working model. It points to an architecture where there is a constant feedback between the user and the building. It has an adaptive property that enables it to work in symbiosis with the inhabitants and the environment, allowing the rearrangement of its configuration according to different requirements. This architecture would be a physical realisation of artificial ecosystems, like some models created in a-life research. In Larry Yeager’s Polyworld organisms develop adaptive strategies for living in the environment having a certain type of response derived from their inherent properties [Levy, 1993, p.5]. The result of this type of approach in architecture can lead to emergent behaviour: an unprogramed response to new situations found in the environment. In the case of the Generator, it has the ability to suggest new configurations derived from the ones used before. One of the numerous examples of emergent behaviour in a-life is Craig Reynolds’ boids. Besides demonstrating flock behaviour, they “know” how to avoid obstacles encountered in the environment [18] . On emergent behaviour and decentralised systems a good source is Mitchel Resnick’s research using Starlogo and LEGO/Logo creatures. In David Ackley’s AL, another artificial ecosystem, a further property is introduced – learning – thus increasing the chance of the organisms developing new behaviours. Frazer had this property of quasi consciousness in the circuits of the Generator, that recorded user actions and considered possible modifications, learning to create modifications and selecting the best suggestions over time [Frazer, 1995a, p.41]. This quality leads to behaviour evolution, extending the evolutionary process far beyond the scope of the architectural design process. Behaviour evolution is demonstrated with the GA’s in Karl Sims’ creatures that learn how to swim and walk, achieving astonishing results [Holtzmann, 1997, p.91]. A wide variety of unexpected locomotion methods emerge. Some similar to the ones found in the natural world, others only possible in the specific virtual environment. If the building acquires the autonomy to modify itself by means of robotic mechanisms, as Frazer considered for the Generator, a trace of character from von Neumann’s Self-replicator can be recognised [19] [20] and also of the self-replicating factories from the SRS Concept Team . These are two visionary examples of introducing advanced a-life organisms in the physical world. “All the parts of the model co-operate and in that sense it can be considered as an organism, but it will only fully exist as such if it a member of an evolving system of organisms interacting with each other as well as with the environment.” [Frazer, 1995a, p.103] Cyberspace – new environment The projects described in John Frazer’s book were developed inside the computer and all the evolutionary process is linked to computational processes. This architecture is based on information and together with all the concepts from the specific Worldview behind it (emergent, complex, interactive, decentralised, procedural) it seems to find its natural environment in the parallel world of cyberspace. Since they feed on information from users and the environment, these virtual objects have full potential to evolve if integrated in this abstract but “real” environment. John Frazer already put this into practice by connecting to the Internet computers responsible for the evolution of the architectural models. The Universal Interactor was exhibited in 1995 the Architectural Association. It was feeding the computer model with data from the physical environment. But in addition to this source of information the model was receiving genetic data from the Internet. Via a web site, remote users were allowed to set up a series of values and then send it to the model. The software could also be downloaded into other machines, where it would keep evolving according to local events. The prototypes can be set in the digital world, vulnerable to the stimuli from remote users, local agents or any type of streaming data. They become part of an artificial ecosystem. This experiment is of relevance to a-life, which searches precisely for the creation of artificial ecologies. These models can become part of it. They are artificial forms with an evolutionary and behavioural ability that allows them to develop autonomously in this new artificial reality. Just like many other creatures from the previous examples of artificial ecologies. Payback – understanding and creativity John Frazer foresees some contribution of Architecture to the a-life field mainly in the area of morphogenesis [Frazer, 1995a, p.20]. Besides that, I would personally extend the concept of Tools for Understanding [Frazer, 1995a, p.33] to all the machines he developed. They bring to the physical world the workings of artificial life systems, and offer the users a possibility to interact with them having a better understanding of the underlying concepts. The Universal Constructor, Self-replicating CA and Autonomous Cells [21] are direct materialisation of CA concepts. The Neural Network Machine is another example relevant not only to the field of artificial intelligence, because neural networks are also used in the a-life field, namely classifier systems where the role of learning is important. John Frazer combines concepts, ideas and models in a creative and uncompromised way, unlike most scientists, maybe because he doesn’t have the pressure