Thesis proposal- Trisha, Siqi, Ruixin

“First we shape our tools, thereafter they shape us.”

PROPOSAL

Movie Factory/ Controlled Environment

Based on the process of an AI dealing with a task: the perception of environment---capable of perceiving human emotions and thoughts; the process of the goal using logic---capable of vague logic/naïve physics/human instincts; the end effector and execution---to push to the extreme, like the robot that could turn everything into paper clips?

There might rise some issues of future human-machine collaboration --- of superintelligence

We could use AI, yet we need to be careful

Some points that we need to pay attention to? What specific points that we should focus on to avoid super-intelligence in terms of future human-machine collaboration?

Idea-

AI controls the spatial experience within the movie factory. We feed the information about the type of scene we’re interested in- it may be the theme, the movie itself, a particular situation. The user can build the scene, input information that defines the type scene he/she is interested in, and the AI calculates the direction in which the movie can head by creating its own set of objects that controls the next scene.

In the movie factory, the movie is partially directed digitally, because the artificial intelligence is stimulated by the already fed in information. The human provides the seen but the machine develops its own story based on a careful curated system and algorithm. This forms the basis for the human AI collaboration that we are trying to achieve.

We are building on the idea of a self learning system, where inputs multiply and transform as we move along the process. This also incorporates the concepts of immersive art wherein the digitally created space blends with the physical environment to produce a movie like world. The different environments could be treated like the scenes of the movie, each of which are carefully  curated to create a holistic experience for the user.

The process of feeding in information and modifying the environment accordingly

The programmatic aspects of the idea is to also bring out the public participation, where the people can visit and physically become part of this collaborative process. As a result, they will not only be able to view the movies, but be a part of it and play a role in directing it.

AI and architecture in the future

“A city thought up and built by AI would be similar to nature,”. “Not visually, but philosophically. Living organisms go through millions of years of evolution to reach the level of sophistication and adaptation to their environments. This process can be accelerated by computational power to design and build cities that are complex, adaptive and resource-efficient.The suggestion is that an AI-guided city would be extremely optimised. Like an enormous organism, it would find a structure that best suits its environment, where movement of cells is at its most efficient. Such a model, however, doesn’t have a great deal of room for you or me, and our individual, subjective experiences. As Mamou-Mani tells me, building with computers is a delicate balance between humanity and the machine.”- Thomas McMullan

Some general thoughts on machine-human collaboration in the future:

In what way would AI collaborate with human in the future? -- Now human gives the commands, the target and AI executed the order automatically

RESEARCH

General Definition of AI

Artificial Intelligence (AI), as suggested by the term itself, generally denotes the intelligence shown by non-human objects, such as machines and systems. Taken apart, the adjective “artificial” restricts the range of subjects that show the traces of intelligent behaviors, while the noun “intelligence” defines the intrinsic characters that AI has, fundamentally defined the scope of AI.

As the key of understanding AI, the concept of intelligence has been studied since the time of ancient Greece, when Aristotle and Plato had a dialogue on what behavior should be considered pious. Aristotle then formed the famous system of syllogisms for reasoning with initial premises and conclusions as an attempt for establishing rules for the rational part of human mind. The study of mind can be then found in the later schools of thoughts, such as dualism and materialism. Whether the operation of our mind is dependent on the physical environment is the focus at this period. Having established the principle that mind is operating by reasoning the knowledge it gained, the source of our knowledge—what we learn using our intelligence and how we achieve that—is later explored in the empiricist movement and even in the cognitive psychology.

Developing from previous studies, the contemporary definition of intelligence is controversial, involving a range of abilities from reasoning, planning and solving problems to being self-aware emotionally. Among them, the ability to sense the environment, think rationally and act purposely is the main character of human intelligence that the study of AI is focused on. Based on this, the definition of AI varies from scientists to scientists. Typical definitions include “activities associate with human thinking such as decision-making, problem solving, learning” and “automation of intelligent behavior”. Generally speaking, those definitions could fall into 4 categories inside 2 major dimensions: the first dimension would be thinking/acting, and the second is humanly/rationally. The differences in the first pair address whether AI is able to interpret the information and reason in a certain manner, or whether it is possible for AI to act and behave. The second dimension discuss the discrepancy between a human mind and a pure logical mind, which lays the theoretical background for the potential future collaboration between AI and human being.

The study of AI

Nowadays in the realm of computer science, AI is a study of understanding and building intelligent agents. The intelligent agent is a system which acts intelligently: it can perceive the situation of the surrounding flexibly, learn from the experience and environment and act given the limitations and the goals. Under this circumstance, the rationality inside the concept of intelligence is emphasized instead of the human characters.

The study of Artificial Intelligence has described the structure of an intelligent agent with two key characters in its definition: the inside composition of the agent(system) itself and its outside ability or behavior to think and act based on the task given. It could also be simplified as architecture and program. Architecture is the physical settings that can allow the perception, reasoning and acting of agents happen, and the most common architecture of intelligent agents is the computing device and hardware. Generally, an agent consists of environmental sensors such as infrared sensors and anti-collision sensors, and effectors like a robotic hand and wheel. The program describes the function that the agent has from percepts to actions, which is the major focus of AI studies.

