Will an AI Take Your Design Job?

Hi friends,

Last weekend I got really into architectural renderings.

Check out the results below:

Not bad, right?

Except I didn’t really do them.

An AI did them.

It did so based on single-line text inputs that I gave it.

I type in something like, “Futuristic skyscraper, solar city, at dawn, architectural rendering, inspired by Zaha Hadid, atmospheric lighting”.

The algorithm then combs a massive image library (seems like it might even be all the images on the internet) and combines source material organically based on your prompt to produce a result.

Given no more than one line of text, this AI-generated coherent-looking, awe-inspiring architectural renderings that in fact, do look kind of Zaha Hadid-y.

It delivered this in less than a minute.

We are living in the future.

About a week ago, a friend gave me an invite to MidJourney, an AI art-generating tool that is in closed beta. Since then, our friend group of artists and designers have been putting it through its paces, with results that range from impressively beautiful, to hilariously absurd, to hauntingly disturbing.

At first, our results seemed rather random. But as we learned to feed the algorithm the right inputs and utilize more advanced features like weighting the prompts differently and seeding it with a source image, our results became more controlled.

The machine could produce high-fidelity visual ideas at an insane pace.

That got us talking.

What do tools like this mean for designers?

Are our jobs and passions going to be automated away?

No, probably not.

Currently, a website called willrobotstakemyjob.com puts the risk of automation for Commercial and Industrial Designers at 5%, compared to 87% for Heavy and Tractor-Trailer Truck Drivers.

Phew.

While tools like Midjourney are impressive in responding to prompts and generating a wide variety of ideas quickly, we have to remember that design is more than just spitting out concepts.

Generating is only one part of the equation. The other part involves making decisions and value judgements about your ideas, then integrating and refining your solutions. Since design is fundamentally a human-centric field, all this has to be done in the context of an understanding of human behavior and emotions with a high degree of nuance.

Current AI is already pretty good at the first part (generating). It still needs human inputs to define the conditions in order to produce useful outputs. But the synthesizing and refining based on a human-centric mindset is where it really fails. It can only produce results based on the rigidly defined conditions that it was given, in the specific realm it has been trained on.

You still need a human to sift through the output and make decisions.

Humans are uniquely equipped to do this because our brains allow us to process information organically, draw connections between disparate ideas, apply our past experience and mental models, and perhaps most importantly, empathize with others and apply their experiences and mental models.

Basically, if you want to be a good designer you have to think like a human.

So why not just program an AI to make design decisions like a human?

Because it’s incredibly hard. So hard, in fact, that in AI research a problem that requires an AI to be as intelligent as a human to solve is called “AI-hard” (great name, folks) or “AI-complete”. There’s also significant doubt among AI researchers that it’s even a possible outcome or worthwhile pursuit.

I’m willing to bet design is an AI-complete problem.

To make a fully automated designer you have to first program an AI that is as smart and creative as a human being. In that case you might as well hire a qualified human.

It’s more likely that AI will become a tool for designers to use in parts of the process where it makes sense. Crunching large amounts of data and identifying patterns. Or generating a wide variety of possible solutions and selecting the best based on inputs and criteria from human designers. In fact, it already is being used this way with generative design tools.

It’s possible that AI tools will replace or enhance some parts of a designer’s craft. But a designer’s tools have always been evolving. CAD software has totally replaced paper drafting. Fast, easy, photo-realistic rendering has made high-fidelity visualization by hand all but obsolete. 3D printing has replaced foam shaving.

Emphasis will shift to more of the synthesizing, integrating, and refining parts of the design process that can’t be easily automated. These higher-level functions are where future designers will hone their craft.

It’s important for designers to know about AI tools and their growing capabilities, and consider how they will integrate into their future workflows, but there’s no need to defenestrate your computer just yet.

Then again, maybe this is all wishful thinking and super-intelligent AI is going to take over the world.

In which case, I for one welcome our robot overlords!

Anson

P.S. Shout out to my friend Brian for unleashing this tool on our friend group. It’s been fun to experiment and discuss the implications. Those discussions fed into the writing above.

P.P.S. check out some more interesting results from my MidJourney explorations below. I’m still new to it, and my results are far from the most impressive I’ve seen.

Interesting things of the week

📖 Reading - Will AI eventually replace designers or change the way we design? (Faisal Risq, on Medium)

An informative article that makes a clear case about why design will remain a fundamentally human-centered activity, utilizing the strengths of AI such as rapid discovery, speed of iteration, and identifying patterns to support and inform the more human side.

👁️ Seeing - #Midjourney (Instagram)

Check out other people’s results with Midjourney. Remember, all this came from one-line text prompts.

👁️Seeing - Chair designs generated with Midjourney (Nicholas Baker, on LinkedIn)

Some pretty cool chair designs made by a fellow industrial designer using the Midjourney algorithm. I’d be interested in what the text prompts were, as the results are pretty clean. I do think soft, organic shapes have an advantage. When I tried to get the algorithm to do hard goods with straight edges, it tended to fail fantastically (see below).

📺 Watching - AlphaGo Documentary (Directed by Greg Kohs, free on YouTube!)

A documentary about the effort to develop artificial intelligence that would eventually beat the then world champion of Go. Go is an ancient Chinese board game played with black and white stones on a gridded board. The rules are simple, but the possibilities are almost infinite, making it a perfect challenge for AI. The documentary puts the story of this impressive technological development through a very human lens by telling it through the voices of the technologists and the players.

I watched this documentary when it first came out on Netflix years ago, but upon looking for it this week I found out it was now free on YouTube. Highly recommended!

More MidJourney

Interiors:

A gaming console?

An autonomous delivery robot…made of cardboard?

Baroque iPhone ad: