HYPOTHESIS DRAFT_AARA, KALIN, YAXIN
HYPOTHESIS (DRAFT)
In reference to the etymology of automation, let’s discuss the implications of automatic action in design, with specific applications in the field of architecture. What can we reveal, about creativity, lifestyle, and survival through investigation of systems that require less human input?
Automation is in our immediate future. Throughout every aspect of modern culture, progressive integration of automatic systems is persistent and profound – beginning with physical tools and structures of manufacturing, and now with smart-devices and computing programs. Our ambitions for an automated future relate to a word: efficiency. Demands of higher productivity, larger volumes, and consistent products command our desire for automation. We seek ever greater possibilities for living and growing our societies – Bigger! Better! Faster! Stronger!
Examples of automation in society are found in our tools of productivity. For many decades, manufacturing has been supported by assembly lines populated by unmanned robots. Repetitive courses in the making of products was an early element to be automated in large-scale manufacturing. In transportation, real-time location data collected by our personal smart-devices run together as input into generating map routes, such as finding the fastest-time route to our destinations. (http://www.drdobbs.com/architecture-and-design/priority-queues-for-motorists/184405153). In communication, spam email filters censor messages containing words that are recognized to be common in spam emails. Automation in agriculture is present in satellite-piloted combines and driverless tractors.
It’s not just statistical data and solutions with straightforward, singular dimensions that are automated; subjectivity and opinion are automated as well. In the review and scoring of standardized test essays, a double-reader system of grading incorporates a robo-reader to assess papers. (https://www.nytimes.com/2012/04/23/education/robo-readers-used-to-grade-test-essays.html) This approach to grading relates to the automation of design – how can subjectivity and personal effects, such as taste, be influenced by automation?
Automation in design exists in two parts: form-making, and form-finding. The aspects of form-making affected by automation are distinctive; like industrial manufacturing, building creative works is impacted by robot-operated sequences and assembly. Constructing physical items and structures is affected in similar ways as commercial processes of production. Digital fabrication – cutters, mills, routers, custom tooling and other heavy machinery in crafting – is the evolution of using machines as artisan tools.
As the next advancement in the making-processes, digital automation is a part of our digital form-making. Digital image processing, such as cropping and color correction, are automated by coded software packaged in digital systems. In principle, robots and coded programs that perform repeatable sequences of action without connected human operation are the breadth of automated form-making.
Automated form-finding is a greater challenge to the meaning of creative work and is the utmost subject of our investigations of automation in architectural design. Design and general artistry are subjects of absolute personal preference and individual taste – or are they?
Design has principles. There are rules – parameters. Courses of color theory, proportion, formal harmony and composition are taught as foundations of art and design. Arguably, these rules do not describe the success of a design in the same, empirical way that the rules of physics describe the behavior of physics, or the rules of chemistry describe the behavior of chemistry. Nonetheless, design throughout time adheres to common attributes and guidelines. Hence theoretically, design can be coded and measured using these agreed-upon principles. To automate design, the degree to which these principles are considered are passed to digital designers: artificial intelligence.
WHAT IS THE DEFINITION OF AI
What is artificial intelligence (AI)? AI is a machine system, or process, of decision-making inspired by the logic of human decision-making. There are three hierarchical levels of sophistication when defining AI (https://medium.com/predict/types-of-artificial-intelligence-and-examples-4f586489c5de):
Super AI: Often the subject of science-fiction and indistinguishable from a human’s mind processes, this system of AI is complete and all-encompassing in processing inputs. This system can make logic connections with levels of abstraction that would be almost impossible for a single human to compute.
General AI: This system is a composite group of categorized actions that operate together seamlessly to mimic complex human ability. Included in this system is the ability to return its own outputs as new inputs, which allows the AI to “learn.” This self-sustaining feedback optimizes tasks and produces that offer better efficiency and results. An example of this system is Google’s AlphaGo machine, which optimizes for the best moves in Go, a board game, to win. This system “learned” the best moves for every possible board position through playing itself repeatedly.
Narrow AI: This system is the most widely available AI that society has contact with. Narrow AI automates a set of unchanging operations or actions to complete a single, 1-dimensional task – though with greater efficiency and consistency than a human. Automated design interventions, such as through generative digital design and digital fabrication, are commonly collaborations between Narrow AIs of differing tasks.
