Archdaily Published the First-Ever Architecture Interview With a Bot. This Is What I Learned

Nicolás Valencia
8 min readJan 4, 2023
© Nicolas Valencia, using image ©Flickr user masakiishitani, licensed under CC BY-NC-SA 2.0</a>

On December 21, 2022, we published on ArchDaily our first-ever interview with a bot — likely to be the first-ever on any architecture media.

Ammaar Reshi inspired the idea on Twitter, who published a children’s book co-written and illustrated by AI after experimenting with ChatGPT and MidJourney over the course of a single weekend.

Since most groundbreaking technologies pass by our lives without even noticing, Reshi’s simple but powerful example illustrates how 2022 might be remembered as the year when AI-fueled tools finally become accessible to a larger audience: from text-to-image generation Stable Diffusion, Midjourney, and DALL-E 2 to the surprisingly smart OpenGPT, a language-processing designed artificial intelligence to generate human-like text.

It is common knowledge that architects often resist incorporating state-of-the-art technologies: it has been happening with this new wave of tools, but it also occurred with the widespread use of hyperrealistic renders and 3D models. Before that, it happened with Photoshop. And before that, AutoCAD and Revit. And before that, computers at large. And before that, photography. And the list goes on and on.

As a symbol of what machine learning could achieve in the short term for architecture and humankind, I ‘talked’ with OpenGPT about the architecture trends of 2023 and I would like to share a few reflections.

Why ArchDaily profusely covers these start-of-art technologies

A render created by text-to-image diffusion. Via ThisHouseDoesNotExist.com

In 2022, ArchDaily has extensively covered the impact of machine learning systems on creative tasks, such as featuring tools like This House Does Not Exist and Interior AI. These articles have sparked a conversation about the reach and potential threat of such systems on creative tasks, as well as highlighting their ability to, for example, map and intervene in urban informality.

We have intensively covered these topics because, as Carlos Garcia Vazquez noted about the modern Western city taking shape at the beginning of the 20th century, “architects didn’t know how to lead or even work with these new elements of modernity.”

On one hand, we want to encourage architects and designers to embrace these technologies for our own good in the early stages of development, instead of later suffering the effects of disruption employment models implemented by someone else.

That’s why ArchDaily’s New Practices have selected seven startups since its inaugural edition in 2020 — U-Build, Urban Beta, Rayon, Monograph, Baupal, CANOA, and ICON — sharing the honor alongside traditional practices. All of the selected practitioners have demonstrated the flexibility and diversity of architecture as a discipline on a global scale.

On the other hand, automation and machine learning are technologies still in the exploratory stage, so editors can open the discussion with a global massive audience — over 15 million monthly sessions worldwide — that historically has been skeptical of technologies, or at least, has underrepresented its potential while overrepresented the human’s creativity skills.

Where ChatGPT gets its knowledge

Photo by Ilya Pavlov on Unsplash

Conducted over two sessions held on December 18 and 19, 2022, the original conversation with ChatGPT extends to approximately 6,000 words. The bot might recall and reference statements made during the current session but is unable to quote from previous conversations nor have access to the Internet, as it explained by itself:

ChatGPT: As a language model, I was trained on a large dataset of text from the internet and other sources. This training allows me to generate responses to a wide variety of questions and topics based on the patterns and relationships present in the data […] My knowledge and responses are based on the state of the world at the time of my training, and I do not have the ability to update or expand upon that knowledge.

As of the time of writing, ChatGPT’s latest stage of training had been completed in 2021, thus it doesn’t know, for example, anything about the results of the Qatar World Cup.

Then, how could ChatGPT predict the architecture trends of 2023?

Photo by Danist Soh on Unsplash

First of all, any prediction, conducted either by humans or robots, is an extrapolation of current, predictable events — events or tendencies that can intensify, soften, or even disappear.

Predictions are, despite their purpose, quite predictable.

Hardly any forecast for 2020 New Year’s Eve presaged a pandemic or a full-scale Russian invasion of Europe territory in the 2022 edition. Forecastings express a stakeholder’s desires (and fears) more than anything else — a Metaverse pundit would seldom describe a future without it included, for example.

Ultimately, the future is always one step ahead of us.

What amazes me the most is that ChatGPT aims to “provide conversational answers to users’ queries” based on an extensive up-to-2021 database, but the quality and precision of its responses lead to thinking that ChatGPT could work as a crystal ball as well.

The logic could be: “If it’s smart enough to know almost everything about both the past and the present, it could be smart enough to create a forecast of the future.”

Or at least, that’s the logic we apply among humans.

Having said that, architecture is a slow-cooked discipline: a built project hardly takes less than one year from the initial design phase to the final construction and opening —that might explain the raise of pavilions and temporary experimental projects since can more easily be funded, materialized, and showcased as part of any emerging architect’s portfolio.

