AI generators and the future of architecture
As if to prove a point that politicians can sometimes be guilty of robotic delivery when making speeches, the start of the year saw UK Chancellor Jeremy Hunt enlist an Artificial Intelligence (AI) tool to write the opening lines of his address on the future of the UK economy. OpenAI’s Chatbot GPT, a piece of software trained on a massive amount of data to allow it to generate human-like text in response to a given prompt, was used by the Chancellor to show how important high-tech business is to the future of the UK economy.
Interestingly, I used the same tool recently to write a blog post about how AI will impact the work of an architect (https://www.stephengeorge.co.uk/blog/artificial-intelligence-ai-in-architecture/), but this is not the only type of AI tool to have suddenly gained prominence in recent months. The emergence of AI art tools that generate digital images from natural language descriptions, such as DALL-E or Midjourney, has seen an explosion of weird, wonderful and sometimes eerily realistic images appear across the web and numerous social media channels.
The advent of text to image AI software is, as you might imagine, of great interest to us as architects and designers. VR and game developers are already using text and image generators to create fully immersive spaces and even entire new worlds, but with a whole new batch of text-to-3D generators recently announced to the marketplace, such as GET3D from Nvidia, Make-a-Video from Meta and DreamFusion from Google, one can’t help but wonder about the future use of AI in the design of buildings and what it will mean for our clients in the broader real estate sector.
We might get a better sense of its value if we step back and consider how these AI image-generating programmes work. In essence, they are fed huge sets of data and use intelligent algorithms and fast processing to learn from the features and patterns in that data to deliver the highly realistic visuals that are now so prevalent online. Whilst architects already use lots of building and design data when starting a new project, AI’s ability to rapidly process huge amounts of it will not only take the labour out of many mundane processing tasks, but will also mean we could be producing and testing concept ideas much more quickly to deliver a more energy and resource efficient end product, for example. Indeed, some BIM software is already using AI to learn from data and to make independent decisions on how to automate and improve the model-building process and it’s likely we’ll see much more AI-assisted BIM in the future.
Given the photo-realistic visuals being generated by current text-to-image software, it’s also easy to see a potential future for AI-generated renders. An incredibly important tool for architects, renders allow us to better interrogate our design ideas, whilst also helping clients visualise and understand our design intent.
However, a number of challenges lie ahead before we see AI generators more confidently adopted by architecture professionals, not least are the growing concerns around intellectual property and data ownership. As AI image generators are trained on vast amounts of data and billions of image libraries, some of that data and imagery could be copyrighted work or considered to be another organisation’s knowledge asset. And if we have tools such as GoArt that can produce artistic works in the style of the Grand Masters, how long will it be before we see tools that can produce entire building concepts in the style of famous architects or draw on data, even if unknowingly by the user, that lifts elements of design from other well-known buildings? This is potentially a legal minefield for users and has already led to the opening of lawsuits against various software companies by artists who feel their rights have been violated. We will examine the impact of some of these lawsuits in a future blog and how they may impact the use of AI in architecture going forward.
In truth, given the speed at which AI technology – and text and image generators, in particular – have appeared in the marketplace, it seems inevitable that the legal system will need some time to catch up. Regardless of the legal, ethical and creative questions that AI poses, its rapid evolution continues to fascinate and represents an incredible opportunity for the architecture profession – one that SGP is fully exploring.
We do not envisage AI as a threat to architects; it will still require a user with experience, knowledge and skill to produce functional and practical results. Nonetheless, we need to recognise it as something that can empower us to create better buildings, for the good of our clients and wider society. It is a tool that will streamline many time-consuming processes and make the whole design process simpler and quicker and, as the Chancellor’s AI generated speech proclaimed, it will be highly important to the future of the UK economy.