This AI Art thing may change Design after all
“Oh, good — another essay about AI-generated art!” 😆 I don’t have much to say that hasn’t been said already on this crowded topic, but here’s a quick personal take given my experiences so far. If you’re reading this as part of my Design Portfolio, you’re likely the right audience.
AI is here. It seems … fine.
I’ve been there with the rest of you, reading about the emergence of Artificial Intelligence (AI) in creative work — “knowledge work” that we think (or, thought) is so precious and “human”! — and trying out new tools as they come. I’ll happily follow the story as it develops, but so far (late 2022) I’ve been somewhat under-impressed with the AI writing tools (like Rtyr), and AI “art” generators (most famously, DALL·E 2).
They’re fun to play with on a “blank canvas,” but it’s been challenging for me to see how they fit into “real work” and purpose-driven creative efforts. Any “prompt” I can give to one of these tools has to be targeted enough to generate meaningful results; but by the time I’m able to conjure up that targeted prompt, I may as well just be writing on my own. Something like: write me Web copy for a product with these 4 features and this main selling point — a real set of prompts on Rytr — is a hilarious exercise given that the human (I) just did all the hard work and the AI just fills out often-inaccurate filler copy.
I’ve had vaguely similar interactions with AI art tools : the quality and “communication” of these pieces just aren’t consistently up to snuff for professional work, and often I have to really work at those prompts to get these machines to generate something workable. At the end of the day, I might as well have spent my time finding honestly pretty great stock photography on Unsplash.com, or just designing something purpose-built from scratch.
But I’ve had reason and opportunity to re-think this lately.
AI art, like human art, can inspire
I was apparently the last one to hear—from Fabricio Teixeira and Ciao Braga’s The State of UX in 2023, which was fantastic, by the way—about the AI art tool Midjourney. And after a mere hour of bouncing things off of that AI, I think I’m starting to get it. In fact, I was agape, laughing, and feeling downright inspired at my interactions with Midjourney!
✋, wait, a quick aside
Ok, I was struggling to continue this section and thought I’d try getting an AI writer, Rtyr, to assist me out of this writers’ block. I was pretty unimpressed with what I got back for my prompt, but after asking it to expand on one phrase, it did give me this:
AI art tools are very helpful for designers because they can be used as a placeholder while the designer is working on something else, or they can be used as inspiration when the designer is feeling uninspired.
Which … seems true! And it’s interesting given that I didn’t include any keywords related to “placeholders,” etc. But I think it still reads pretty dull, and wasn’t really the topic I was hoping it’d pursue from my prompt.
Funny enough, I first posted this article the week of the ChatGPT launch, but I haven’t spent much time with it yet. I’ve heard mixed things about it so far, to the extent that I’ll assume it has the same constraints I mentioned (“writing” over “creating” per se), but I’ll update here once I have more experience with it.
Anyway … back to it . What was I talking about, again?
Oh, right, my feeling inspired and energized by the AI-created art I was getting out of Midjourney. I chalk this up in part to Midjourney’s clever setup and resulting UX: it’s a Discord instance where lots of “newbies” are all issuing prompts to the AI at the same time 🤯. You all see everyone’s prompts and all of the resulting images printed out in real time. And each prompt starts with a menu of diverging concepts that you can build off of. It’s an exciting “wide generation,” plus “accidental collaboration” from the community; it really drives home the sheer diversity and fidelity you can get from the AI.
But it’s not just the UX: it was the results. I mean, look at this ☝️! It’s kind of a whiff on the prompt that I gave, and really not per the expectations I had for it. But it was one of a diverse set of images I initially got back for a simple prompt. And while the topic is … vague … the style and tone that’s being communicated is something I couldn’t have come up with myself. Something about, primordial humankind melding with machine; ancient cave paintings of an android race, with each of us plugged into the pipes, plugged directly into our work.
It’s kinda creepy! And it almost starts to communicate something around the edges.
AI has obvious inroads for “art” and traditional creative visual work
So, yeah: the era of the human artist is over! 😱 😂
Well, maybe not yet, but I can see where AI-generated visual output is going to be a tempting substitute for its human-generated counterpart. I see “creative visual work” as often accomplishing some specific things:
- 🤖 Setting the tone — using visual signals to contextualize, most often, the visual word. Setting the emotional landscape for a piece: is it aspirational? Futurist? Alarmist? Optimistic and bright? These are very much things we try to achieve in Web design.
- 👩🎨 Communicating meaning—this is, in my eyes, the hardest one: composing a piece of visual work that conveys a specific concept that you expect someone to understand based on the visual alone. This comes up a lot in professional design settings—from infographics for a sales deck, to illustrated, abstracted expressions of what an automation SaaS product actually does (sounds familiar 👀).