WHAT IS FUTURE OF AUTOMATION IN DESIGN
Utilization of artificial intelligence in form-finding is the imminent progression of design. If the craftsmanship and historical knowledge of foundational design principles are being passed to machines, what is the role of the designer? As production of viable and interesting design forms are generated from algorithms using accepted commonalities, is the designer left as the mere caretaker of the code, only present to adjust and make tweaks as needed?
The responsibilities of the designer changes as follows: because the repetitive, codified foundational concepts of design are taken from the designer’s burdens, the designer is liberated to focus on issues of higher complexity, addressing existential questions related to the goals and priorities of design. Designers will think of solutions as less process-based and as more gestural and baseline. Design, as a mindset and problem-solving philosophy, remains.
LITERATURE STUDIES FOR AUTOMATION IN DESIGN
Lit study 2:
Optimization -> Genetic/Evolutionary algorithms -> Machine Learning -> Neural Networks (Deep Learning)
The theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.
John McCarthy (1956) - “The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”
Merriam-Webster: A branch of computer science dealing with the simulation of intelligent behavior in computers. The capability of a machine to imitate intelligent human behavior.
- Build systems that think exactly like humans do (“strong AI”) <machines with human sensibilities, ie, HAL and TARS>
- Just get systems to work without figuring out how human reasoning works (“weak AI”) <performing specific tasks, ie, pattern & facial recognition => MACHINE LEARNING>
- Use human reasoning as a model but not necessarily the end goal
- Artificial Narrow Intelligence ( ANI )
- Artificial General Intelligence ( AGI )
- Artificial Super Intelligence ( ASI )
2Amazon builds a lot of its business on machine-learning systems (as a subset of AI) and defines AI as “the field of computer science dedicated to solving cognitive problems commonly associated with human intelligence, such as learning, problem solving, and pattern recognition.”
2In 2016, several industry leaders including Amazon, Apple, DeepMind, Google, IBM and Microsoft joined together to create Partnership on AI to Benefit People and Society to develop and share best practices, advance public understanding, provide an open platform for discussion and to identify aspirational effort in AI for socially beneficial purposes.
Those working with AI today make it a priority to define the field for the problems it will solve and the benefits the technology can have for society. It’s no longer a primary objective for most to get to AI that operates just like a human brain, but to use its unique capabilities to enhance our world.
Artificial Intelligence in daily life
- Current uses in AI (outside of design):
- Traffic - real-time location data from smartphones as input into generating routes (fastest-time route suggestions)
- Autopilot - airplanes, ships, future cars
- Email spam filters - adaptive censoring to update with messaging techniques (new wording, new senders)
- Mobile check depositing - pattern and image recognition
- Loan risk assessment - terms on loan applications generated on applicant meta-data to reduce loss risk
- Targeted online experiences - shopping, advertising, social media suggestions
Essentially, ML makes suggestions. Each subsequent round of suggestions is evolutionarily/genetically generated based on user reaction. Commonalities breed more suggestions, with each generation having greater fitness to the user’s personal interest.
Computer Vision, AKA image recognition & sorting
- Requires lots of upfront hand coding
- Nuances and contingencies must be known beforehand (eg, STOP sign on a foggy day)
Deep Learning, AKA really really big (deep) Neural Networks
- DL = ML self-selecting correct/incorrect parameters over millions of generations of input
- DL may see patterns (duplicate data/geometries) that humans cannot, & develop novel results based on hidden connections
WHY IS AI IMPORTANT IN THE CURRENT SCENARIO
In the age of information and data, data storage has a reached a point that is inconceivable for the human brain to process efficiently. By enabling an artificial intelligence to process this data and ‘think’ for us, we can strive to make our complex data driven problems, easier to solve.
We are trying to design ‘the Hidden Layer’ i.e. the portion of our brain comprising of neurons, that takes in information (input) and processes it to produce an output. If we successfully manage to create this ‘Hidden Layer’ consisting of a sequence of highly complex processes and procedures, we have created artificial intelligence.
Simple Hidden Layer
Complex Hidden Layer (the AI)
LITERATURE STUDY FOR IMPLEMENTATION OF DESIGN IN AI
Lit study 1:
Using the code logic of Neural Networks to develop a Neural Style Transfer code.
Initial Technique: The first published paper on neural style transfer used an optimization technique — that is, starting off with a random noise image and making it more and more desirable with every “training” iteration of the neural network.
Here, we aren’t training a network to do anything. We are simply taking advantage of backpropagation to minimize two defined loss values. The tensor which we backpropagate into is the stylized image we wish to achieve — which we call the pastiche from here on out. We also have as inputs the artwork whose style we want to transfer, known as the style image, and the picture that we want to transfer the style onto, known as the content image.
