Such variations are actually something humans have to take time to do.. It is able to make it within 1-2 minutes What keyword should I type? Let’s see if it works. It does not. You can stop now. Hello, Watsons! In this episode, we will discuss about the topic regarding artificial intelligence and architecture. The world has been in awe ever since ChatGPT has been recently released. Around July 2022 last year, a program called Midjourney was released, which can be used to create drawings. Such drawings created by AI even won first place in an art competition, which has led to huge controversies. Point. 1 AI and Architecture! Any Possibilities? To be frank, architecture can be considered as a field where it has contemplated how artificial intelligence can be applied since the 1960s. The logical aspects and imaginative aspects are both essential factors that we need in design. In terms of the imaginative aspect, when using a program like Midjourney, it can produce drawings that maximize and exceed the human imagination. The possibility of adapting this in architecture is infinite. At first glance, the examples takes form of a building, mushroom, and tree being mixed together. Honestly speaking, architects would record and draw a sketch whenever an image comes into mind. After the first sketch, during the process of becoming an architectural painting, a lot of ‘programs’ have to go through simulations. The ‘program’ I am referring to is its purpose. We simulate the purpose such as deciding whether this is a suitable space for a bedroom, or if it is appropriate to put the kitchen here, or placing an office or an executive office. This aspect is where logical thinking is necessary. Before that, The inspiration intuitively takes 1% and the rest would consist of 99%. effort. Regarding the first 1%, After seeing a video or image as such, we can also construct a nice building that matches. It is definite that this will influence in many aspects. In a way, you can get such influences from AutoCad as well. In the past, the tasks that require around 100 people were made possible with just around 25 people after a program called AutoCad was released. Just like how the workload for 100 people became more efficient, if the current workload for 25 people rely on AI, the workload will then require only 5 people. Specifically, rather than the later process, the initial plannings and conceptualizations can be relied on AI. The unfortunate part is that this part of the process is the most fun. The most fun parts can only be enjoyed by a small number of people. Alternatively, it can also be a world that everyone can enjoy. It can turn into a world where anybody can take part in. We would have to wait and see how it will be applied. Architecture is not something that’s executed from just an image. There can be several approaches. In the past, when clients are unable to express their ideas, after utilizing such programs and creating an image, they can convey what they want to the architectural firm or architect. During the architectural field back in the day, they would bring scrapbooks consisting of several images. One of my clients would bring a scrapbook with around 200 images in it. Among those, there were images that are hard to combine, and some would even contradict each other. We would have to sort them out in order to find and make the space that the clients wanted. For instance, after making just one image as such, the client could keep drawing for one year and rather present it first and request for a similar design. From the architect’s perspective, it would be an unfavorable situation. In other words, in the past, the architects worked on tasks that require their heads but now it turned into a situation where we only need to use our bodies. I am definite that there is a possibility of the formation of design firms that will specialize in those areas. Point. 2 Evaluating AI Architecture Judging from the samples, these works give out a very surrealistic vibe. If it was 10 years ago, this would be turned into 3D models before creating an image. Currently, the image is created first before the modeling process, by using pixels. When looking at such images, there are two main types, one made with pixels, and another made with vectors. When made with vectors, XYZ coordinates would be used to create 3D models which are then captured by the camera to turn into images. That’s the type of image we make with computer rendering in the architectural field. This image would be pixels, which are combined together on a two-dimensional paper. By doing so, it is much easier and faster for computer simulations since 3D modelings are no longer necessary. After that, if there is a preferred design, you can continue to the 3D modeling process in reverse, which can be more effective. To be honest, I personally believe that, there is not much difference between what humans create using their imagination and the images that Midjourney produces. The reason why the previous architects were able to plan better than normal people is because they have seen many works within that field, from magazines, their studies and in history. If that much data is stored inside your brain for more than 10-20 years, your imagination will be more improved as it gets mixed into our heads and unconsciousness. We can respond to the land that we’re given and come with new ideas based on it. Deep learning is actually the same exact system. In the case of ChatGPT, after we type in a word, it stochastically chooses probable words that may come out consecutively. I actually think human brains would work in the same way. But of course, when looking at the latest studies, our human consciousness is similar to the superposition of the two. I’ve heard of this saying from somewhere. It is not something you can make certain of using logical data from computers. It overlaps at a much finer level in the quantum domain It is a unique feature of the quantum world that does not produce definite results until observed. Because human consciousness comes from such place, I heard that some say it’s going to be a bit different, In that sense, some people may be able to refute that AI is completely different from the human consciousness. During the 1990s, when such debate existed during my college of architecture days, based on the conversations in MIT, an artist