Digitally. How the age of neural networks is changing architecture

The future is already here

Ever since Midjourney, DALL-E, and other software officially launched beta versions and became available to users, there has been a boom in the generation of images based on linguistic descriptions. Millions of users around the world have started experiments to create realistic, artistic or futuristic drawings. It seems that the limits for imagination simply do not exist either in stylistic directions, or in storylines, or in color solutions. Upon specific request, the neural network can reproduce anything: portraits and landscapes, plots from comics, famous films or cartoons, post-apocalyptic art or science pop.

IT technologies have long played a key role in countless industries, including architecture, design and the visual arts, but artificial intelligence appears to be changing the very nature of creative processes. In particular, new platforms for creativity have become a mouthful of inspiration and visualization of the innermost desires, which sometimes experts simply do not dare to demonstrate to the general public.

Thus, browsing the Internet for architectural creations created with the help of AI, here and there you come across something that seemed impossible to combine. Entire underwater cities in the style of steampunk, tall skyscrapers as if designed by Gaudí himself, surreal buildings with intricate facades and sophisticated elements - new tools open up unlimited possibilities for creative imagination and self-expression.

However, there is a place for completely realistic objects that can be confused with the work of a real architect. And most importantly, AI can offer several options at once upon a certain request, and the generation will take no more than 5 minutes, while a person may need hours, weeks, or even months to offer the customer several options for architectural solutions. And the question arises: will not artificial intelligence replace the work of real specialists in the future?

Prophecy and pragmatization

Matias del Campo, an architect and associate professor at the University of Michigan, says that he first started using AI in his work about 4-5 years ago. Then he mostly used GAN (Generative adversarial network) and Revolution Networks. They are more serious AI algorithms of classical machine learning and were created even before the appearance of the well-known Midjourney or DALL-E.

The essence of the idea is in the combination of two neural networks, in which two algorithms work at the same time: a "generator" and a "discriminator". The task of the generator is to generate images of a given category. The task of the discriminator is to try to recognize the created image. GAN can be used, in particular, to improve the quality of blurry or partially corrupted photos.

Matias del Campo mentions that he applied an algorithm to correct the images. "I don't think the experience was very successful, but you know, when it's just starting, it rarely works well," says del Campo.

The architect learned about Midjourney later in the company of media artists. He says that artists are usually the first to discover interesting technologies and apply AI in art.

"I found it very interesting. If you enter a sentence in Midjourney, the program creates a group of matching images, then you can work with one of them and get the next group of images generated based on the selected one. Architects traditionally work in the same way: you sit down with a piece of paper and a brush and start creating a sketch, then you try another and another. You are trying to find a final solution that you will like. It is unlikely that you will get a perfect drawing on the first try. AI also works according to this principle. So, Midjourney has something of an architectural process,” says the architect.

According to Mathias, the volume of data with which artificial intelligence works is more than 5 billion images - this is a huge amount of knowledge about human culture.

"This is a completely unique application. But everything just seems amazing. If you are an architect and start exploring AI as a tool, you realize that it doesn't make any sense because there is no semantic information in the Midjourney components. He does not understand what a living room is, what a bathroom is, what a toilet is. The application is very provocative, interesting, beautiful, but it is not aimed at creating functionality or, for example, sustainability and environmental friendliness," says del Campo.

However, Mathias sees prospects for the development of AI for architects, if in the future it will be integrated with information about the environment, places where construction is planned, environmental conditions, project financing and other "more pragmatic things".

Only with the analysis of the specified data will the use of AI make sense, and the program will be able to offer the best solution. Another approach that can be effective is to combine the two systems. First create the image in Midjourney and then send it to another network that will analyze those images and draw conclusions about the quality of the design.

There are two things that AI is very good at: prediction and optimization

“So you can take the images created in Midjourney and feed them into another AI program that will optimize or analyze the illustrations and make suggestions to the architect about whether those solutions are feasible for the project. Yes, I think Midjourney can become a part of our daily work, but only as one more tool in a whole range of tools we use. This can allow us as architects and the architectural discipline to make progress in what we offer to the public,” says Matías del Campo.

A new dimension of investment (in yourself)

When it comes to modern technologies, their creation and development, when we start a discussion about new developments and achievements of mankind, we involuntarily look back towards the West. For centuries, Europe stood at the forefront of evolutionary processes in human life and its interaction with the environment. In the last two centuries, some countries of North America joined it. But an amazing fact: Ukrainian architects used neural networks in their work more than 10 years ago.

