Artificial intelligence in hospital design
Christopher Shaw
Founder of Medical Architecture (now retired) and past Chair of Architects for Health
The eagle-eyed will have noticed that my last few posts included Open-AI generated images. I've been using the open-source DALL-E tool which takes text input and translates that string into a (mostly plausible) image from a vast array of data and some very interesting artificial intelligence. Log in here and have a go yourself. https://labs.openai.com/
Back in the early 1980’s I was using an early automated architectural plan programme called GOAL running (evenings and weekends – honestly) on the polytechnic’s state of the art Norsk Data mini computers. Feed it a schedule of accommodation along with a proximity matrix and a few hours later it produced an atrocious plan. In retrospect these were quite reminiscent of chaotic outpatients departments produced in the PFI era.
Things have obviously advanced and some of the AI generative tools now have the potential for optimising building envelope, daylighting and room layouts. This is not yet at a stage where they can orchestrate the vast numbers of parameters, experience, insight and understanding brought by engineers and architects, however the technologies are becoming interesting again and I will be having a closer look alongside recent process simulation tools.
So for amusement let’s look at kind of the images, in each case the caption is the text used to generate the image:
Perhaps a bit less successful - but there is the bones of something here.
Not surprisingly things fall apart when asked to generate a 3D representation of a patient bedroom in axonometric.
MIoL.
2 年At the moment it's still a bit (as I've seen ChatGPT referred to recently) "Dunning-Kruger As-a-Service". As Alex Senciuc RIBA says, it'll need a two-way conversation, at least once, to teach the thing the rules of what definitely is/isn't viable but might help with modelling options more quickly, with the human adding both experience and creativity (and thus the real value) to the process.
Associate Director at Llewelyn Davies
2 年We ran a similar test in the office the other day and it sends a chill down my spine! AI should stand for 'Architect's Intelligence' otherwise the result will always be...'artificial'!
PhD in Strategic Healthcare Planning, Senior Architect at Medical Architecture
2 年Using AI to shape hospital designs is more of a two-way conversation between designer and machine. Obviously we cannot expect AI to churn out perfect plans (or even ones that makes sense). However, we can expect AI to do the grunt work if we give a few simple rules (grid size, room types & quantities, adjacencies). But that should not be taken as the finished product, it should only be a starting point. As designers we could then take the generated plans and turn them into a design. For example, take a tool like this one (although this example is not exactly AI, it could easily be an AI application): https://www.sciencedirect.com/science/article/pii/S0926580521005537#f0115 which can be used to understand the rough shape and volume and could quickly rule out a (say) courtyard type hospital in favour of a podium and tower type hospital (for a specific site and program). Wouldn't this be useful to know before going down rabbit holes in the design process? Or wouldn't this be useful to test the designer's intuition?
Director at Fleet Architects Ltd
2 年I was wondering where/what
Associate Professor of Practice at Xi'an Jiaotong-Liverpool University ? Programme Director MSc International Professional Practice (MIPP)
2 年Will share the outputs from our masters programme, one module of which looks at the use of AI in form generation