Image Image Image Image Image Image Image Image Image Image

Bartlett School of Architecture, UCL

Scroll to top

Top

One Comment

Synthesis of Intelligence and Embodiment

Synthesis of Intelligence and Embodiment

Nicholas Negroponte’s quest to impart intelligence to machines has led architectural research to be focused on building quasi-living environments. At present these Quasi-Living environments are presently artificial but they behave like living entities which modify themselves in the surrounding landscape, they breathe in the atmosphere, they sense the world around them and respond to human intervention within them. Softness thus is an immaterial approach to impart intelligent behavior to systems.

I will explore in this research thesis the basis of imparting intelligence to machines. The various examples and research undertaken will establish the background of the study and help to understand the aspects of intelligence and factors leading to it. My aim is to create artificial systems which are intelligent to survive, adapt, cause minimum disturbance to ecology and interact with human intervention.

Buildings around us are quite unsustainable, static and rigid. We bring in a lot of material and put it on the small surface of the earth, destroying the surface forever. We also cause a considerable change to micro climate which cannot be restored even by demolishing the structures we build. The rigidity of such buildings that are alien to its surroundings give rise to instability in the micro climate and therefore to the sustainability of the ecology. We overdesign structures to be capable of withstanding all sorts of environmental forces even if the forces do not act on them.

Even when we incorporate several smart technologies within the internal environment to make them interactive in changing circumstances, yet there is no intelligence imparted to the system. A sensor operating lighting system detects the movement of a person in the surrounding and switches on the light for a specified time. It does not distinguish between the number of people entering, the effective lighting requirement for the task to be performed. A system that can perform multiple operations also is restricted by the combinations of operations to choose from. They behave according to tasks assigned to them through a  predefined list of operations that they can perform.

Architecture has a complex responsive external and internal environment. It has multiple inputs to determine the output and several factors acting at the same time. The response to this environment is usually control over openings in a building to let in light and ventilation, thus buildings are objects placed on the face of the earth which do not interact, sense or respond to the ever changing environment. We search for a softer approach to architecture to create structures which inhabit the environment and exuberate life .

 

2. DEFINING INTELLIGENCE

[Phiefer et al.,1999] Since mid 20th century researchers have been finding ways to impart intelligence to machines. The Turing machine laid down a language for computers. Computers were compared to human brains and the principles of human intelligence were transferred to machines. Computer simulation revolutionized the way intelligence is imparted to machines by their ability to do complex reasoning, logical operations and problem solving, giving birth to artificial intelligence. It has to be considered artificial, since it arises from synthetic elements, and exists in a virtual world. This paved a way to understand human intelligence via computer simulations.

The Turing Machine, one of the early devices to derive binary system of computation to calculate mathematical problems.

The Turing Machine, one of the early devices to derive binary system of computation to calculate mathematical problems.

But this was far from intelligence as we observe in humans and nature. Computers have been unable to do simple tasks of recognizing a face, walking or multitasking. Tasks which are very easy for humans to do.

[Phiefer et al.,1999] In 1980’s Intelligence was redefined by researchers in artificial intelligence, psychologists and cognitive scientists in various ways. The most comprehensive definition in my view is by J. Peterson, who defines it as the ability to adapt oneself by responding to a complex stimulating environment in the form of a unified behavior. This definition of intelligence finds similarity with Gordan Pasks conversation theory which describes intelligence derived out of constant communication feedback loop between the object and its environment.

It has been practiced in the realm of cyberspace to generate loops of interactions between a user and its environment. Therefore a system must be considered intelligent in context of the surrounding.

 

3. REQUISITE FOR INTELLIGENT SYSTEM: EMBODIMENT

The body and the environment cannot be considered separate from each other. As [Brodey,1967] mentions that the object and environment can replace each other can be considered in a mutual  relationship. This is important in the context of real and virtual environments. If the system is tested in a real environment, it is important for the system to have a physical presence. [Pheifer et al.,1999] Brooks called such an intelligence imparted in the presence of a physical body as ’embodied intelligence’.

The virtual world though gives immense possibilities to build an object and its environment in virtual space to test it for various alterations. It can be considered unintelligent to test a computer simulation in a real situation.

Embodiment can also be interpreted as segregating the various levels of intelligence over the different parts of the body as observed in humans. The activity of walking is an example of this process. As [Pheifer et al.,1999] describes that the memory for walking is not entirely stored in the brain but the knee and the ankle joints are independent of nervous control and rely on the gravitational force to swing the leg forward when it is lifted off the ground. This is a process humans learn while growing up. The passive dynamic walker exhibits the walking mechanism without any central controller, solely moving on the inclined plane depending on the gravity and the morphology of the body.