In 2011, the founder and chief architect of Dmytro Aranchii Architects, Dmytro Aranchii, together with his team, presented the renovation project of the Kyiv-Passenger depot in Kyiv.

The huge territory had to be reorganized into a comfortable space that would connect two large microdistricts — Chokolivka and the Polytechnic Institute microdistrict — thanks to rationally organized roads, pedestrian zones, and bike paths. The experts used crowd simulation, which made it possible to generate pseudo-optimal paths between two neighborhoods cut off from each other.

"I wouldn't narrow down the concept of artificial or computational intelligence to 'trendy' applications like Midjourney, DALL-E and the like. This is just the moment when the most fun and spectacular, but not the fact that the most productive part of what was happening in the professional environment broke through to the general public. However, the architect is increasingly turning into an operator of technology, which is becoming more and more complex and "smarter".

Our team, for example, intensively uses artificial intelligence in the work of our workshop. In September 2022, we won a closed competition for the design of the apartments, drawing out the layout by hand and generating the visual series with Midjourney. We also actively connect artificial intelligence to post-processing, working with the environment. Currently, we have started experimenting with the generation of three-dimensional geometry and we have high hopes for this technology.

Nowadays, the architect increasingly has to understand a huge arsenal of tools that make life easier. The time that was spent to do something manually can now be invested in finding and learning a tool that will simplify/optimize/give unprecedented results in design," says Dmytro.

He notes that the further, the more the architect has to rely on at least elementary programming and, most likely, it will become an integral tool of modern man, as mathematics once became.

"In this case, the architect can create or modify algorithms, including artificial intelligence, to solve unique problems in architecture, which, after all, is much more reminiscent of haute-couture than pret-a-porter," says Dmytro Aranchyi.

The future is based on the reconciliation of conventional and anthropomorphic intelligence, and AI itself could be considered as an expansion or improvement of our capabilities

"Advanced global practices have been automating large, especially repetitive processes for a long time," adds Aranchii, "in particular, they are also automating the creative component. The cornerstone seems to me to be feeding artificial intelligence with algorithms for form creation and engineering solutions so that they operate not only with images from Google, but also understand the logic of design - from the pattern on the curved facade and the profiled glazing system to the perception of the load and energy efficiency nodes by the structures. We live in the most interesting of times, you will be surprised what will happen tomorrow, and maybe even this afternoon."

In addition to global issues, such as the prospects for the development of the use of artificial intelligence in the field of architecture and design and the revolution that computer programs will be able to implement already in the near future, there are other, more down-to-earth ones. Like copyright. Who should it belong to? Directly the neural network that reproduced the image? Neural network developers? Or to the architect who made the request? Or the authors of those works on the basis of which the neural network generated these images?

Dmytro Sivak is also concerned about these issues. He mentions that previously it was not allowed to place AI-generated illustrations on photo stocks, because there was simply no understanding of who was the author of the work.

"Currently, it seems that there is already permission to place them on photo stocks, but there is still a lot of misunderstanding about what will happen next. I am more pessimistic - it will be sad for all of us. In short, the profession will become low-paid. Some processes will be done much easier. Mass dumping will lead to the fact that we will have to work more for less money," the architect shares his concern.

Obviously, one concern flows into another. Will specialists be able to stay in the profession? Will the new technological boom cause the loss of jobs for tens of thousands of specialists in design, engineering and other diverse fields?

Dmytro Sivak explains his pessimism in this matter with the fear that soon architects and designers of elementary levels, who have just received education and graduated from universities, will not be able to give themselves advice so easily, because what they have studied for years will be able to do for them artificial intelligence, the learning of which is carried out much faster.

"The fourth version of Midjourney is already a head higher than the previous ones. This is like a student of the 1st and 5th years. Neural networks learn ten times faster. Nowadays, they generate impressively realistic drawings and photographs. To work with images in this way, people study for 5-6 years, and not everyone succeeds. A person draws one or two options and gets tired. You need a break, to think, to find inspiration. And Midjourney makes 10 variations in 5 minutes, and each of them is brilliant in its own way.

