Architecture in the Age of Artificial Intelligence

With the rise of digitalization in the early 2000s, architecture entered a significant transformation in terms of production, self-expression, and representation tools. Today, we are experiencing a completely different process that is shaking the industry: the era of artificial intelligence (AI). While “artificial intelligence” is a broad concept, how can we redefine it specifically for the field of architecture?

The new century, often referred to as the “millennium” but perhaps more accurately the “third millennium,” heralded new horizons in architectural production, as in many other fields. However, parallel to the process we call digitalization, the process of “individualization” also seemed to accelerate during this period. Among many developments and regressions, considering this “digitalization x individualization” pair together is appropriate, as they mutually reinforce each other.

These processes required technological support, both in software and hardware. In technological development, hardware deficiencies are remedied with software, and the thresholds that software cannot overcome are addressed with hardware advancements. Therefore, this pair also had to develop in an important and mutually reinforcing way. As a third pair, we can add “human x machine” here. Everything said for the previous pairs applies equally: an important pair that feeds and advances each other.

Today, we think we are witnessing the development of artificial intelligence, but perhaps the energy of the “human x machine” pair has shifted to the machine side as a reflection of a blockage in human development. When the machine strengthens, it will likely transfer its power back to humanity.

Here, artificial intelligence predominantly represents the software side in the machine’s development system that depends on the software and hardware pair. In recent times, we have seen how developments like quantum computers, VR/AR, and others have created excitement but could not progress quickly enough to affect our habits. Because progressing and obtaining practical results quickly with existing software is not possible. Therefore, the power flow in the background of the machine shifted to the software side, and it’s natural that AI studies are now starting to bear fruit.

Does architecture not have an important pair in terms of computer-aided productions? Of course, it does. We can present this as the “simulation x application” pair in architecture. If asked what will be the result of AI support, which is especially effective on the simulation side, in architecture, one might think we are starting a software update in the field. Designing with smarter (although the term has become overused) programs through artificial intelligence could be the next step in architecture.

The Image of AI Today

Today, there’s an “AI image” stemming from everyone’s state of “knowing everything.” The most frequently mentioned platform in the field of artificial intelligence right now is ChatGPT, developed by OpenAI. The fact that a major company partially owns it may be a bit unsettling, but this platform is the closest candidate to becoming AGI (Artificial General Intelligence: AI that can develop capabilities beyond human intelligence). It currently has a public demo, and we interact with its fourth version.

You may not have opened and used ChatGPT, but rest assured, it has indirectly, perhaps even directly, written an email response to you or communicated with you through some data input. This might cause a shiver, but it’s important to recognize its pervasive presence. An initial test might reveal that while ChatGPT boasts of knowing everything, it actually has fragmented knowledge, filling in gaps with plausible content and strong rhetoric. It’s like that impressive friend who seems knowledgeable but sometimes gets facts wrong. When caught in a mistake, it has a humble tone, apologizing instead of trying to dominate—a seamless approach in terms of customer relations.

The End of an Era for Certain Programs

AutoCAD was a revolutionary program during its inception. With artificial intelligence, which applications or programs like AutoCAD will see their era come to an end? We are currently slaves to cumbersome and tiring programs in design, prototyping, visualization, and construction. Using the analogy of “literature x printing press,” many firms employ workforce in “printing”—implementation documentation and visualization—rather than the “literature” of architecture, such as design philosophy and ecological consciousness. Naturally, the programs they use are oriented towards this.

The significant workload in implementation documentation and visualization needs to be shifted to AI-supported processes so that architects can allocate necessary time to design philosophy and other important matters. This means that programs currently used for implementation drawings and obtaining impressive visuals will either become nostalgic antiques or transform completely. For example, Adobe Photoshop has undergone a significant transformation, and companies like Microsoft and Adobe have caught the AI wave. However, large companies may not adapt quickly enough due to their ingrained habits.

