LOOKING AT TYPOGRAPHY developed by artificial intelligence is like looking at lettering submerged in deep water, warped and fuzzy. It looks like a copy of a copy of a copy. The words are recognizable, barely, but the original form has been lost. AI typography is, charitably, bad.
A recent example of this phenomenon is Word-As-Image for Semantic Typography, a paper in which anonymous authors propose a tool that morphs text into an image of what that text represents. Type in “yoga,” for example, and the word will appear garlanded with wobbly vectors of stretching women. The resulting jagged, blurry text is emblematic of the shortcomings of AI type. This experiment sacrifices readability and accessibility, two of the pillars of good type design, in a misguided attempt to innovate. We could hardly expect much more from AI, however, when it has only a surface-level understanding of how humans read.
As a designer and typographer of more than 10 years, I’ve watched the progress of AI-powered design with a mixture of amused curiosity and subtle dread. Where typography is concerned, it’s becoming clear that AI innovations are focusing on the wrong ideas. Right now, some are playing with using this technology to try to redefine visual language—in the case of our Latin letterset, one that’s existed for over 2,000 years—but ultimately this is an unworkable course. The key to setting AI typography on a better, more accessible path is to think of it as assistive rather than generative.
Word-As-Image isn’t novel. After the Industrial Revolution brought machines to the forefront of manufacturing, designers in post-war Europe started exploring how technology could influence the future of art and type design. In his 1920 book Sprache and Schrift, engineer Walter Porstmann proposed that language could be amplified by introducing one character for every sound, ordered by tone, sound length, strength, and voice. László Moholy-Nagy at the Bauhaus later adopted and refined Porstmann’s concept, anticipating in 1925 that typography would be supplanted by advancements in film and, especially, sound. In response, he suggested, typography needed to evolve to express these new technologies.
Perhaps the most interesting response to Moholy-Nagy’s phonetic proposal was Kurt Schwitters’ Systemschrift. First published in 1927, it was a unicase alphabet that used character weight to denote phonetic emphasis, conveying vowel sounds with boldness. This experiment was remarkable for its visual eccentricity; it stood apart in a school that favored more standardized typography. But that doesn’t mean it was effective. Not even Schwitters used these phonetic elements in his own work.
Looking at both AI type and these 20th-century typographic innovations, one can reasonably ask: Who is this for? Certainly not readers. But like earlier experiments that fused technology and typography, it’s possible that AI could lead designers to create better type. If AI can be used to help typographers, rather than to try to supplant them, generative models could just be a blip on the way to a more efficient and accessible use of this technology as an assistive tool in the type design process.
Think of how the digital revolution put typography in the hands of everyone with a computer and made the process of creating it more efficient than ever. AI could be applied in similar ways, assisting typographers and making their work more accessible. But it’s important to consider where to place that assistance.
“Typography is such a nuanced practice that relies heavily on the human hand and optical illusions,” Craig Ward, a design director, points out. “And much of it doesn’t even make sense to those well versed in it.” We’re not one hundred percent sure why we make horizontal strokes thinner than verticals nor why circular characters sit below the baseline and overshoot the x-height. Short of studying the optical science of how we read, the truth is we do these things because it would look odd if we didn’t.
That indelibly human and instinctive influence on typography is a major obstacle to AI’s application to the type design process, even on the mechanical side. “One visible and pervasive flaw among AI-generated type—and it’s a biggie—is the lack of consideration for a type design as a system and not an image,” Zeynep Akay, creative director at Dalton Maag, says. To date, AI fails to recognize that typography is a series of systems with its own conventions, not just a picture or visual representation of the spoken word.
“At the moment,” Akay adds, AI “is not sophisticated enough to make adjustments on a given design based on parameters like legibility, readability, and likability, and make them with that systematic consistency in mind.” But if we reframe AI as a tool for assistance rather than generation, we can potentially make the design process leaner, more approachable, and more accessible to a wider group, as other technologies have done before.
One potential AI application I’ve found in my own practice pertains to written languages that are losing speakers (mostly due to colonization) and subsequently dying out. Endangered alphabets like these need workable digital representations that are archived and made available for use so they aren’t lost from the historical record.
In 2019, I digitized Kayah Li, a language decimated by the genocide in Burma and the persecution of its Karenni speakers. While doing this work, it became clear that projects like these are struggling for personnel. An automated solution could be helpful in such undersubscribed work. AI could analyze the letterforms of these alphabets from existing materials—digital scans, photos, or handwritten documents—and create accurate digital representations that people who wish to speak the language could use.
Can we trust AI in its current form to respect the immense cultural impact of an endangered language? Perhaps not. “AI, as it has been proven time and again, can exaggerate inherent biases and have an overwhelmingly Eurocentric approach,” Akay told me when I broached the subject with her. “Global type design has come a long way, and it would be a step backward if AI were allowed to infuse any such biases into languages that deserve thoughtfulness, sensitivity, and specificity.”
By repositioning AI as a mechanical rather than creative typographic tool, we may be able to further streamline the design process in a way that ultimately helps speakers of imperiled languages preserve their native tongues. This technology could take over laborious yet precise tasks like spacing, creating new weights of a design, and analyzing kerning pairs to make the process more efficient and the work more accessible. History tells us that this could happen. One could understand AI as part of a typographic revival like the 1920s upheaval. We’ve seen that attempting to disrupt visual communication (as generative AI does) is bound for failure. But if we think about this technology in terms of how it might assist humans rather than usurp them, it could help us to create a leaner, more accessible, and more pleasurable type design process. To build from the past a more readable future.
No comments:
Post a Comment