Artificial intelligence is posed to change the way we design, develop, document, and possibly construct our projects. A concept that has been around for a few decades, AI recently took a spectacular leap, mainly thanks to astonishing outcomes from a few very powerful text-to-image models, and of course the chatbots.

The current models are largely based on the research done in the 70’s on Parallel Distributed Processing models, founded on the intuition that running multiple simple computations at the same time, and creating feedback loops (the base of “learning”) would get a step closer to what a cognitive system does.

The immense amounts of data, together with computational power, that are available today, make it possible for these algorithms to perform in the way we are witnessing for the first time. The topic is not only very broad, but also constantly evolving. With this article we try to peek a little beyond the buzzword and take a glance at what AI for architecture looks like at this moment.

Understanding AI in architecture

Some kind of machine intelligence has been with us for a long time, and in architecture several software make use of automation and “smart” features in different ways, assisting architects and designers to fulfill specific tasks with great speed and precision.

Think about how a BIM tool can generate a door schedule without the need to manually draft it yourself. Parametric design has also been a quasi-generative tool allowing designers to materialize forms with greater freedom. Even a calculator can be regarded as a simple intelligent machine, one that is immensely better than humans at the specific task of doing calculations. 

What has now overtaken the attention of the public are systems that, though very far from being close to human intelligence, can perform tasks usually regarded as the domain of humans. Tasks that involve creativity, and originality. These are diffusion models, or generative AI models, able to translate text-to-image, image-to-image, and soon text-to-video in a very convincing way. 

As images are very powerful in architecture, since they are the almost universal medium with which we convey our ideas, text-to-image models stormed the industry with their nearly magical capacity to generate images (and videos) of fantastic photorealism and beauty.

In fact, these models are trained precisely to appeal to human taste. But this is a very tiny part of what architecture is, and AI does not stop at arranging pixels together to generate pleasant visuals.

A perhaps more interesting way in which the power of AI could be harvested in the architectural industry lies in how we develop our work, for example by: 

• Giving early insights on preliminary design iterations.
• Finding alternative solutions to complex issues when a large amount of information is influencing the problem. 
• Assisting in the checking of authority compliances. 
• Making it easier to coordinate the different trades.
• Optimizing structures and MEP services routing.
• Implementing sustainable solutions which are performance based.

Software companies, architects and engineers are rushing to find new solutions and we are expecting to see great development at a very fast pace, and almost every computer program out there is incorporating AI tools (if they want to survive in the market).

Advantages of AI for architecture

Like any technological advancement in the way we process our work, AI will undoubtedly offer great benefits to the architecture industry. There have been several “revolutions” before, such as the transition from pencil drawings to CAD, and from CAD to BIM. The progressive development of AI-aided tools will probably make an even bigger impact. Some obvious advantages are:

• Testing ideas more quickly, being able to generate breakthroughs and unexpected solutions.
• Giving more freedom to designers as they can experiment with different workflows (for example, using sketches to generate renderings, or using parametric massing iterations to test materials and concepts at the same time).
• Organizing data more efficiently, and making use of those data in an effective way.
• Generally making certain lengthy processes faster, freeing up time for more creative tasks.

Applications of AI in architecture

New applications come up almost weekly, and it is a very exciting time to be in this space, with new possibilities and opportunities paving the way for greater creativity. Some of these applications include:

• Image generations, from text, from other images, and with several ways of controlling the outcome (for example using ControlNet for Stable Diffusion, of which we will talk another time).
• Generating plan layouts, given a certain set of inputs, parameters and constraints.
• Documentation, which is the act of translating the design into a set of formal drawings, tables, specifications, lists, and so on, with the objective of forming the basis for pricing, tendering, and constructing. 
• Not only images, but also words are very important in the architectural industry, so the current chatbots are already aiding at a high level. For example, we can use chatbots in synergy with parametric modeling in BIM.

There are many designers and architects, including us, who are experimenting with deep learning models based on diffusion (Midjourney, Stable Diffusion, Dall-E, etc..) and trying to incorporate those into their workflow.

At the same time, we are witnessing a booming of proprietary applications based on those models that help rendering images, for example. And many of the “traditional software” as we already discussed are embedding AI into them. This will make AI tools much more accessible to a larger audience.

The applications are going to be limitless. It is up to the next generation of architects to decide how to make the best use of it. The development is happening so fast that a deep debate around it is challenging, but we think there should be a wider conversation on the advantages, benefits, costs and disadvantages of embracing the use of AI in architecture (and in even much more critical domains as well, like social media or weapons).

Disadvantages of AI for architecture

First, a perception risk, that is to banalize the work of an architect and represent it as if reduced to the production of pretty images. There is a larger issue here, related to architects communicating the nature of their work effectively. However, the dangerous and improper logical construct would be that since producing an architectural image is so easy why should we value the work of an architect. Of course there is a lot more behind that image.

Another disadvantage is the idolization of AI models and their real nature; in short, what diffusion models really do is anything but creative or ingenuous: they make use of a huge set of data (gigabytes of images) as training sets, combined with algorithms that try to predict how a certain pixel arrangement would please a human eye (again based on a set of data). 

There is not much magic to it, and it is essentially strength in numbers. In fact, a large amount of what comes out from those AI generative models is utter rubbish, but people generally showcase the best results, not the dozens of useless and discarded images.

To us the biggest disadvantage is the risk of stripping away one of the most essential elements of our existence, that is, creativity, or the ability to create something from nothing. To delegate the agency of envisioning and imagining to an AI model would be, we think, greatly damaging. Imagination is something too important to give up. 

And because it will become so easy, and so fast, and possibly so cheap, to start to heavily relying on AI models when it comes to creativity, we may start to lose the capacity for complex thinking, effectively becoming less intelligent. As such, AI must be an ally, not a surrogate. A tool to complement our thinking, not a shortcut to come up with a fast and furious proliferation of images (whose actual utility is very questionable). 

Human Intelligence, Artificial Intelligence, and Value

Today, there is no human being on earth able to beat an average computer in the game of chess. However, humans did not stop playing chess, and national champions still make the news. Why is that so? The answer is very simple, because people like playing chess, regardless of whether there is an artificial system that would inevitably beat them.

Although we are not there yet, and will not be for a long time, It is conceivable that in the future, an AI model could produce an architectural design, from scratch to the end. However, it is also very conceivable that there will be architects still enjoying the process of designing, and clients still very willing to pay for their services, perhaps a higher price than the design produced with pure automation. 

Value is classically understood as a direct consequence of something’s rarity. One would pay a lot for a hand-made carpet produced by artisans over several months of meticulous work, but way less for a similar product manufactured in minutes by a machine.

In our society as presently constructed, higher value leans towards scarcity, not abundance. So, being able to produce thousands of architectural concepts within minutes, or seconds, may not actually translate into a valuable proposition, but quite the opposite.

We would also argue that there is a slight mystification when it comes to AI and how it compares to the general intelligence of a human being. Human intelligence is something different altogether. But for the very choice of the name, Artificial Intelligence, it is common to compare the operations and functionality of a generative algorithm to that of the human brain, of which we still know very little. The choice of words is critical to direct our thinking.

Beyond the buzz and the excitement, AI for architecture is still at the beginning and we believe there will be wonderful ways to enhance the quality of our services and products, liberating imagination to new horizons, while we have to be careful to pursue the best possible outcomes framed within purposeful intents.