 

4. BEHAVIOR: AN OBSERVANT ASPECT

It can be concluded that for a system to be intelligent we can provide a combination of central preprogrammed controller and a well designed morphology to carry out tasks in an environment. [Pheifer et al.,1999] Braitenberg phototropic vehicle is a simple example of such a system. The vehicle consisted of a chip based controller programmed to react to light and light sensors in body to sense and move around according to pre defined stimulus. But it is observed that the vehicle shows complex trajectories in complex lighting environments, therefore imparting it an unexpected behavior. Though humans have a limited sensing ability, we can predict and make relevant choices in an unpredicted situation depending on our behavior.

[Bondin,2013] suggests that behavior should be the governing aim of design. In order to observe intelligence in a system, we need to define characteristic behavior that a system should exhibit. Behavior seems to be a response of the observer and might be difficult to impart to a system. Since it is in the eyes of the observer and not pre defined in the system. We can study the behaviors observed in independent systems to find processes that make the system acquire a certain behavior.

 

5. BEHAVIORAL ASPECTS

For a system to survive in complex dynamic environment it can be studied with respect to adaptable, autonomous and evolutionary behavior. Though these terms are inter related, they have various individual characteristics.

5.1. ADAPTABILITY

As it is discussed that a dynamic environment needs a real time response, adapting to the changing environment is an important aspect. A system can be observed to be adaptable via dynamic response to senses or modifications to morphology.

 

Dynamism in an object provides it a characteristic to adapt via stimulus responses in real time, an important observation in Theo Jansen’s strandbeests. The object behaves like living creature and is imparted intelligence to sense and respond to the beach environment independent of any computer control. It adapts to stimulus responses by the ability to move around.

Adaptability in morphology is a long time process as observed in genetic modification of ape to human beings on centuries of evolution.

5.2. EVOLUTION

Evolution of living creatures is an extraordinary behavior which is required for the survival of the species. It has been observed over centuries to have provided intelligent creatures comparable to previous generations. Evolution can be discussed by a basic division into regenerative, self organizing and self learning characteristics.

5.2.1. Regeneration

Regeneration can be considered most evolutionary principle as it provides morphological changes with evolving generation. This ensures making a fitter individual in every new generation, helping it to adapt to the environment it is going to imbibe. This process has been simulated in virtual environment via genetic algorithms. These algorithms try out various possibilities of permutation as defined by basic rules and test and compare each outcome for their survival rate, thus obtaining the fittest combination. The trial and error method might not be the best way to test complex systems. But in robotics, this approach is quite efficient because of rapid computation timing.

Since architectural environments are fabricated, they can be compared with evolutionary robotics which has been studied by Prof. AE Eiben [2014]. He believes that the use of evolutionary processes in artificial systems has a great potential for research and applications in robotics. He studies reproduction mechanism of simulated roombot modules to provide autonomous ability for robotic ecosystem to evolve by reproduction and self learning. These are modular robots designed by researchers in Switzerland, capable of reconfiguration of their shape. Though the study is only via simulation and in coherence with a central controller, the independent learning is difficult to observe.

We can observe evolutionary process in living organisms life through reproduction. Though there could be different ways of evolution in terms of artificial environments. Humans are composed of materials which decompose in a shorter span than the materials composing the artificial environments, therefore we might not need to mimic the biological system but devise an evolutionary way for artificial systems which can live for centuries.

5.2.2. Self Learning

The mind has to evolve with time and encounter new experiences in the lifetime of an intelligent living creature. Hence self learning is an important aspect for the system to teach itself to grasp knowledge while  interacting with the environment. The experiences need to be stored and recalled when a similar situation arises.

Similar to biological neural networks there are artificial neural networks (ANN) which receive the sensory stimulus and generate the most relevant response out of the previously stored information.

[Sher et al., 2013] use ANN based data to compare generated values in a physical prototype of a load bearing canopy which modifies its shape with changing load patterns. The learning abilities developed in the system were quite similar to the ANN data and improved with further practice. Hence, similar examples can be tested with responsive systems and embedded in the memory for recollection later on.

Kinetic pavilion is a student project to research external and internal environment responses by altering the volume of spaces within by modifying the roof shape. It shows transformation in the structure derived from changes in external lighting and to the movement of people inside the structure. Though these two responses are independently monitored, in practical cases they might need a simultaneous response, which may provide conflicting behavior.

 5.3. AUTONOMY

Since the system needs to take decisions in real time which provides it complex sensory perception, I realize that autonomy is an important aspect to be imparted. The system should have control over its variety of learning data to generate most appropriate response.

Symbrion is an example of robots showing independent and collaborative behavior at the same time. Even when the modules attach themselves to each other, they do not get pulled by the front module but move at their independent pace to preserve themselves against obstacles and getting dragged along surface.

In case of physical agents, this an important property to be able to survive without depending on external agents and maintain a continuous supply of energy to perform thus being self sufficient. Theo Jansens creatures are an example displaying autonomous behavior by deriving its own energy from environment to move and independently react to the external stimulus in real time.