of scientific validation for his experiments. Scientists tend to stick to a certain strand developing it deeper every time, trying to prove things to be right or wrong. John Frazer just uses the principles for his personal purposes and in the end evaluates the results from a design point of view. Even if he uses concepts in a personal and less orthodox manner, the results yield “food for thought” for both fields: architecture and a-life. Some experiments dealing directly with CA introduce new areas of application of a-life, not necessarily related with natural systems [22] . Conclusion The projects presented by John Frazer are an expression of logic in space [Frazer, 1995a, p.45], moreover in cyberspace, when considered as a space of logic and information. This evolutionary architecture has to be seen as the design process, not as a finite architectural object. John Frazer’s work is a report on an empirical use of the knowledge of a-life, detached from the philosophical concern around the existence of life in a-life systems and from the recreation of any natural phenomena. He only approaches the principles behind their complexity to develop a new paradigm for architectural design. After all this is architecture, in essence an artificial creation. With this background it becomes a form of Artificial Life. John Frazer’s work is of great coherence. Every step, model and project is an experiment that shares the same Worldview with a-life, becoming deeply embedded in the development of the field. It opens new doors in architectural design for the development of projects in a world that is more and more depending on circuits of information and conscious of the relevance of complex dynamic systems, like the ones found in nature. Notes [1] Langton [Levy, 1993, pp.108/118]; Oppenheimer [p.239]; Packard [p.251]; von Neumann [p.22] [2] [3] [4] [5] [6] [7] [8] [9] On Alan Turing’s machine [Levy, 1993, pp.22-25] Cellular Automata techniques [Frazer, 1995a, p.51]; Evolutionary techniques [p.57] The rules of the Life game [Levy, 1993, p.52] The workings of Genetic Algorithms explained [Levy, pp.162-164] Classifier systems: a definition [Levy, 1993, p.243]; a schematic [Levy, 1993, p.245] The biomorphs explained [Levy, 1993, pp.172-173] Steen Rasmussen’s crib sheet [Levy, 1993, p.145]. Several machines for modelling are described [Frazer, 1995a, pp.38-39] Description of the Universal Constructor [Frazer, 1995a, p.44] Description of the Universal Interactor [Frazer, 1995a, p.75] Images of different sensors and effectors [pp.76-77, pp.80-81] Mobot Lab research [Levy, 1993, p.282] Machina Speculatrix [Levy, 1993, p.283] [10] [11] [12] [13] [14] [15] Examples: Evolving Sequence [Frazer, 1995a, pp.46-47], Evolving and Growing Form [p.79] and Surface Evolution [p.75] [16] Examples: Hierarchical Cellular Automata [Frazer, 1995a, p.89], Generative Sequence [pp.92-93], Emergent forms [pp.104-105] and Emergent forms under solar influence [p.64] [17] [18] [19] [20] [21] [22] Description of the Generator project [Frazer, 1995a, p.40] Description of the boids rules [Levy, 1993, p.79] Description of the self-replicator concept [Levy, 1993, p.26] SRS Self-replicating factories [Levy, 1993, p.35] Autonomous cells [Frazer, 1995a, p.56] Examples: Simulation of Turbulent Flow, Dicing with Music, Seeding and Response [Frazer, 1995a, pp. 52-53] Reference List Frazer, John [1995a] ‘An Evolutionary Architecture’ London, Architectural Association Frazer, John [1995b] ‘The Architectural Relevance of Cyberspace’ in Maggie Toy, Martin Pearce and Neil Spiller (ed) ‘Architects in Cyberspace’, Architectural Design Profile No 118, pp. 76-77, London, Academy Editions Frazer, John [1995c] ‘Architectural Experiments’ in ‘Architects in Cyberspace’ op. cit., p. 78 Frazer, John and Frazer, Julia [1998] ‘The Groningen Experiment: Architecture as an artificial Life form’ in Maggie Toy and Neil Spiller (ed) ‘Architects in Cyberspace II’, Architectural Design Profile No 136, pp. 12-15, London, Academy Editions Frazer, John and Rastogi, Manit [1998] ‘The New Canvas’ in ‘Architects in Cyberspace II’ op. cit., pp. 8-11 Frazer, John; Rastogi, Manit and Graham, Peter [1995] ‘The Interactivator’ in ‘Architects in Cyberspace’ op. cit., pp. 79-80 Holtzmann, Steven [1997] ‘Digital Mosaics – The aesthetics of cyberspace’ New York, Simon & Schuster Levy, Steven [1993] ‘Artificial Life – The quest for a new creation’ London, Penguin Books Negroponte, Nicholas [1970] ‘The Architecture Machine – Toward a more human environment’ Cambridge MA, the MIT Press Resnick, Mitchel [1994] ‘Turtles, Termites and Traffic Jams – Explorations in massively parallel microworld’” Cambridge MA, the MIT Press Saunders, Robert [1997] ‘An evolutionary Architecture - book review’ www.arch.su.edu.au/~rob/study/AnEvolutionaryArchitecture.html (last accessed May 2000) Toffoli, Tommaso and Margolus, Norman ‘Cellular automata machines: a new environment for modelling’ Cambridge MA, the MIT Press Wolfram, Stephen [1982] ‘Cellular Automata as Simple Self-Organising Systems’ Caltech preprint CALT-68-938 Wolfram, Stephen [1988] ‘Complex Systems Theory’ in ‘Emerging Syntheses in Science: Proceedings of the Founding Workshop of the Santa Fe Institute, Addison-Wesley, pp.183189 Woolley, Benjamin [1992] ‘Virtual Worlds: a journey in hype and hyper reality’ Oxford, Blackwell Web site of the Exhibition at the Architectural Association www.ellipsis.com/evolutionary/evolutionary.html (last accessed May 2000)