called Jackson Pollock coincidentally created a very abstract painting. Next to it was a painting created by a monkey using watercolors. When comparing the two, will the painting created from deep thought through human consciousness be any different from the painting that is randomly drawn by a monkey. That was the key agenda. Although the results may be similar, one is made from the creativity originated from the consciousness created in the quantum realm, and another is a result made from a similar simulation from a computer. However, these two aren’t distinguishable. I believe that we are at that point. Of course, although some several images made by the computer may be completely different from those by the human consciousness, if the results are different to differentiate, it would mean that since this is faster, more effective and affordable, there is a possibility that people will switch to this. This looked interesting; Apple Office Building. It is in a form of a cube made of transparent glass, with a curved cube version as well as a pyramid-shaped option. This design looks more Eastern, like an Apple office building constructed in China. There are designs with a literal shape of an apple. These variations are actually something humans need to take time to do, but since it can be produced here within 1-2 minutes, there would be a definite sense of competition. Therefore, we have tried an experiment. There is a project called ‘CNF House’ in our office. It is a house with 3 floors consisting of a terrace. The reason why I would like to test it out is because when we design such residences, we would design residences with similar purposes. When we presented it to the client, they commented that it nice but yet too straight, preferring curves and thatched houses. We received a request saying that they would like a house with more thatched curves. I uploaded this picture to Midjourney and typed ‘make a house with a curved roof’ as a command. There were 4 images that came out as a result. At first glance, they look somewhat similar. I continued to remove a few more keywords from the command prompt such as “3D printed, glossy white.” These images came out as a result. After looking through, the first and fourth images were similar to our designs in terms of the direction. The final results came out in these shapes. At first glance, the results came out similar to the actual design. Of course, I would not be too pleased to use what this produced, since there will be a lot of changes. When referring to architecture, most users would think of complicated building purposes. It would be difficult to put all those complicated requirements into keywords as if it’s a natural language and produce the perfect result. It’s just about receiving the images at first. When looking at the results, what AI produced and what created would definitely have differences nevertheless. However, during the thinking process, if it took us 6 months to come up with this design, with the help of this technology, we may have been finished within 3 months. And so, since the employees only need to spend half the time, the overall scale of the organization will happen to decrease. It could be also be an area of concern. Rather, until now, since Midjourney is still lacking for us to use, we would not get to that point yet. Before that stage, what we do is we use Pinterest. When searching for keywords in Pinterest, tremendous amounts of images will appear. Once you click on them, more images in a similar fashion will appear. After combining them and placing it in your head, we are able to make a design as well. It is different in my generation. In our generation, during the recruiting stage, if they told me to plan a staircase, I would start by sketching in my sketchbook. If we tell the newer generations to plan a staircase, they would type in ‘Stair’ into Pinterest and develop their design. As I mentioned before, since such processes made data more accessible to the public, the architectural field became a field where more and more people are capable of doing. Even people who did not attend architectural schools are now able to say that they can design or plan things. When looking at a group of architects, they would argue there are more important things, and refuse to acknowledge them as architecture. It is a situation with several debates. We can consider this as something that can occur over time. Point. 3 An Attempt at AI Architecture One of the designs I am currently working on is designing a house in an orchard. What keyword should I type? Lets type in ‘House’ and maybe ‘Frank Lloyd Wright style.’ Let’s see if it works. That’s totally wrong though. It just created a weird falling water pond. It is very different from what I imagined in my head. Let’s remove Frank Lloyd Wright. Is it okay to just type in Richard Meier? I don’t see much fun in this. It may be enjoyable for people who aren’t in this line of work, since I have my own way of doing things, it feels like I am talking to an employee that’s hard to communicate with. ‘Why can’t they understand what I am saying?’ I’m sure the interface will improve to be better. As time passes, it may reach to the level of Jarvis from Iron Man. Then, the story would be different. But for now, it’s too frustrating.. I’ve had that kind of experience before. When I first used AutoCAD in 1989, it was during the release of AutoCAD R10. Back then, I felt like AutoCAD was unable to keep up with my imaginations, so I ended up not using it. I felt too frustrated to continue using it. The moment I started using the computer again, was around 7-8 years after this. By that time, it was faster and more efficient than me just using my hands. And so, judging from this current level, not a lot of architects will be use this. I believe it may have to wait for a few more years. This software really is bad at designing, in my opinion. I believe the software needs a certain something. Although it is able to create something, it would need to be trained to differentiate between good and bad. To be honest, getting education from schools, are that type of training process where you receive evaluation. We get grades during school, enter and fail in contests. Such processes are actually one of our deep learning processes. Midjourney does not have anything like that. ChatGPT also have people