How would an agent make decisions and finally complete the given task step by step? A typical process for an intelligent agent to act is dependent on four features: the given performance measure related to the accomplishment, every information it has perceived so far as the percept sequence, the information of the current situation that it should react to and the actions that it can take. Ideally, the agent is expected to maximize its performance measure by taking whatever action, based on the overall percept history and the built-in knowledge it has at the beginning. For contemporary tasks in computer science, it is achieved through algorithms including heuristics, nearest-neighbor search, etc.

Regarding the actions of intelligent agents, the assigned goal is also very critical. Ideally, agents are designed to be able to complete their goal with limited information and commands, learning from the environment themselves. Their tasks range from explicitly defined targets such as chess playing, to complicated, implicit goals as speech processing and image understanding. For the future development of AI, the range of achievable tasks might be largely expanded. Tasks that can only completed by humans in the past are already under experiment with AI. AI has successfully made a film with its ability of learning and synthesizing past screenplays, and artworks with assistance from AI can also be found in exhibitions. Though the quality of AI works calls for questioning, the possibilities of more intimate human-AI collaboration exist.

Common tasks of AI today

AI has been implemented in many fields of industry today, while some commonalities can be found in the tasks it is set to complete. These common tasks are basically sub-fields of intelligence that humans seeking to achieve through artificial intelligence. Yet some might be far from the study of space and experience in architecture, some fundamental problems of AI are related closely to our field.

1)     Reasoning, problem solving

Reasoning and problem solving is the general underlying problem of all the tasks in AI study. Early in this field, a simple reflex agent that could react directly to the situation given the built-in rules is a clear example of this, while in that phase, the logic and algorithms that researchers developed for agents are relatively straightforward, imitating the step-by-step of human reasoning. Examples of this can be a simple system that reacts to the situation of brakes in the car in the front: if it brakes, then start braking. More recently, the study of reasoning and problem solving of AI deals with uncertainty and incomplete given information, mostly of qualification problems. The current method introduces the probability theory, offering a way to quantify the uncertain problem.

As abstract the task might be, the uncertainty of contemporary reasoning problems connects closely to design decisions. How AI deals with the uncertain metrics of the design result and how its logic reacts to the incomplete information would be a crucial problem for AI to be integrated in architectural tasks.

2)     Planning

Planning problems are similar to the classical problem-solving tasks, while it requires agents to divided the ultimate goal into steps of subgoals, predicting the consequences of each action and planning the process of accomplishing the task. It concerns with the stronger ability of the agent to discover the environment and optimize its actions. It is the key issue of more complicated AI tasks, including self-driving cars, autonomous industrial robots, etc.

To a great extent, the planning problem reflects the designing process of the spaces. Based on this strategy, AI can offer an alternative strategy for optimizing design decisions.

3)     Learning

Machine learning is almost ubiquitous in the industry nowadays. It emphasizes on the agent’s ability to learn from the experience with respect to the task and the performance measure given. The agent will improve its own ability to perform tasks through learning, which is from both the observation of past decision-making process and the general experience of itself. Patterns of learning that happen among human beings are experimented in agents as well, including supervised learning and reinforcement learning.

The ability of intelligent agents to learn makes it possible for AI designing unpredicted alternatives for spatial problems and improving the outcome based on feedbacks.  For the role of architects related to AI, the selective environmental information set for learning would be the key to differentiate human and machine, and a new set of standards might be introduced in the future.

4)     Perception

Perception problems deal with the ability for agents to sense the world they are placed, which is achieved through sensors such as infrared cameras, microphones, sonar and radar. The issue is two-fold: the design of the sensor itself and how agent process the collected raw data. Nowadays AI has already achieved certain level of accuracy in speech recognition, facial recognition and object recognition while tackling the perception issues.

AI’s ability to perceive the environment is revolutionary for architectural design problems, as it helps define the spatial experience precisely. However, the human experiences of space are complicated and even ineffable sometimes, and how AI can reflect human perception is an important issue for designing architecture with AI.

5)     Motion and manipulation

Robots are an important application of AI dealing with motion and manipulation issues. It is largely based on a synthesis of abstract, logical tasks of AI, with a special emphasis on the interaction with physical surroundings.

What role is AI playing in architecture?

There are various methods in which automation or artificial intelligence can take part in architecture, whether it is before the design process, during the process or after it.

1. Site studies and social research- remote networks researching information and storing it on clouds for future use. Remote storage technology enables us to derive information from distant sources and helps us take decisions throughout the process.

Eg. How Facebook and Google have ‘smart’ technologies to predict information about certain events or people, or their choices.

2. Design decision making

This is mainly based on the initial information fed in by the user. Based on given commands, the AI program could develop the design further by tracing the style and techniques of the initial designer.

