No items found.
This article was first published in the Architects Journal, AI Edition
When I was studying, ‘parametricism’ was the big buzzword. Despite the hype at the time, working with computational fluid geometry was mainly the preserve of the starchitect elite and a small band of experimental niche practices who also taught diploma units in London.
In the 00s, some people began to identify as exclusively ‘parametric’ architects and the others, like myself, who tended to draw in with straighter lines, carried on as they were. The technological shift, allowed by modelling software like Grasshopper, had become tethered to a style dogma and was largely ignored by mainstream practices. Labels that end with ‘ism’ usually divide opinion.
However, the application of new Generative Design and AI tools is showing all the signs of becoming far more transformative for the way architects will work. It’s not hyperbolic to state that the physical environment of our future cities will be moulded largely by the work of computer scientists and the algorithms they’ll write in the next few years.
With this as a backdrop, our best architects and placemakers must be more than part of the new conversation, but also active in the development of this consequential software-driven shift.
Some fears need allaying. Because, despite what you see on social media, ry wide application from administrative tasks to provocative conceptual design. It is available for everyone, often at low cost and it spans the spectra of 3D, text and image realms.
I’ve both seen how it is being used and have been given an insight how it might be used better. The following observations are taken the many meetings, training workshops and deep discussions I’ve had with practices around Europe since founding Arka Works six months ago.
Learning: AI is not the next style war, it’s for everyone.
The first step on the road to adoption is deciding to take action. Our brains try to protect us from the risk of the unknown. We often fear change. People largely ignore new phenomena for several months until, on the seventh or eighth nudge, they decide to act. I’m seeing many practices make this step now.
The AJ survey results (see pages 10–14) show that more than half of respondents are already on some kind of adoption journey and the ‘early adopters & Innovators’ cohort (those using AI regularly make up 15% of the overall total. This figure maps closely to the classical distribution predicted by the Diffusion of Innovation Theory model by EM Rogers in the 1960s — a bell curve diagram that is applied to any new transformative technology.
Before practices begin to adopt new tools, they should have an open discussion about how this technology aligns with their core mission and values and, therefore, what their adoption journey will look like. This will be different for all practices.
Setting a trajectory will mean trying some new things out for yourself, before forming a rigid opinion. It’s very important that board-level design leaders are part of the testing. It is also best done in parallel to support real-world project work, so that actual and practical value can be assessed. Leadership needs to come from the top. Your direction must be design based, because these techniques may begin to fundamentally impact the way you create. Ultimately that is the single most important thing in any architecture practice.
Learning: Don’t run hypothetical ‘research’, start small and apply to real-world problems.
AI is often compared to the invention of CAD or of BIM. This is largely wrong. AI will most likely bring horizontal change to your organisation, namely it will, in some way, touch every department of the business. In this way it is fundamentally different from CAD or BIM which are more vertically managed phenomena. Think of AI as a wide, horizontal umbrella covering all parts of the business; projects, support, bids, comms etc. Yes, you need technical expertise at the core of your approach, but mainly you need your designers and project leaders leading the conversation and doing actual testing on live projects to decide what is useful to their method (or not).
A common approach is for practices to view AI as a new, purely software-based problem and put it into their IT departments who then do research and ponder which apps, if any, they should direct a budget towards. This is a misstep.
Yes, those tech specialists need to be closely involved, but this isn’t purely a ‘software stack’ question, it’s more a culture of practice and working behaviour question. Once you’ve decided the desired scope and depth of adoption, forming a working group is usually the first step.
Learning: Set up a working group led by your best designers
The AJ numbers suggest that only 8 per cent of respondents are reporting any kind of controls being put in place around AI tool usage. This uncontrolled use accords with what I see when I meet practices.
People are experimenting, which is great. But there are risks. We are in the Wild West where opportunity and pitfalls abound. If you are using a basic subscription on Chat GPT or Midjourney then the images, PDFs and words you upload are likely being harvested and will be part of future model training material.
It is possible to host and run your own private models using open-source models, or to pay extra for privacy. But if you haven’t taken a position, you may be in breach of an NDA or data policy and you may be allowing commercial/GDPR sensitive data to leave the business. Conversely, simply banning things probably isn’t going to work longer term either. That may put you on the back foot while your competitors progress at pace.
If you decide to dive in, then an adaptive approach that is permissive but cautious in approach will reduce confusion about which tools are — and are not — endorsed officially by the practice. You must also allow yourself to change your mind when new information suggest a shift in strategy is needed. Many other guardrail issues should be considered; including bias, detecting errors, Intellectual property and copyright, environmental impact and costs.
Learning: Scope things out and build a policy framework for responsible use.
Clients don’t want vanilla and generic solutions and this is the likely outcome of using a closed source, ‘black box’ generalised AI tools. The solution? Don’t use general tools, pick them up and customise them, pull your own IP into them, get answers no one else can get (for example Heatherwick have trained their own in-house image models).
All technological breakthroughs lead to unintended consequences, so we need to keep a proactive eye on change and engage it head on. Let’s not just let it happen to us. While practice leaders are wondering whether they should look into this AI thing, the next cohort of architects are graduating, and they are already using it everyday.
Some respondents to the AJ survey, expressed concern that AI will lead to a lazy design culture and shorter collective attention span. Others have pondered what the point is in practice if we begin to dilute our core value proposition. I would agree with the sentiment on both points. However, as long as we are working with humans on behalf of other humans, then our human empathy, unique experience-based insights and ability to communicate ideas with our clients is the most important service we offer, so we should lean into it.
Conclusion: We shape our tools and then they shape us.
LoRA model credit: Ismail Seleit
In the second post of this series we are looking at what I’m calling “Auto Optioneering”.
