Associate Kirk shares his thoughts on AI in architecture before embarking on an online course.
I’ll admit it, I am a natural procrastinator and over thinker. I want to know everything about something before I act on it. That said, I’ve been playing with AI, trying different tasks with it, and I’ve been impressed with what it can do, particularly the new ChatGPT Astra.
I’ve had it generate research, summarise briefs, provide a first draft of project programmes, generate and edit imagery, summarise notes and, more recently, generate 3D models from images and plans.
I’ve also immersed myself in AI podcasts such as AI Daily Brief, Better Offline, 80,000 Hours, and Design + AI, but to be honest, it’s overwhelming and sometimes quite scary. I recently listened to a podcast with people who have worked within one of the big LLM providers describing how they don’t entirely know how it works, and I found it quite shocking. The opinions on it’s prospects vary widely: will it kill everyone? is it all just a glorified next word predicting machine? Is it a benign but useful tool? Is Artificial General Intelligence really a thing? Are all the above true? Unfortunately, it seems difficult to find the nuanced information through the noise. The best source of information I’ve found is a YouTube content creator called ‘House of El:AI’, a computer scientist with a PHD who seems to genuinely understand the context and provides genuinely good-quality, nuanced information.
I hear some people say that you don’t really need to understand the theory of AI to know what to do with it, but I do think, for me to use it properly, I need to know what it's doing and why. This is particularly true when it comes to governance, i.e., what checks and balances are needed versus the need for experimental autonomy.
Then there are the ethical and sustainability issues. Even considering the use of AI in our business is a big decision. Most LLMs are trained on data available on the internet, some of which comes from creative people who worry it will steal their IP without compensation. Also, if it is trained on historical data, it will be biased toward past discriminations. Benefits may also be concentrated with those in charge of the models, and then there are the vast amounts of energy and water required to run and cool the data centres when these resources are already in short supply and should be used carefully. It feels like when I ask it to do something, it shouldn't be done frivolously.
Looking at social media, particularly LinkedIn, it becomes more apparent that procrastination, or not engaging with AI, isn’t something we can afford. Like many people, I constantly feel like I’m behind with AI. So many people are talking about what they can do with it. I have seen people show how they generate an architecture project in a day, but I’ve actually had it struggle with generating a model of an existing house from photos and plans - the house in question did have a particularly unusual roof shape, however...
It got me thinking about how essentially AI is a next-word prediction machine. The roof shape, although somewhat traditional, was unnecessarily weird and illogical in places, but maybe this was what stumped it. Being a prediction machine, it was expecting the roof to behave in a certain way, so it modelled it as such. To me, it was clear what the roof shape was doing - ‘My job isn’t lost yet’ I thought! Maybe this is just a matter of time, or maybe there is something fundamental about the predictive nature of AI that means it can’t work this roof out. Maybe the human element will be the ability to be unpredictable.
With all the above in mind, I’ve enrolled myself on the Oxford University Said Business School’s ‘Organising for AI programme’ in the hope that it will give me a framework for implementing AI in our business in a structured and considered way.
I want to understand exactly how AI can help our specific practice, specialising in collaboratively designed, bespoke, low-impact projects. What parts of the job will be left for us humans? What makes us different? Are we just a pattern prediction machine with some add ons? I tend towards the belief that humans are complex but not special, so theoretically there is nothing that AI couldn’t eventually do. The question is what people will want humans to do, not because of ability but simply because they are human.
In some ways, we feel that our job is becoming more about the process from the perspective of the client anyway. Acting as a spatial counsellor or therapist and advocate is a big part of the job, here, the human touch is essential. So the big question is, how does AI supplement this? The programme started in early October and continues for 13 weeks, and I hope to feed back on what I’m learning and how AI will feed into how we work.