Specifically, BIM programs led by Autodesk Revit seem quite outdated. Although there isn’t a widespread AI-supported alternative in BIM yet, significant work is being done in places like China. Visualization leaders like Chaos have uncertain AI integration paths, while chip manufacturers like NVIDIA are leading in hardware for the AI era. Rhinoceros, an important tool in post-spline architecture, doesn’t yet have an AI-supported software claiming its space, but developments like text-to-3D continue to advance.

The Professional Impact of AI

Professionally, what has artificial intelligence brought into our lives, and what else awaits us? What should we fear? The requirements of the profession remain the same: creating designs by considering the client, user, immediate environment, and nature; development, implementation, transformation, etc. Sometimes this involves creating a compromise environment despite the client who provides the financing. Producing and presenting visual and technical documents for compromise, and having options according to different desires, wishes, and needs, are necessary.

In light of all these questions, artificial intelligence promises—or at least smart usage of it will help—that if simulations can be created quickly, there will be time to think. Perhaps it will not be daunting to tear down a simulation and rebuild it from scratch. Design and other processes are extended with cumbersome tasks that cause simulations to lose adaptability according to changing problems, trapping them in the requirements and priorities of a certain time frame. Removing unnecessary requirements and priorities allows architects to focus on where they should actually spend time.

Curious individuals who educate themselves need not be afraid. In fact, some of the frightening “monsters” created by this period will gradually disappear. For example, the once-favored profession of “data scientist” was touted as one of the most important future professions. Recently, “prompt engineering” became popular. However, AI doesn’t want you to jump through hoops to understand it; it’s being developed to be in dialogue with you and to understand your requests.

For instance, when producing visuals for social media with ChatGPT, efforts previously required to get desired results with “word salads” on platforms like Midjourney turned into a more enjoyable process of making literary space descriptions. By exploring literature on space descriptions, one can produce scenarios through dialogue. This indicates that professions like “prompt engineering” may gradually disappear. The place of parametric design and coding experts might be taken by professionals familiar with literature, art, environmental awareness, understanding the psychological effects of space on humans. While there will always be people handling complex underlying tasks, it’s unnecessary for everyone to know coding.

If something has a high learning threshold, it indicates room for development; there are unnecessary bureaucracies and sluggishness that should be automated.

Adapting to AI in Practice

Are there applications or programs that were actively used before but have been abandoned after the introduction of artificial intelligence? While no programs have been entirely abandoned yet, there’s a noticeable decrease in the usage frequency of visualization tools like Enscape and Twinmotion. Heavy visualization processes done through 3ds Max have already diminished.

Now, more sketches and models are being made; traditional tools like pens and rulers are being brought back into use. For tasks like plan drawing (currently done with Revit), which might be delegated to AI in the future, certain people are still employed, so nothing has been entirely displaced yet. Programs like AutoCAD are used for specific tasks like examining rules and frameworks provided by public authorities.

New programs being used include Midjourney, DALL·E, and ChatGPT, primarily because of their speed in visualization. Midjourney is more “creative” but communicates in a more obscure language, while DALL·E and ChatGPT allow progression through dialogue but still have room for improvement in image processing techniques. Plugins like ControlNet with Stability AI-based software enable viewing sketches or 3D models in different materials and situations. Problematic parts of obtained visuals are corrected using “Generative AI” in Photoshop. There’s also LookX, which gives surprising results in important studies in China; it has high potential but results will be obtained in the longer term.

Voice dialogue with platforms like Pi.ai is enjoyable and transformative, though patterns and repetitions can sometimes break the magic. Good results are obtained from text-to-video and text-to-3D platforms like RunwayML and PikaLabs. Given the constant evolution, time must be allocated wisely, and strict selection processes are necessary.

Midjourney as a New Representation Tool

Programs like Midjourney are seen as new representation tools, similar to how a book transforms into a movie, telling stories from another perspective. The Midjourney platform was something the industry was ready for because it uses space description as a production method. It’s like an illustrator who draws what you describe, though often the results are “not quite there but not bad.” It’s important not to get carried away and to strive to get closer to what is desired.