who are evaluate whether their generated responses are right or not. Since Midjourney does not have that kind of process, what they create right now feels like something created by a student who is still trying to improve their design skills. It will probably still learn fast. But it still needs to go through that process. Back in the 1990s, when I attended classes related to artificial intelligence, the professor would always say that highly evaluated artists and people would survive, and that the intermediate class would disappear, such as the jobs that utilizes design to only a certain extent. For instance, even in architecture now, most jobs where you design a building according to the law may not need to be done by humans anymore, such as planning parking lots. These are done by artificial intelligence. Such jobs could be replaced. You should not type in any specialist architects in the prompt. If you type it in, it just picks out a representative work of the architect and alter it repeatedly. Is it being created? Is this it? It does not work… Hey, you can stop now. I don’t think you’re suitable for this. Come back after a few years. I think of it as a relief. Point. 4 The History of AI Architecture The reason why there has been many interest in AI architecture since before is actually when referring to architectural planning, there is a part where you have to plan intuitively and also have a logical and reasonable process. It is a decision-making process with these two combined. In terms of the work of artists, if the painting reflects their intuition and unconsciousness, since architecture has a function, we have to execute that functionality, as well as combine it with some intuition. Therefore, architecture planning originates from the combination of rational and intuitive processes. It actually first started to navigate with Peter Eisenman, an architect during the 1960s. While working on his doctorate dissertation in Cambridge, he wrote about an Italian architect, Giuseppe Terragni, who made reasonable designs. Since his designs are made to be reasonable, Eisenman thought that if this person’s designs are reasonable, he would be able to read through all the processes. That field back then was probably ‘Shape Grammar.’ Let’s say one particular architect named A would make similar designs every time. Just by one look, you can recognize that it’s their work, like Frank Lloyd Wright’s. You can “Louis Kahn designed this.” or “Ando Tatao created that.” This is because these specialist architects have the same algorithm in their heads. This is called ‘Shape Grammar.’ The grammar used to create form. So, when Giuseppe Terragni plans and designs, Eisenman thought he would have a very logical and reasonable approach, so he recreated his thought process by inferring. After creating the processes, just like entering into that person’s mind, if there’s elevation, he would divide it into 8 parts, with 6 on one side and leaving 2 behind, and so on. It is explaining the process of developing the elevation. After that, in the 1970s, there was George Stiny and William J. Mitchell. They both can be considered as the masters of Shape Grammar. When looking at their dissertations from the 1970s, they wrote about someone named Andrea Palladio. They used the square floor plan that he designed to explore and reveal what kind of logical process was used in his design. Since the 1960-1970s, this architectural field called Shape Grammar tries to figure out the thought process of one architect. In the end, if this technique gets passed on to artificial intelligence, they can create designs based on person’s style. William J. Mitchell’s prodigy, is Takehiko Nagakura who is a professor. He was my thesis advisor. It is interesting to study his floor plans. Floor plans are developed by computers. After designing one on the first day, we change on it on the second day, and develop it again on the third, and again on the fourth, and continue designing this for one month’s worth. Sometimes, after a month as passed, there are times where you realize you planned it wrong. For an example, when I first designed it, with the entrance placed on the south, after some thinking, since sunlight must come in from the east, I realize moving it to the west would be better. After moving it to the west, instead of designing it again a month later, if the entrance is moved to the right in the program, by simply pressing the enter key, since the design process that I’ve made within a month have already been recorded, the program was able to alter it within 10 seconds, by developing my design. In that sense, my thoughts and the computer can be seen as mutual communication, needed to get to know my thoughts, through that software. The conclusion of that professor’s dissertation is that after 10-20 years, when working on your designs through such programs, among the 1-100 alternatives, if I like No. 1, No. 50 and No.70, the computer will combine those 3 to make another 100 alternatives. If I prefer No. 35, 72 and 93 among the new batch, the computer will be able to combine and design them again, through collaboration. Even in the case of Midjourney, when it creates an alternative, it made it possible for me to have choices, such as choosing Version 1, or Version 4 to develop it more. I believe we have just reached to this level, but since this is still pixel-based, it only creates images. If it gets more developed, it could go into more depth. When the students read the dissertation about ‘Shape Grammar’ they continued to study how to implement a two-dimensional design into three dimensions, and figure out what their way of thinking was. Although this is what the computer science field mostly do, when doing so, what made it possible to adapt is the architectural planning. There were such processes like this. And so, during the 1990s, the people used to go to libraries for research As we enter 2000, since the invention of the internet, we can find several documents just from searching one keyword. In the 2020s, Pinterest fulfilled that role, and Midjourney is gradually on the process of becoming an interface with the role of the collaboration between humans and technology easier. In this episode, we tested on how AI can be used in artificial planning, and discussed about the meaning it holds. I will see you in the next episode.