We use the outputs of various intermediate layers to compute two types of losses: style loss and content loss — that is, how close is the pastiche to the style image in style, and how close is the pastiche to the content image in content. Those losses are then minimized by directly changing our pastiche image. By the end of a few iterations, the pastiche image now has the style of the style image and the content of the content image — or, said differently, it is a stylized version of the original content image.
New Technique: This technique is feedforward — train a network to do the stylizations for a given painting beforehand so that it can produce stylized images instantly.
Lit study 2:
Image based AI design
- Package design
2. Minimalist effect brand recognition
3. Banner design using Luban
Generate 1 billion banner on 11.11 for Taobao.
4.UI design
Zeplin
5. Logo design
logojoy
Lit study 3:
Image processing
- Edge drawing algorithm, photo to low poly (space divide logic)
2. Adobe photoshop image crop automatically fill blank
3. Photoshop Apply style to another image (style apply to façade, color, material)
4. CycleGan
Software that can generate photos from paintings, turn horses into zebras, perform style transfer, and more.
Lit study 4:
AI painting
- Daizi shrimp painting with iconic style of Baishi Qi
2. Generate image based on description
Lit study 4:
- Xkool
Generate building based on regulations, FAR, etc., one project in 6 min.
2. AI divide space
Evolving Floorplans
http://www.joelsimon.net/evo_floorplans.html?utm_medium=website&utm_source=archdaily.com
WHY ARE WE DOING THIS ?
What are the potential influences of AI in architecture ?
- Self-designing
- Self-constructing
What is our preliminary goal/ strategy ?
We are interested in using the computer intelligence to be able to optimize and design space at the same time for greater efficiency for creating spaces for humans, building cities, accommodating population growth, designing new intelligent spaces for entertainment.Our aim is to be able to a find a middle ground or area of collaboration between human and machine, in the process of working toward full automation.
To find new ways of subdividing and designing space.
- Based on pattern recognition by an AI software.
- Based on optimization/ efficiency of the AI software.
- Using a hereditary or genetic code to evolve the best result (chosen/ biased) result.
WHAT ARE THE POSSIBLE AI SOFTWARE THAT WE COULD USE FOR THIS PURPOSE AND HOW WOULD THEY IMPLEMENTED IN DESIGN ? HOW WOULD WE USE THE GENERATED PATTERN ?
- Based on pattern recognition by an AI software
- Technique: We use two existing references, one architectural and the other an artist’s stylistic technique. We use AI to apply material and style to the architecture building.This would help to generate many renderings quickly for client to see if the building fits the context. This translated, new 2D image is then modeled into 3D by us. But through this, we are able to show how in future, the AI could generate 3 dimensional spaces.
- Possible AI software that we can use to achieve this.
- ‘CycleGan’ as the AI platform. It is currently used to translate an image of one style into the artistic style of another reference image to generate a completely new 2D painting.
- To use ‘Google T-SNE Map’ as the AI platform that sorts and categorizes paintings based on similarity in patterns, composition, color, etc using its own developed complex algorithm.
- Advantage: AI software/ code to perform this stylistic translation/ matching already exists.
- Disadvantage: This process is currently limited to 2 Dimensional pattern making. The AI cannot translate to 3D on its own.
- Possibility: Use this technique with another 3D AI form making process to achieve a full pattern AI translation.
- A tensor based self written code that allows the machine to learn and develop its own geometry
- Technique: Using a multi-agent system where one agent reacts in response to another.
- Self written code
- Advantage: Allows for customization of code to meet our design requirements. A tensor, enables a series of complex calculations in the ‘hidden layer’ similar to that of AI. These operations can be highly sequentialized to produce an intelligent output.
- Disadvantage: Technical difficulties of achieving the code.
- Possibility: achieving a simple 3D AI translation, holds great potential for future use.
- Using Autodesk Dreamcatcher as the AI platform
- Technique: AutoDesk Dreamcatcher enables the user to input a variety of needs ranging from “natural language, image inference, and CAD geometry.”. By outlining the input data, the program optimises and develops a satisfactory output by taking into account all the solutions. The designer can then guide the programs by refining the variables and discarding the unnecessary solutions.
- Arguments for the incorporation of AI into design often default to concerns around replacing the human designer. While many elements that are commonly modeled from scratch such as brackets, adapters and stiffeners may be created more effectively by a system such as Dreamcatcher, complex elements and aspects that are difficult to quantify will require new types of interaction to leverage human intuition and computational rigor in partnership. Dreamcatcher is pioneering new methods for interactive synthesis and optimization with industry leaders from the automotive, aerospace and manufacturing fields.