It is hoped that this will become a tool and, as before, architects will do their work. Neural networks make a lot of cool pictures, which you then have to model, outline and do a lot of additional work. You have to be a highly qualified specialist to work with it later. Another thing is that AI-generated images can bring a person out of a creative stupor, and for some it helps a lot.

However, if earlier, starting from junior year, you could work and earn money, now you are not needed. You will study for 5 years at the institute and then you will work for free for 5 years, because your work can be done faster and more efficiently. Earlier, the improvement of elevators led to the fact that the need for elevators disappeared, the development of communication caused the disappearance of the telephone operator profession. But it affected a relatively small number of people, it was not so global.

Now, Notion (a smartphone app that provides databases, kanban boards, calendars, and reminders. —Ed.) has already started embedding neural networks, and we're losing copywriters and a bunch of related jobs. I may sound like a psycho saying "Repent", but very soon many professions will be useless. A lot of unemployment is waiting for us," Dmytro shares a disappointing forecast.

A neural network can create an infinite number of image variants, and the most unpleasant thing is that it can do it without you and beautifully

"Now there is a neural network that creates videos on demand. And there is also a neural network from Google that creates 3D models on demand. You enter, for example, the text "vase with fruit", and it immediately creates this vase in 3D. Yes, now neural networks create just pictures, and they can help architects in the brainstorming process. But after half a year, we will be able to receive a 3D model of the building on request, and in the future - to download the plot as well. And in another year or two, you can get all this with planning. And that's it, you have 5-10 options for a ready-made building. For 90% of people, this will be enough. People don't always want something complicated. And why then an architect, if for 10 dollars you can get a complete package of drawings? And, most likely, such startups will appear," says the architect.

Matias del Campo does not share the concern of his Ukrainian colleague. In his opinion, such discussions have been taking place for more than 250 years and are being conducted in parallel with the development of certain technologies.

“In the XNUMXth century, artists were very confused by the advent of photography, which could potentially put them out of work. I don't worry about that because people have a lot of imagination, which means you can always think of ways to use this technology. It's just a change and you have to adapt to the situation. People adapt very well. Our profession will change, but it will not disappear,” says del Campo.

Dmytro Aranchii: The digital era is creeping up quickly and imperceptibly

"Leading scientists agree that, first of all, artificial intelligence applications will replace low-skilled workers, such as sorting and unloading workers. This is already happening in the same Amazon. The management of Tesla and other factories is trying to automate work as much as possible, leaving people on the ground rather for quality control. Only opinions about the rest differ.

Some pessimistically believe that at least a person will run out of work, and he will be bored, losing the meaning of life; others, on the other hand, think about how a person deprived of routine operations and responsibilities will turn into an absolute creator: he will create more and deal less with utility calculations on weekends and household duties.

Be that as it may, it seems to me that there will be a little bit of all of the above, and it will happen so gradually that we will not notice the difference, and instead of strikes by people who have lost their jobs, we will see the movement of workers, as in the past, from the fields to factories, and from factories to high-tech production and the sphere of digital services."

AI vs Human beings

2057 year. People have practically stopped communicating, handing over this function to high-tech robots. They are stronger, younger and can do things that their owners cannot even dream of. At the same time, computers are becoming smarter, learning, they have feelings and emotions. In addition to kindness and love, they can feel envy, greed, contempt, aggression. Due to accumulated grievances or huge ambitions, the confrontation between robots and humanity begins.

It was an average plot of a Hollywood movie with exciting special effects in the cinema during the viewing session. And these techniques were used by screenwriters many times, because even today humanity wants to dream of what the world will be like after its robotization.

And although Judgment Day has not yet come, heated discussions continue to rage around the question of who will be the winner in the fight for creativity: human creative imagination or super-speed capabilities of computers? Answering this question, Matias del Campo turns to the British scientist and theorist Margaret Boden, who has been working on AI research for more than a dozen years. At one time, she announced the verdict: artificial intelligence itself is absolutely not creative.

"And I understand why. All these artificial intelligences and programs, they are all inspired by what our brains have already generated. We do not understand how creativity works, and we do not know how to describe these processes mathematically. Every time we see something new, we try to put it in a certain context, just to identify it,” says the professor.

