Why AI Can Never Replace Designers

Visual averages and human ingenuity

Randal Cumming
The Generator


“Visual average” illustration of a coastal road via Midjourney

As the CEO of a visual media company, I’ve had hundreds of conversations about AI over the last year. Many of our clients, photographers, and designers rightly see the emergence of generative AI as a seismic shift in the design space.

Although AI will impact designers, it’s important to put things in perspective. Our company has been around for 20 years, so we’ve seen a lot of new, impactful technologies come to market.

Based on that history and the conversations I’ve had over the last year, I know that AI will never replace human designers–but it will make them vastly more productive.

Here’s why.

The Visual Average

Users often assume that AI image generators create wholly new visuals. When you type a prompt into Midjourney or DALL-E, you do indeed get a new, never-before-seen image out of it.

In reality, though, while the image itself may be new, the visual concepts driving it are not.

That’s because of the way AI systems are trained. Systems like DALL-E and Midjourney learn to create images of places or concepts by reviewing billions of existing images.

As they pour through these massive numbers of visual examples illustrating a specific place or concept, they pull out little bits of meaning from each image they see.

When they’re later given a prompt, AI generators essentially produce the visual average of all the related images that the system has reviewed. All of the system’s learning gets condensed down into a single visual that reflects its training data in as balanced a way as possible.

If you ask for “a car”, for example, the system will pull little bits of meaning from all the millions of car images it has seen — vintage cars, race cars, concept cars, cartoon cars — condense them down into a single perfectly “average” car, and spit out the result as an image.

I don’t mean that the images look average — in fact, they’re often quite striking. I use the word “average” literally, in its mathematical sense, as the mean value between many points of data.

The images produced by AI generators are the visual, mathematical average or mean of all the training data the image generator has seen related to a visual concept.

As a creative, I find that fascinating.

If you type a prompt like “coastal road” or “modern architecture” into Midjourney, you’ll get a visual that shows the average of what the whole of humanity thinks such a road or building should look like.

It’s like polling the hive mind; using an AI generator, you get an instant idea of how people, taken as a whole, think the world is supposed to look.

The visual average of Modern Architecture, via Midjourney

The New York Times wrote about this in a more flowery way, saying that AI images are like the ideal visual forms imagined by the Greek writer Plato.

The ability to find the visual average of a concept is an interesting exercise for a designer. Using an AI image generator, you get an immediate sense of what the typical person might visualize when they think of a term

It’s important to note, however, that the images you’re seeing aren’t really new.

Again, there’s no spark of creativity here. A system designed to return mathematical averages can’t create something unique.

The AI is just mirroring back to you a visual average of everything it’s seen before. It can give you a barometer for what the masses think something should look like, but it will never come up with a concept or idea that’s truly unique or novel.

The Value of a Brand

Likewise, while AI image generators can create visuals at a massive scale, for most brands the visuals are only a tiny part of the equation.

A strong brand like Aston Martin or Volkswagen is not just about products, marketing, copy, or beautiful images. Brands are about history, emotion, craftsmanship, people, and community.

Certainly, images created by AI can help a brand convey its message. But they can only do that when they’re carefully curated by humans who fully understand the brand’s vision, what it seeks to convey, whom it plans to reach, and (perhaps most importantly) the topics it would avoid.

Brands have gotten into trouble when they’ve forgotten this. If you let a computer dictate what your brand stands for, you’ll end up with messaging that’s dull and flat-footed, or at worst, offensive.

Computers don’t understand the kind of human emotion and history that’s tied up in a brand. Only human designers understand that.

If you’re a car person, try to describe how a Ferrari makes you feel, in a few words that a machine could parse and understand. It’s impossible.

Although AI can serve designers as a powerful tool, it will never replace them and their knowledge of what motivates humans to connect with brands.

The Value of AI Images

Although AI will never replace the human ingenuity of a skilled designer, AI will cause massive changes in the design industry.

While it can’t imagine things that are wholly new, it can take a seed or spark of human imagination and expand it exponentially, often in no time at all.

AI excels at creating versions. If you give an AI generator an idea that’s wholly new and human, it can give you 100 or 1000 versions of that idea in a few minutes.

That’s an incredibly valuable capability for designers. We can bring our ideas to AI, and see them built out or expanded upon almost instantly. From the vast number of options this gives us, we can apply our human creativity to curate the best results.

We can also validate our ideas by using AI to see how close or how far they are from norms and visual averages.

Good design is often about striking a balance between innovation and familiarity. Designs that are too familiar are boring, but designs that are too innovative are often impractical.

Because AI provides a visual average taken across millions or billions of images, it can help designers see where their ideas lie on the spectrum between familiar and totally new. This can be an important reality check in the design process.

Comparing your ideas with a visual average can help to ground them–or to reveal that they’re truly new and different.

All that’s to say that AI is an incredibly powerful design tool, and designers need to start engaging with it now.

AI will rapidly expand the capabilities of the best designers, and help established brands create far more visual content, much faster.

But AI will never replace the brilliant spark of ingenuity that a designer brings to the table. That spark rests only with us humans.

Read more of Randal’s writing and more about the visual design space by joining the CGI.Backgrounds newsletter or email him your thoughts at randal@cgibackgrounds.com