AI and machine learning in architecture

Over the last few decades, computation has given architecture an edge advantage over the understanding of the structure, solve complexities with ease, and better utilization of resources. The evolution of the application of computer or computational design in architecture is divided into era’s which are: 1^st era is of 2D Drafting, 2^nd era of 3D Modelling, 3^rd era of BIM, 4^th era of Algorithm-based design, and finally, the 5^th era of Machine Learning (ML) and Artificial Intelligence (AI).

Before understanding the impact of ML and AI, it’s important to understand the evolution of the application of computation in simpler terms. 2D Drafting (software like AutoCAD, Sketchpad, etc) comprises of drawing of lines with help of computation, then in 3D Modelling (software like Blender, Maya, etc) the surface, volumes, NURBS, meshes came into play.
When it came to BIM (it’s a workflow, but the most prominent software that helps in BIM are Revit, Tekla, ArchiCAD, etc), an object can store data which can be used productively, for example, a roof when made, the data like its area, volume, physical properties, materials, texture, etc are stored within it, with can be used in BOQ, structural analysis, etc. Algorithm-based design (software like Grasshopper 3D, Dynamo, etc) came to play after the popularization of works of Zaha Hadid and the parametricism movement.
In algorithm-based design, a design is made with some input parameters and let the computer give design output as per the direction coded by the designer. And finally, the era of ML whose concept exits from the late 90s, but the application of it is happening now. Software like Lunchbox, Finch 3D, Sketch Graphs, etc is being developed based on ML.

Machine Learning (ML), which is also known as Statistical learning, is a type of Artificial Intelligence that utilizes a set of data to predict the result with a certain percentage of accuracy. It requires a training data set (larger the data set, more accurate is the result), based on which ML gives the output. Several mathematical models like Artificial Neural Networks (ANNs), Support Vector Machines (SVMs), Bayesian networks, Radial Basis Function networks (RBFs), etc are used under ML to achieve different types of results.

So, here is the list of activities with the mathematical models that will generate output.

1. Generating Design Concept from the client’s demand

The ML will generate the concept of the building from the description given by the client as the input. The ML model that helps to get such output is Artificial Neural Networks (ANN) which will capture the data patterns and correlations concealed in the data. ANN will give many conceptual ideas based on the description and the different contexts related to the project. This will give the architects to explore more concepts and find the best one without spending much time in research and context-based information like the vernacular buildings of the site. After just a few years only, we can also find ML models that will generate text to 3D Models which will further increase the explorations and exposure.
When the architect is using ML to generate concepts, ML also enables the user to get recommendations in the process of design. ML will help the architect to clarify his design intents and recommend choosing truss types, floor plan layouts, and façade treatment.

2. Mass customization and urban planning

In the case of urban planning and mass customization systems, the style transfer technique which uses Convolutional Neural Networks (CNN) comes handy. CNN is mostly used to analyze visual imaginaries and uses neural links, like our brain’s image processing technique from the visual cortex. To use this technique in the urban design field we need to produce the building styles we need, a base design on which variations will be performed, and a previously trained neural networks panel. Architects must have control over the output and use ML just as a tool, as this system may produce undesirable variations.

3. Program synthesis

Algorithms are required to produce design at various stages of an architectural project. Order to create a program that will solve a particular design intent requires knowledge of programming, and in some cases, it’s difficult to find Application Programming Interface (API) experts to execute the idea. In some recent developments, an API expert is made which will guide, recommend and predict the step of the designer who uses it. This kind of system can be seen in auto-completion of texts while using G-mail. The API expert made with ML will suggest to the user which API is best suited for a particular problem, identify bugs and fixing them, and finally learn algorithms that solve a particular problem.

4. Analytical Modelling

Modeling is nowadays an integral part of architecture projects. It let us design a building in detail, run simulations with computations, look for clashes in between members, and prepare views before the building is made on site. These processes require a lot of time when the geometry becomes complex and heavy for the file to run and execute simulations. Due to this, we need to have low poly models of the heavy models prepared by us and lower down the size and number of geometry present in the file for faster simulations. These processes could be carried out with AI, and this system will also segregate different members like beams, columns, floor plates, etc. Similarly, the extraction of geometry could be done from satellite images for urban scale modeling and planning works.

5. 3D Modelling, Production, and Labelling

When it comes to production and labeling tasks, Generative Adversarial Network (GAN) model is the appropriate model to perform the task. GAN is an ML model, developed by Ian Goodfellow which is capable of unsupervised, semi-supervised, fully supervised, and reinforcement learnings. The GAN model can be trained with data consists of names of spaces, their geometry, etc based on which it will generate and label appropriate spaces. Running this model with style transfer will generate a better model that will be easy to understand.

6. Finding hidden correlations

While evaluating a design, based on certain parameters using an ML model, it becomes important for the ML model to separate the data that are used for evaluation from the rest of the data, which will improve the efficiency of the system. Models like Surrogate-Based Optimization (SBO), ANNs, SVMs, Gaussian Processes (GPs), Radial Basis Functions (RBFs), and some ready-made tools like Opossum, Lunchbox are used for approximations, segregation of data, and evaluate the data. This will result in the execution of faster and meaningful simulations and also will generate a similar type of design solution for the user.

The application of ML and AI is all around us, and its application in the field of architecture is vast. Objectives like the development of an adaptive façade control system, the creation of desired materials, and the development of more smart systems can be achieved with ML. Introduction to ML in architecture will not let the creativity of the architect down but would allow the architect to explore possibilities and find the best, faster simulations, automatic organization of files, and much more, its just a matter of time after which we just need to ask the correct question, to get the best of ML.

With all technology, there come drawbacks. The major drawback of the application of ML in architecture is the training set of ML systems needs to be continuously updated, else it will generate a similar result. Unfortunately, we don’t have any large training set available in architecture. An unbalanced dataset might make ML biased towards particular decisions and research on minimizing the biased character from unbalanced data set is going on. There are several abnormalities related to ML which are being solved, without which application of it in architecture can be disastrous.

The paper “The role of Artificial Intelligence in architectural design: a conversation with designers and researchers” by Giuseppe Gallo, Giovanni Francesco Tuzzolino, and Fulvio Wirz comprises of interviewing the leaders of the architecture industry and trying to figure out the impact of AI in architecture. It’s quite interesting to see that out of 10 interviewed leaders, 4 industry leaders have put Machine Learning, Other computational methods, and Digital Manufacturing in their first place of choice while ordering the technologies that will prove their usefulness in the field of architecture in the next 10 years. The other options that are put in front of them are Building Information Modelling (BIM), Internet of Things (IoT), Augmented Reality (AR), and Virtual Reality (VR). So, it’s quite clear that the amount of impact of ML and AI that the field of architecture will have in the next 10 years has a very high probability.

There are lots of architecture tools that use ML and AI technology to some extent and have become a part of our daily life. Like Google Maps, E-mail filters of Google, Linked In, Google Search Algorithm, etc. In architecture, some of the software that uses ML and AI is Unity 3D (which uses AI to find the shortest distance of fire exits), Lunchbox (it uses general ML for regression analysis, clustering, and networks), Opossum (uses ML for functional evaluations, near-optimal solutions, and simulations), Project Dreamcatcher by Autodesk, etc.

Being surrounded by technology, and predicting the possibilities of the future. The question of “What’s next?” remains the same all through the ages and that’s where AI and humans have the gap.