Creator vs Creation: Can AI platforms like MidJourney replace designers?

In the previous article, we covered the shortcomings as well as the appeal of MidJourney as a software and how intuitive it actually is to text prompts. In its nascent stage, the software’s archives seem to lack the database for culturally nuanced references and inspite of having an extremely simplistic UX, like most softwares – it requires time and effort to master MidJourney. In short, there is a long way to go before the platform becomes a part of streamlined design processes. But the question still remains, ‘Can AI ever replace designers?’ – just how capable is it?

Inventive, not Intuitive

“I think AI tools, parametric, generative tools, simulations, 3D printers – these are basically just that – tools. The “good prompts” are just that – prompts. The human mind brings nuance, the drama, the poignancy into design for everyday users to experience.  And for architects, we also work in parallel to add to the continuity of our discipline’s evolution and its role within society.  There might be a day when AI can even simulate this, but as Tom Cruise said in Top Gun Maverick: ‘Maybe so Sir, but not today’”, says Edward McIntosh, Design Director, Middle East, Atkins.

Hassan Ragab, seems to in-sync with this tune as well, “One assumption people make is that AI is infinitely creative and does things on its own. But AI has no intuition, it is a tool like any other digital tool. Of course it’s really easy to get fancy looking images using it, but it takes time, patience and knowledge to master and control. The AI text to image generators is just a python code. It has boundaries. So at least for now we are safe from a ‘West World’ scenario”, 

MidJourney as Super Photoshop

Currently, Edward’s team uses MidJourney as a concept generation tool, “We were very excited at Atkins when we started experimenting with text-to-image AI solutions like Midjourney. We did many tests and used the tool in many scenarios to try to find out for ourselves where AI can be utilized best.” 

Here are a few findings from Edward and his teams:

The text-to-image AI tools are not currently geared specifically towards architecture, they are image focused. “It’s more like a Pinterest inspiration board,  for references that you could not find easily, like “a building made out of cotton candy”, explains Edward.

Regardless of the prompts used, the result led to similar-looking images. “We noticed different colleagues using different prompts would arrive at similar-looking images. This might be because people have many ways to describe ideas through human language, but AI can only come up with a limited set of solutions that respond to many different ways of describing a scenario.  Because of this, we started fearing homogenization of any “design” coming out of the tool. This also happened in a way when parametrics, simulations and complex geometries started making their way into architectural practices some 20 years ago. Designs looked “Rhino” or “rhinoscripted” or “grass-hoppered” or “Maya”, or “Houdini” based on what software they were created in.  As our architectural solutions tend to respond more to human use, local construction techniques, tectonic expressiveness and materiality, we prefer to not create designs that are not highly-influenced by the tools we use. Thus, similarly with AI, we use it purely for internal inspiration in terms of design explorations and conceptualisation,” Edward makes known. 

Revealing more, Edward goes on to say, “The area where we have found great use for these text to image tools is in narrative, we use the tools as a “super photoshop”.  For example, if we want to create a rendering of a building in the Costa Rican jungle, we create the background of the jungle using MidJourney  and then we use this background to render our project. In the past, if you used actual geometry to render a jungle the number of polygon faces representing the complex leaves, branches, etc would demand high-rendering times – not to mention the actual modeling of the geometry. Apart from this, we also use MidJourney to “tell the story” of our projects through the creation of human characters that inhabit our projects and describe the concepts we use to guide our design decisions.”

He reveals what in his opinion is holding MidJourney back, he says, “Nobody can claim they control the tool 100% yet; that’s why many users describe the interaction as ‘collaboration’ or ‘co-creation’ with the tool. I wanted to have a brief, and a narrative that would actually limit me, and give me a specific aim. That way I wouldn’t be able to just create a random cool image but rather, I would be forced to create images that responded to my narrative. Narrative support is where I see the current value for the tool. This is really where the tool saves me time,” Edward finishes.

MidJourney as an aid for Design Inspiration

Although, AI platforms such as MidJourney do a tremendous job in creating effortless conceptual renders, there still exists a gap between the realities of design problems and fantasy of the ethereal pictorial outputs of MidJourney. Today, there exists no better tool to come up concepts, schematic themes and impressive contexts, however, the future might bring more. Stay-tuned, as we explore the possibilities of AI platforms in the realm of architecture, and explain why these platforms are colloquially called ‘Architect’s Drug’ in the next article.

All images, including featured image, courtesy @Edward McIntosh

Read more about MidJourney
Previous in series: Creator vs Creation: What the users say about AI-platforms like MidJourney in architecture and design?
Next in series: Creator vs Creation: What will the future of AI in Design look like?