- Advantage: Customization of data input and output using a high intelligence existing AI platform.
- Disadvantage: No prior experience with Dreamcatcher.
- Using a Genetic Algorithm to iterate and evolve to the best possible result.
- Technique: Develop a code that runs through multiple iterations and self learns to optimize either on preset configurations or through a fitness function that is iteratively chosen by us. To then pick an option from the generated population that best suits our purpose.
- Possible AI software: To use Peter Von Buelow’s Genetic Algorithm to generate pattern forms optimized based on some chosen factor i.e the fitness function. Fusing two parents and achieving a resultant.
- Advantage: Can use 3D forms and iterate through genetic mutation. In house professors and resources that we can use to help us achieve our result.
- Disadvantage: We have not fully developed a code and run it through the AI platform yet.
- Possibility: Use the AI’s potential to produce and analyze iterations in 1000s
WHAT IS A GENETIC ALGORITHM IN AI
Evolutionary algorithms are a set of generic meta-heuristic algorithms which solve optimisation problems by imitating aspects of biological evolution.Genetic algorithms are a subset of Evolutionary algorithms inspired by Charles Darwin's work on evolution by natural selection.Genetic algorithms, or evolutionary algorithms, are the foundations of machine learning (ML). They are meta heuristic search algorithms relying on bio-inspired operators such as mutation,crossover and selection.
PROCESS OF A GENETIC ALGORITHM
APPLICATIONS OF GENETIC ALGORITHM IN AI
GA searches from a population of points, not a single point. GA supports multi-objective optimization (optimization based on a number of parameters).
Lit Study 1:
Use of Genetic Algorithm to produce multiple iterations of clothing in Fashion Design
HOW IS A TENSOR DIFFERENT FROM A MATRIX ? ( IN REFERENCE TO TENSOR FLOW)
A matrix is a grid of n × m (say, 3 × 3) numbers surrounded by brackets. We can add and subtract matrices of the same size, multiply one matrix with another as long as the sizes are compatible ((n × m) × (m × p) = n × p), and multiply an entire matrix by a constant. A vector is a matrix with just one row or column (but see below). So there are a bunch of mathematical operations that we can do to any matrix. The basic idea, though, is that a matrix is just a 2-D grid of numbers.
A tensor is often thought of as a generalized matrix. That is, it could be a 1-D matrix (a vector is actually such a tensor), a 3-D matrix (something like a cube of numbers), even a 0-D matrix (a single number), or a higher dimensional structure that is harder to visualize. The dimension of the tensor is called its rank.
A tensor is a mathematical entity that lives in a structure and interacts with other mathematical entities. If one transforms the other entities in the structure in a regular way, then the tensor must obey a related transformation rule. This “dynamical” property of a tensor is the key that distinguishes it from a mere matrix. It’s a team player whose numerical values shift around along with those of its teammates when a transformation is introduced that affects all of them.
AUTODESK DREAMCATCHER
Arguments for the incorporation of AI into design often default to concerns around replacing the human designer. While many elements that are commonly modeled from scratch such as brackets, adapters and stiffeners may be created more effectively by a system such as Dreamcatcher, complex elements and aspects that are difficult to quantify will require new types of interaction to leverage human intuition and computational rigor in partnership. Dreamcatcher is pioneering new methods for interactive synthesis and optimization with industry leaders from the automotive, aerospace and manufacturing fields.
ADVANTAGES OF GENERATIVE DESIGN
APPLICATIONS OF AUTODESK DREAMCATCHER
Lit Study 1:
Lit Study 2:
https://www.autodesk.com/customer-stories/airbus
CHALLENGES IN METHODOLOGY
We do not know enough about AI or coding AI to fully implement something specific to our requirements. Instead we are looking to adapt or modify existing AI codes to generate useful output.
Our basis of research will depend heavily on a software, process or code that works within our capabilities and coding skill level. For this reason, we are unable to narrow down our research at this point.
BIBLIOGRAPHY
- http://www.sagradafamilia.org/en/geometry/ ---- GEOMETRY OF SAGRADA FAMILIA
GENETIC ALGORITHM LINKS
TENSOR LINKS
GOOD RESOURCE FOR AI AND ITS IMPLEMENTATION
LITERATURE STUDIES
Comments
Post a Comment