AI in Design- Helping out with architecture designs, create design tools, modify softwares,

1. Design through exploration
2. Design by description
3. Conversational interfaces

3. Client user engagement

4. Robot craftsmanship, computational architecture

5. Integrated systems involving AI in the finished product to enhance spatial experiences

Examples of different ways AI can be used in architecture-

1. Designing a small component/prototype in a larger building

2. Use of AI in the functioning of a home after the inhabitants have moved in; eg. alexa/ siri to control

3. Robots to control the building of a structure

Taking an example of the software, Dreamcatcher, we can explore how design inputs by humans can turn into projects. This software takes in inputs like client databases, environmental information, site information, sketches, etc. and and converts them into outputs of optimized 3D design solutions. This is based on a self- learning algorithmic pattern, which repeats and multiplies to enhance the solutions. In the process, the design is further augmented and enhanced as the process repeats itself.

Computational Architecture Digital Grotesque Project

Artificial intelligence is named as the upcoming social connections of future of  human lives. It has pros and cons with respect to the kind of control it may have on our lives.

Pros- 1. Convenience

2. Liberate the productive forces of human beings

3. Efficiency of work production

4. Enhance the solutions beyond human capacity

Cons- 1. Limitations of creativity

2. Technical difficulties

3. Possible loss of control by humans

1. Convenience & the Liberation of the Production forces

Laziness stimulates innovations, and innovations lead to convenience. Using tools is regarded as the symbol of evolution, in ways of saving body energy, easier daily works, and comfortable enjoyment, to bring more and more convenience into our lives. From stone ages, farming, to stream energy and capitalism, our economy and social relationships are constantly moving forward by means of intelligence of creations and productive tool innovations.

AI is named as the fourth industrial revolution after the computer and internet. As its human nature in our history, that our intelligence makes our lives easier in creating new tools and products. AI would be a reproductive supporting tool to help our lives in advantage of human creativity. In this case, the production forces of human beings will be fully liberate into more sophisticated thinking, managements and innovations, with the replacement of duplicated simple operations by AI-managed machines.

2. Efficiency & Solutions beyond Human capacity

Human effort of productions are eliminated by biological persistence and body structures, which is weak comparing with giant machines and efficient robotics.

AI has already been used to create superior processes in healthcare, finance utilities and e-commerce as an improved “more intelligent” version of computing and calculation. While data-driven decision making already exists, collecting adequate sample sizes has been a painstaking process in the past. Collection work that once took months or even years could happen in a matter of minutes, particularly with the growing economy of data sharing. Repetitive tasks or massive information collecting, which human brain won’t be able to finish in a short time, are now handled by AI learning and producing, in seconds. The Annual Manufacturing Report of 2018 from The Manufacturer discovered that 92% of senior manufacturing executives believe that "Smart Factory" digital technologies such as artificial intelligence will allow them to improve their degrees of productivity and empower their staff to work more intelligently.

3. Subjective and Objective Decisions

In the near future, AI will become more intelligent presenting as a life and work supporter in our society.

When facing the situation of making choices, it comes to the essential debatable part of the using of Artificial Intelligence: Are you responsible for AI to make a decision for you, or it’ll be counted as AI’s liability? Base on your daily behavior and personal informations, AI will be able to make a choice that perfectly suitable for you, but eventually, it’ll not turned out to be your own conscience to that specific behavior.

“We don’t want to accept arbitrary decisions by entities, people or AIs, that we don’t understand,” said Uber AI researcher Jason Yosinkski, co-organizer of the Interpretable AI workshop. “In order for machine learning models to be accepted by society, we’re going to need to know why they’re making the decisions they’re making.”

As these artificial neural networks are starting to be used in law enforcement, health care, scientific research, and determining which news you see on Facebook. Social researchers are now debating that there’s a problem with what some have called AI’s “black box.” Previous research has shown that algorithms amplify biases in the data from which they learn, and make inadvertent connections between ideas.

CONCLUSION

This makes us wonder whether there is an actual possibility of artificial intelligence control our systems 100% or replacing the human mind altogether. This has been a much explored question in the recent times, and in fact has stirred some amount of fear too. There are a few logistical questions to be dealt with. If it is the human who is feeding information into the machine, who has the final control, the human or the AI. Are we controlling the AI or is the AI controlling us? As projects get more complex than just making an architecture plan, how will AI handle the creative aspects of the project, or make the decisions? Will it not become repetitive if the software creates solutions out of the same inputs?

Our stand in this sea of questions is that AI has the capacity to execute design processes efficiently and precisely. However, the seed of design, which is very subjective and inherent, lies in the mind and soul of a human designer, and we are envisioning a future of human- AI collaboration where the human provides the prompt and helps the machines to take the design one step further. The job of the AI is not to replace human thinking but augment it. We would like to take the idea of the human- machine collaboration further. The human mind keeps evolving and learning and there is the added factor of emotion that impacts decisions in one or the other. Can technology mimic this 100%?