Therefore, architectural trends may not change significantly from year to year, but there are rising factors or variables that might be considered in the future when designing, analyzing, or evaluating the built environment in the next five years. In fact, critical writing, ephemeral architecture, media, photography, performance, and architectural theory can test, adopt, or question these movements.

ChatGPT seems to agree with me:

Nicolas Valencia: […] architecture is quite slow when it comes to embracing new trends.

ChatGPT: It is true that the field of architecture tends to evolve relatively slowly, and new trends and technologies may take time to be adopted and integrated into practice. However, it is important to keep in mind that the pace of change can vary widely depending on the specific area of architecture being considered and the specific context in which it is being applied.

An unexpected challenge: editing a bot’s responses

Photo by Lukas on Unsplash

I had a great time chatting with ChatGPT, but when I went to edit the conversation for the article, I realized it was challenging to edit the bot’s responses without introducing any (human) bias. ChatGPT might have solid answers, but it’s up to the human editor how they would be finally represented in any article.

In my experience, ChatGPT's answers were Solomonic every time I asked for deepening its trends forecast. Here’s an example that didn’t make it into the final version of the piece published on ArchDaily.

Nicolas Valencia: LEED certification has been discredited since focuses “overwhelmingly on operational emissions rather than emissions from the construction supply chain”, as expressed by Architects Declare member Andrew Waugh.

ChatGPT: It’s true that the LEED certification program, which is administered by the US Green Building Council (USGBC), has come under criticism in the past for focusing more on operational emissions rather than emissions from the construction supply chain. However, it’s worth noting that LEED has evolved over time and now includes credits for a wide range of sustainability factors, including materials and resources, indoor environmental quality, and innovation in design.

It’s important to recognize that no single rating system or certification program is perfect, and it’s always a good idea to do your own research and due diligence when evaluating the environmental performance of a building or product. However, LEED and other similar programs can still be useful tools for evaluating and improving the sustainability of buildings and construction projects.

ChatGPT’s answer can be narrowed down as follows:

  • First, acknowledged the shortcomings of LEED certification.
  • Secondly, explained how the certification has evolved by embodying the criticism.
  • Later, recognized that “no single rating system or certification program is perfect”.
  • Finally, admitted those programs “can still be useful tools”.

Afterward, when asked its opinion on automation in architecture, ChatGPT replied:

It’s important to approach automation and other forms of technology in the architecture industry with a balanced perspective, recognizing both the potential benefits and potential drawbacks.

This answer structure repeats throughout the conversation on diverse topics, but the latest quote encapsulates what ChatGPT tries to answer when I asked for a personal opinion: “a balanced perspective, recognizing both the potential benefits and potential drawbacks.”

Of course, what balanced means isn’t unbiased. Thus, as a human editor, I had to preserve the bot’s intended meaning when editing its answers. In addition, I may need to include an Editor’s Note to ensure that readers do not misunderstand formatting such as hyperlinks or italicized or bold text as being decisions made by the bot:

“The following conversation has been edited for length and clarity. Hyperlinks and stylizing have been added by humans.”

The note taps so many portraits of the future.

Finally, ChatGPT does not have an opinion on architectural projects but rather interprets what humans have said about them.

Photo by Andy Kelly on Unsplash

To be honest, the 2023 architecture trends brought up by ChatGPT remind me of actual ArchDaily forecasts published in 2021, 2020, and 2018. This is not surprising, since ChatGPT has read millions of websites under the supervision of humans — using a technique called Reinforcement Learning from Human Feedback (RLHF) — in order to create a human-like written answer, not to develop critical thinking.

Therefore, being an architecture critic is not a strength of ChatGPT.

As a specific example for this text, I asked it to “show [me] an example of a good architectural project”. Its answer was Sydney Opera House’s Jørn Utzon, “a multi-venue performing arts center that is considered a masterpiece of modern architecture”… by humans, not based on their own judgment.

Similarly, despite the subjective nature of their overall quality, when asked for examples of biophilic design, ChatGPT mentioned The Singapore Botanic Gardens and The New York Times Building because both projects “feature a range of biophilic design elements”.

ChatGPT has read a lot about architecture and provides examples when being asked for a specific topic, but this confirms that the bot is able to provide information and examples based on the descriptions it has read, rather than providing its own aesthetic analysis.

Due to the high speed of artificial intelligence development, this interview might get outdated sooner than later. In fact, OpenAI CEO Sam Altman humbly described ChatGPT — its own product — as “incredibly limited” but “it’s a preview of progress”.

The architecture trends described by the chatbot should be taken with a grain of salt, but for the purpose of the interview, it’s a sneak peek of what Artificial Intelligence could achieve in the short term for architecture and humankind.

--

--

Nicolás Valencia
Nicolás Valencia

Written by Nicolás Valencia

Architect. IAAC Faculty and UTalca Guest Faculty. TERRAZA Podcast. Former Head of Editorial at ArchDaily | Articles in English y Español

No responses yet