The integration of DALL·E’s third version with ChatGPT 4.0 is a significant development. Unlike Midjourney, DALL·E 3 doesn’t consider commands to imitate specific architects or artists, encouraging users to understand what they truly want to achieve. Dialogue with AI means being familiar with literature; examining things that arise from different words is crucial. Using different word variations can help understand their implications in visual production.

The Evolution of Representation

Representation in architecture has always been a complex issue. Unlike art, it doesn’t have a state in itself. While the object of representation in art can be considered separately from its creator, this separation is challenging in architecture. This hinders the design and production process from being collective. AI platforms like Midjourney produce selections based on vast data, resulting in entirely collective outputs where the creators are ambiguous.

This multitude and ambiguity might seem chaotic, but it could lead to architecture becoming more collective, free from the creative ego, and inherently self-sufficient. The infinite representation forms in AI and the ambiguity of their creators might open new doors, allowing for more collaborative and collective design processes.

Working with Midjourney

Copying is essential in the training part of working with platforms like Midjourney. However, it’s crucial to move beyond imitation and focus on the “nature of things” by starting space descriptions. Copies can create dangerous and addictive blurs, leading to average-oriented acceptance.

A healthy usage method involves asking AI what it understands about materials and their forms, essentially reverse engineering. This helps in examining and understanding architectural works or visuals. It’s important to avoid merely taking the most flashy visual versions created by “word salads” in prompt engineering.

Learning new languages or terms can aid in better communication with AI platforms. Sometimes, common words have dominant effects that need to be navigated carefully. Exploring different styles, textures, and architectural movements by breaking them down into components allows for hybrid integrations.

Curation is vital at every layer of the process. From initial productions, selections are made, errors corrected, and quality controlled. A significant portion of productions might remain archived, emphasizing the importance of selection in achieving desired outcomes.

The Data Issue

The subject of data is crucial. It’s undoubtedly possible to collect data ethically, but we need to ask ourselves if we want to do this ethically. Since AI programs are text-based and creative, accumulation through reading, writing, and observing is essential. Obtaining an ethical product is open to debate due to the trio of text, collective database, and algorithm involved in production, making control challenging.

When using AI, every data input is embedded in the program, contributing to the collective database. This is similar to writing a thesis or paper, where integrating others’ ideas without plagiarism is essential. Understanding AI’s operations requires experimenting with all its aspects.

In isolated environments, unorthodox methods might be used to understand the nature of things, always keeping ethical considerations in mind. Outputs obtained through certain processes are used to make ethical affirmations. For instance, using a well-described building from a magazine to examine results and then deleting it ensures that ethically questionable outputs are not utilized.

New innovations allow users to have their productions removed from AI platforms’ base working libraries. If not just a PR move, this is a positive step towards addressing ethical concerns.

Authorship and Future Implications

Authorship is currently an issue with weakened ethics and invalidated rules. It will remain important in the future, but perhaps not under the current guise of patent protection. Since information can be accessed quickly, unreferenced use of someone else’s work can be automatically challenged. Ethical and aesthetic boards in future municipal structures might address such issues, especially as AI becomes more integrated into processes.

Projects like “Content Credentials” aim to trace the source of visuals, emphasizing the importance of publishing and establishing authorship. Sharing ideas and productions publicly acts as a time marker, aiding in the acknowledgment of authorship. In respectful, civilized societies, problems can be solved with such methods without the need for extensive legal measures.

Conclusion

In this fast-paced era, reflecting on the production process and gathering thoughts is invaluable. The integration of artificial intelligence in architecture is reshaping the industry, offering new tools and methods while posing challenges in ethics, authorship, and professional practice. Embracing these changes thoughtfully can lead to more innovative, collaborative, and ethically sound architectural endeavors.