- 🤖 ❓ Setting a “topic”—I see this as a bit of a mix between the two prior points: it’s not quite “a specific, explicit concept,” but it’s not just setting the tone either. I think this is the role that stock photography usually plays in professional content: computer on desk means “work,” a container ship means “shipping” or “commerce.”
So, how do some quick Midjourney queries accomplish the above? Here’s the initial responses to my trying to see “teammates working on pipes”; images to support a blog post about the automation tool I’m working on at Parabola:
Well … they’re doing something already. I think it’s easy to point to point 2️⃣ above ☝️—communicating specific meaning—as an outright fail here. And I think that’s something that human visual designers are going to have the leg up on … at least for a while.
But I think accomplishing 1️⃣ — setting tone—is already on the table (and expressed in a few different ways). I could see simply dropping that top-left one into plenty of blog post headers or backgrounds: it has notions of industry or progress (pipes), some sort of horizon(s), and a general enough “hotel abstract art” 😆 aesthetic that you could put on a website and … nobody would bat an eye! I could say the same about the top-right bit.
The bottom two bits … likely need some refinement or retry (which you can, fortunately, easily do with AIs). But depending on the context, they’re creeping closer to 3️⃣: conveying tone and something of “topic.” If we were talking about the Designers’ ultimate fate of being plugged into their computers (machines, pipes, gears) and turned into productive automatons, then we’re getting partway there. (That hypothetical piece sounds a little dark for me, but these images would suit it).
This ☝️ is another riff from the image in the article header: “Space miners working on a workflow, oil, ambient, aspirational”. It’s weird in a few ways—these almost-human shapes are apparently a staple of AI-generated art for the time being—but it’s already conveying a lot of tone and even a little bit of “topic.”
The result is, again, a bit darker than what I was going for. I see: the future “Knowledge Workers” wandering a content-filled wasteland (once called “Earth”), scrounging and collecting ideas instead of generating them. The post-creative human existence. The cycle of civilization, somehow reset.
But maybe it’s just getting late 😅. The point is, some tone and vague topic is being implied: if I went to an online article and this was the first image at the top, it’d set some expectations as to what I was reading about—the future, and not a “bright” one—and the posture of the author towards this topic. And … that’s 1️⃣ and 3️⃣ from my list. That’s what a Designer is often asked to do.
One thing that I think could close the gap on 2️⃣, though, is to have AI tools better able to interface with design tools: to have an AI generate some ideas, then hand off the Design “files” to a human Designer. But AI art these days all appears to be simple pixel/raster art these days, likely related to the way that all of this visual data is collected and then re-rendered based on prompts. Maybe someday this actual “handoff” will be made possible, though, and this whole “a Designer can use an AI when they get stuck” notion will have more legs.
AI may not be overtaking “Product Design” (etc.) yet, but the roots are there
I’ve heard the notion that AI art is “going to take down traditional creative design” before it gets to “Product” or “UX/UI” Design. And I think I get that reaction along a few lines.
Barriers to the imminent AI takeover of Product Design
- 🙋♂️ UX/UI/Product Design emphasizes a “target user(s),” and thereby the role of interacting with those humans to figure out what they need (not just crunching a bunch of imagery and spitting something out).
- 📐 This kind of design is trying to solve things similar to 2️⃣ above: to communicate exact meaning, not just tone/topic.
- ✏️ Human Interaction (input/output) is a dimension than “art” or “visual design” doesn’t take on (it’s, rather, often output-only)
- 🎛 (Related to the “target user” thing) it’s not just concept, tone, or information that a Product Designer need to convey in a static way: much of the job is about which things to convey and when. How much to conceal or reveal certain elements, for whom, and in which ways.
So, it feels like this design has more of a “moat” from AI than, perhaps, more purely creative/generative art and visual design. But it’s not hard to see where AI can start to take over for us, too:
Signs that the AI takeover will creep in anyway
- 🦜 Conventions are key in web and UI design: users increasingly have expectations for what controls do, where to find certain information, what icons mean, etc. etc. Conventions often benefit the end user — even if they can feel “uncreative”—but our dependence on these quasi “best practices” means that a computer could likely learn to design a pretty good UI.
- 🏗 Design systems imply a top-down, logical application of design. I love Design Systems, but it’s easy to see a computer picking up that kind of componentize-able, centralized design and taking away a lot of the “logical work” that Design Systems Designers do.
Off-the-shelf design systems are already showing the playbook for this takeover, in some ways: the system provides the parts, and the Designer (or often, a non-Designer) chooses what pieces to use and how to use them. So maybe that “system/AI does one part, Designer finishes the job” is a natural fit, once AIs get better and better at this.
Pardon the train-of-thought here; I’ll continue to update and refine this as I spend more time with these kind of tools. Stay tuned 👋.
