Re:Create Q&A with Benjamin Lorentzon

How do you use AI in your work? 

Most of the time I use AI to gather knowledge and research topics, replacing a search engine almost completely.  
But also, I’ve been using AI to help with coding tasks, everything from creating simple Python scripts for common office tasks, converting data formats and creating automations, to more advanced tools or full stack apps that can hopefully help me and my colleagues in creating easier and more intuitive workflows. 
I’m also using AI techniques, such as training 3D gaussian splat scenes of existing places to use as digital excursions, documentation and annotation for preexisting places or buildings.  

This is helpful when we need a good starting point in creating a new architectural design in an existing context.  

Or using generative AI models to create images that serve as internal visual references, inspiring and supporting our design process during team discussions. For instance, using simple 3D shapes or hand drawings as a base, I use AI models to generate façade proposals, like an in-context Pinterest. 

Why? 

As an architect, BIM and computational design specialist at Wingårdhs I use lots of different software, 3D modeling and BIM tools, AI can be my personal interactive manual in this case. AI can also help me write small scripts or plug-ins that expand the functionality of the software we use at the office, to help streamline existing or new workflows, or to help the software “talk” to each other. This is enabling me to do things I otherwise wouldn’t have the time to do, like learning coding.  

I already have tools that enable me to automate 3D modelling tasks (Rhino3D, Grasshopper) but AI helps me get even further in creating the results I’m striving for. In this sense AI enables me to be more expressive, or free to create whatever tool I need to complete my task.

What works? 

Most of the paths I’ve taken down the AI rabbit hole have been fruitful for me personally, but most of them are not what the hype suggests. I can see the potential in future workflows for architects. But for now, what really works for me is the ease of learning new things, text editing and AI coding, which is huge!  
 
At this point, generative AI images serve primarily as a source of inspiration, but they do not yet provide the level of control over the design process that we require as an architecture firm. There is however great value in complementing our designs with AI generated 3D props for renderings, enhancing or replacing parts of photo montages to give even more immersion to our architecture proposals.

What do you struggle with? 

Two things; 

1. Making the AI model understand my intentions. It is easy to think AI will read between the lines as a human coworker with all the context and body language we share in the real world.

2. Making the AI model think outside the box, generating edge cases and coming up with something completely new. Generative AI models and LLM: s are trained on common available data, which makes them good at creating more of the same. But coming up with something completely novel is something today’s models are struggling with in my experience. The only exception I can think of is that the AI models can combine knowledge and generative media in beautiful ways, and in that way create something new.

What is your advice to industry colleagues? 

Be precise and guide the AI model. Give the AI model as much context and detail about what you are trying to achieve as possible. Even better, get help from AI to summarize what you are trying to achieve in a format that other AI can easily read.

Get the latest news from Re:Create delivered straight to your inbox