Dmytro Sivak: Neural networks showed the farthest and dirtiest thoughts of architects

"Until today, neural networks have shown the farthest and dirtiest thoughts of architects. In the sense that the pictures that appear are more related to art deco and classics. Very decorative architecture, which is difficult from a technical point of view. Nowadays, however, the time of minimalism and neomodernism prevails. And when people realize that we can design complex architecture - classic, art deco, only 5 times more sophisticated - they will start doing it en masse. And if 3D printing begins to be used on a large scale, which allows us to make new complex forms cheaply, we will see a new round of architecture. Or the return of the era of the 3th century. If you have a lot of money, but no taste, you build yourself a castle on the outskirts of the city, a la "I am the king". The high cost may hold you back. However, the combination of these technologies — AI and XNUMXD printing — is everything. Kapets," says Dmytro Sivak.

AI and urban space

For now, architects are pondering the future of their profession and debating how the new tools will affect the industry as a whole. Will they make some processes easier for them, or will they be able to completely replace architects in their work - let's try to consider this issue even more broadly.

Perhaps soon neural networks will reach such a level that they will be able not only to reproduce individual objects: country houses, high-rise buildings, skyscrapers, bridges. Is there a chance that technology will go so far as to be able to design entire cities? And, for example, on the request "small city, 80 thousand population, moderate climate, the distance from the nearest big city is 15 km" a few minutes - and everything is ready. We receive a project of an entire settlement with all communications and planned zones — from the central lively part with shopping areas to residential quarters, parks and squares, recreation areas, sports facilities, planning of the entire transport network, etc.

Dutch architect and urban planner Fulko Treffers

"I'm a big fan of using computers and data for design. Data is valuable, and computers are very smart. Sometimes even smarter than people. They can calculate a lot of things and that really helps. In our planning, for example, knowing mobility data or heat and temperature data or water systems is very important, and if the data is correct, all these systems can be regulated much better during design. This especially applies to the design of transport infrastructure.

For example, you enter the entire network of existing streets into the program, and the computer can calculate which street it is on and what can be done better. Computers can offer something more efficient than we could do ourselves. But when we discuss what style it should be, what materials are available to us, and other things related to taste and the reflection of a country's memory and identity, I don't know how to ask a computer to give me some advice. It is still more about discussion with the public and specialists. Maybe you can use AI in the process, but the fundamental issues have to be discussed and decided by humans.”

Practical embodiment and search for inspiration

"I don't want such a future," was the first thing said by the architect of the archimatika company Mykola Morozov, who we asked to generate an image in artificial intelligence on the request of "a residential quarter of the future" in order to conduct a creative experiment.

"When you ask Midjourney to generate something futuristic and something from the future, the program offers something between Avatar, Zaha Hadid, Kyiv buildings and a typical Hollywood movie about the future. In people's minds, the future is mostly still something like The Fifth Element, just not as terrible. These are tall and large volumes, massive and monstrous. About five thousand people live in one building, and in order to somehow live there, trees are installed on the balconies. No one suggests that Ian Gale be remembered in the future," Morozov notes.

According to him, for the further generation of images, artificial intelligence is not yet able to respond to clarifying requests, for example, if it is necessary to lower the height or make a more quarter-sized building.

"The program does not yet understand this and continues to generate illustrations of the height format of heavy and massive housing. You look at the picture and it seems interesting. An interesting, unusual shape. In your perception, it is similar to the architecture you have seen in the movies. But I wouldn't want to live like that. Not only did we generate the query you asked for, but we also tried to use artificial intelligence during normal operation. We entered a request: a residential complex, average floor space, plaster, metal-plastic windows, 50% glass, and he started building us... Poznyaks," added the architect.

However, Morozov, like the rest of his colleagues, is not inclined to believe that artificial intelligence will deprive him of employment in the future.

"Any developing software is actually an improvement of the architect's work tool. All the programs in which we even now work — Revit, Archicad, 3D Max — are basically the same, a more convenient tool for the work of an architect in the 20st century. The architect of the last century could not even think that there could be a program that could comprehensively build his ideas. Perhaps in 30-XNUMX years, AI will also become a common tool. The only thing is that the architect will still remain a specialist who comes up with the most optimal solution. The computer will simply help him find it," concludes Mykola Morozov.

 

Main artwork: Atelier Olschinsky Projects

 

See also:

Metaversion of the new world: from interactivity to virtuality

A painting created by artificial intelligence won an art competition

"Environmental and moral crime". The urbanist criticized The LINE vertical city project in Saudi Arabia