 

6. ASPECTS TO EMBODIMENT : MORPHODYNAMIC DESIGN

Autonomy is an important aspect to embodiment. Since it defines various levels of control to various parts of a system, imparting several levels of intelligence to them. This aspect can be designed as loosely coupled processes or could be a feature of material composing the object.

6.1. PARALLEL SENSOR MOTOR COUPLING

[Pheifer et al.,1999] considers transferring intelligence to the body as an off load of intelligence on the central controller. Thus making the process faster and efficient for the system. Hence the loosely coupled sensor motor system off loads the processing from the central system imparting it to various parallel processes making it more reactive to sensory perceptions.

Implications of embodiment as described by [Pheifer et al., 2007]

Implications of embodiment as described by [Pheifer et al., 2007]

[Pheifer et al. 2007] Apart from the interaction between mind and the environment through the body dynamics, it is interesting to note the independence of feedback-response of the environment and body (independent of the mind) This requires a morphology where a part of the body is capable of independent mechanism when necessary.

The form of an object is inconsequent in a dynamic system since it is the morphology and the kinetic movement involved with it which are perceived by the visual representation. In an embodied system the design of the moving form is extraneous than the ability for various degrees of motion it achieves. The multiple joints of bending thus is of more importance to study the dynamic aspect of the body.

Mechanical automata and principles of gear combination are quite an explored branch for further study to understand various ways of transforming a movement.

6.2. IMPARTING INTELLIGENCE TO MATERIALS

Certain materials have been designed to show intelligent behavior. The shape memory alloys are an example of dynamic transformation on external impulse. They store a memory to modify themselves into another form in the presence of external temperature rise.

The Shape shift is an experiment to make surfaces capable of transformation via electric impulses. Therefore imparting certain intelligent behavior to independent parts of surfaces, offloads several processing commands to be sent from the central controller, making the system robust.

 

7. CONCLUSION

It is difficult to transform intelligence in nature to artificial systems. We essentially use motors and sensor couples to impart such a behavior. But motors are imprecise and sensors are noisy. These system lack fault tolerance and sifting of relevant data from the abundant data it gathers. Getting a real time agile response is also a big challenge.

These aspects are better managed in biological and natural systems. But a replication of these systems do not impart intelligence, rather a derivation of the principles in such systems to form characteristics suitable to the artificial world is an advanced approach. Just like the design of wheel which is much faster and friction less process of movement as compared to walking on smooth surfaces, we need to find superior ways to exhibit intelligence in artificial systems.

Since versatile learning and an agile response to autonomous systems can impart an intelligent behavior, embodiment of intelligence becomes a powerful driving force to future interactive systems.

The literature review gives a basic understanding of various behaviors to be imparted to systems capable of showing intelligence as understood in context of the environment it belongs to. There are certain examples of embodiment in different systems which pave the way for future study. Morphological transformation and genetic algorithm are key elements of design to be researched further in detail. These principles have examples in mobile robotics but needs an understanding in the realm of immobile architecture. There is a need to understand the ways to impart knowledge, and question the existence of artificial intelligent behavior. A study of epistemology and ontology will help to build on the machine learning aspect and transfer the characteristics to the morphology of the structure. A further exploration is required in the methodology to impart behavior and a reasoning of generative design.

 

 Key References:

  1. Negroponte, N. (1975).Soft architecture machines. Cambridge, Mass, The MIT Press.
  2. Pfeifer,R, Scheier,C. (1999).Understanding Intelligence, MIT Press, Cambridge, MA, USA.
  3. Bondin, W., (2013). Embodied Dynamics : The Role of Externalised and Embodied Cognition in Kinetic Architecture.
  4. Rosenblueth, A., Wiener, N. & Bigelow, J., (1943). Behavior, Purpose and Teleology.Philosophy of Science, 10(1), p.18.
  5. Eiben, A E., Kernbach, S. & Haasdijk, E., (2012). Embodied artificial evolution: Artificial evolutionary systems in the 21st Century.Evolutionary Intelligence, 5(4), p.261-272.
  6. Pfeifer, R., Lungarella, M. & Iida, F., (2007). Self-organization, embodiment, and biologically inspired robotics.Science (New York, N.Y.), 318(5853), p.1088-1093.

Supporting References:

  1. Brodey, W. M. (1967). The design of intelligent environments, soft architecture. Landscape, Autumn 1967, Vol. 17:1, Pg. 8-12
  2. Sher, E., Chronis, A. & Glynn, R., (2013). Self-Learning Algorithm as a Tool to Perform Adaptive Behaviour in Unpredictable Changing Environments — A Case Study.
  3. Von Foerster, H. (2003).Understanding understanding essays on cybernetics and cognition. New York, Springer.
  4. Rossi, C. & Eiben, a E., (2014). Simultaneous versus incremental learning of multiple skills by modular robots.Evolutionary Intelligence, 7, p.119-131.

Comments

  1. “intelligence is a property ascribed to a conversation between participants by an external observer if, and only if, their dialogue manifests understanding”

    Gordon Pask

Submit a Comment