Artificial Intelligence is Upon Us...
There’s a big hubbub in the Design community about AI taking people’s jobs, rendering the profession obsolete in the near-to-late future. With all the commotion, I’d be lying if I said the concern hasn’t invaded my brain a tiny bit, as well. So, I thought I would put some thoughts out there based on my own experience, arguments I’ve seen others make, and maybe even some cope.
Let’s start with what artificial intelligence is already doing well, to get some insight into why some designers are worried. AI tools like DALL E and Midjourney have proven themselves capable of generating some astonishing imagery based on text.
The concept is simple. You type any words you like into the submission box; the AI comes up with a series of images that it thinks fit those words. It is truly world-changing stuff. Want your image to look like a painting by Van Gogh? It’s got you covered. Need something more photorealistic? No problem. Cartoon imagery? You bet. This technology feels like something entirely novel and it is very impressive.
That’s not to say that there aren’t some issues. For example, these image generators don’t tend to handle text very well. Strangely, the same technology that can spit out an image of a dragon with chicken feathers riding a tricycle has a really difficult time putting a single legible word together (see cover image of this blog post). Surely, this is something that it can learn with proper inputs, but the heart of the issue comes down to the way that the machine has learned to do its job.
These AI image generators have been given millions upon millions of images from the web to analyze. Essentially, they are looking at how pixels interact with each other and using that data to form new images that, to us, look realistic, or at least passable. When analyzing the input images, though, it isn’t taught what words look like or anything about language. It’s not reading any of the words in the images it learns from, only how the pixels that form those words interact with the pixels nearby.
What you get, then, are images in which words look foreign, incomplete, or outright strange. It doesn’t know that the pixels in the source images are parts of letters, which are parts of words, which are parts of phrases. It only knows that this group of black pixels was near this group of white pixels.
I think this is a hurdle we will see overcome in the relatively near future. Surely, someone out there is working on an AI typeface generator to create fonts based on text input. I’m a little surprised that this didn’t come first.
Another thing that current AI image generators sometimes can’t handle well is phrasing. You need to be careful with the way you phrase the input you give to it. The AI is drawing connections between the different words you give it, but it doesn’t really speak your language. For example, last week I was having a conversation with my wife and—who knows how we got here—but we arrived at a concept that I thought was funny: Swiss cheesecake.
As soon as I heard the phrase come out of my mouth, I got excited to see what DALL E 2 could come up with given this phrase. In my head, I pictured a slice of cheesecake with holes in it. Easy enough, right?
It turns out, this was a difficult image for DALL E 2 to come up with. I am a little embarrassed to say, I have used a huge chunk of my credits trying to come up with the right phrase to get the image that I was looking for. The winning search? “Photo of swiss cheese with holes in the shape of a slice of cheesecake.”
Even this image isn’t perfect, but I am willing to call it good enough. The point is, the information that you give to the image generator isn’t given any human thought. Sometimes, it seems to know exactly what you’re looking for and serves up four great options. More often, it gives you four options that each contain a different interpretation of the phrase you input. Occasionally, it just won’t understand until you find a new way to phrase things that is incredibly specific and kind of a hassle.
There is a learning curve to this, like all new technology in design. The average person can have an absolute blast playing with this technology, as I have, but for commercial applications you will need to know how the tech works and how to use it. Even then, it’s not going to always give you something you can use.
That alone is enough to convince me that this is not going to make designers obsolete, but let’s take it a step further by looking deeper at some history in the arts. In the mid nineteenth century, a new technology cropped up that allowed people to capture a moment in time that was more realistic than anything people were capable of before: photography. At the time, many painters loathed the new technology, fearing that it would make their profession obsolete. There was an outright refusal among many artists to accept photography as an art, maintaining that the arts were about using your hands and mind to create something beautiful—as if this is something photographers were incapable of. They said that it was too easy and that it required no skill, thus photographers had no place in their creative space.
We all should know, by now, that these were silly accusations. In reality, talented photographers used knowledge of art and composition to refine their craft and push artistic boundaries. Having a realistic image of a moment, in turn, ended up teaching painters and illustrators a lot about perspective, lighting, and—funny enough—composition. In the end, the two forms of media learned from each other, and both got better. I see the same thing happening with AI image generation. It’s a tale as old as time, so let’s check another example.
When computers and software like Adobe Photoshop came along, it revolutionized the design industry. Designers had long used rulers, protractors, knives, and skilled hands to create their work. Now you’re going to tell them that they can create their work without hand-cutting and collaging? Steady hands were no longer needed to do the job, so steady-handed professionals worried that their job was now obsolete.
Those who refused to learn made that statement true. The ones who chose to learn to use a tool that would make their lives easier thrived.
Great design evokes feeling in its audience. It grabs attention, it tells a story, it suggests action.To say that AI image generators are going to make anything in design obsolete is, in my opinion, panic. The only jobs that will be eliminated belong to people who are more willing to throw their hands up and admit defeat than to learn to use a new tool.
When you break down design into form and function, at its best these AI tools can help with form. However, design is so much more than what something looks like. Like any art, great design evokes feeling in its audience. It grabs attention, it tells a story, it suggests action.
In the coming decades, as the technology grows, it may briefly take over the way that we see image generation. It is obvious that AI gives us the opportunity to mass-produce imagery at a scale that we have never seen before. This will be a popular and cheap option, but if we look at history one last time, we see that the Industrial Revolution gave birth to the Arts & Crafts movement as people got tired of seeing the same mass-produced stuff everywhere.
I think we will see a similar progression with AI, especially as it begins to use its own images as a source for future images. Over time, conformity will present itself. The cost of the product will be reduced beyond anything previously possible, but the product will be missing a soul.
This technology is simply incredible and game changing. However, I don’t view it as something to be feared. For designers who feel like they’ve learned all there is to know, perhaps they should be worried. Always having something to learn, though, is part of why I joined this industry. I love to learn new things, test them out, and find a useful way to implement a new tool. This is an impressive tool, and I look forward to continue learning how to use it.
Oh, and since you asked... here's that dragon with chicken feathers on a tricycle: