The Daily Heller: There is Nothing Artificial About AI Type Design

Posted inThe Daily Heller

For Andrea A. Trabucco-Campos and Martín Azambuja, their book Artificial Typography is born out of curiosity about the AI space: They had both received access to MidJourney, an interaction experimental lab, early on in 2022, and began exploring it. The duo’s initial probes led them to prompt the AI to imagine alphabets by famous artists, architects and designers who never really practiced typography—people like Corbusier, Picasso and Bridget Riley. The results were often interesting and unexpected, but somehow familiar to the visual voice they originated from. Trabucco-Campos and Azambuja’s research has now been put together in a book. As we enter a new phase of AI, which will doubtless impact how designers will practice, it is time to address all the possibilities available.

I asked Trabucco-Campos and Azambuja to take us further into the weeds of this unique project.

The Rosetta Stone was their inspiration.

What is the origin and rationale for this project?
Martin and I met at Pentagram, and both currently work full time at Porto Rocha and Gretel, respectively. Over the years since meeting at Pentagram, we’ve shared a lot of loose project ideas, and finally decided to sit down this summer to go over them and give them structure, resulting in a small editorial imprint, Vernacular, through which we’ll publish various small and big explorations of ideas related to form, design, typography and culture. Artificial Typography is the first volume published at Vernacular; it is Volume 0 (v0).

These early AI explorations converged with the founding of Vernacular, and we saw an opportunity of making a book that was at the intersection of seemingly disparate things: art history, artificial intelligence and typography, which is at the heart of both of our design practices.

The book contains 26 letters reimagined by an AI through the lens of 52 iconic artists across various media (painting, sculpture, textile). The typographic space is especially great for this exploration, since it’s a world of ideas where general conventions rule, but where there is endless opportunity for unexpected interpretations of letterforms—always new ways of remaking the recipe of a letterform. At heart, we were curious and excited to see how wild and different letterforms were possible through AI.

How is AI being used to create these typographic forms?
Of course, AI is not new, it is something that has been developing for years. But there was a major breakthrough in 2015 with automated image captioning. As the name suggests, this study allows describing the content of an image in words. After that, it was natural to play the other way around and see what image would appear depending on the word selection. As it happens in various fields, we’ve seen how rapidly this technology has evolved from weird/grotesque results, to incredibly detailed and aesthetically pleasant results.

We are no experts on the inner workings of these AIs, but as far as we understand for our typographic images—as well as any other AI-generated image—there are massive diverse training data sets, collecting almost every image on the internet with their text description. Those captions come from things like the “Alt text” on websites (the text that is written in the actual code for accessibility and search engines). All those images are indexed in a process known as “deep learning” that collects all the mathematical information from this image, and basically assigns numbers to each pixel. After that, images are organized on the basis of hundreds of criteria, in a process called the “latent space,” which consists of a bunch of variables that track characteristics like shape, color, texture, letters, artists, styles, etc. The last phase is called “diffusion” and is the generation of the image based on a series of iterations, which connects text prompts and metadata to create new imagery. That process is random and it is what makes this tool incredible, since you cannot predict the outcome and can never expect the same result.

Is “artificial” the best way to describe this material or is there another word that better describes what you are doing?
Artificial Typography is an obvious wordplay we loved as a book title due to its immediacy and relation to Artificial Intelligence (AI). It also lends itself to a series of books we’re thinking about publishing in the future, such as Artificial Architecture, Artificial Objects, etc.

Beyond that, “artificial” is an accurate way of describing these images if we consider the mode of creation and how this word is commonly understood: something not existing/originating in nature, rather produced by human processes. It is doubly removed from us since it is not just created by our hands with tools (think a letter drawn by a pen, brush or Illustrator), but instead produced semi-autonomously by machines and systems we’ve created.

Initially, we were also enamored by the idea of “conversation” and played with it as a title. The exchange that happens with AI machines is a form of conversation, and perhaps one of the most intellectually satisfying visual-verbal connections that have been devised between human and machine. As mentioned, you feed it a text prompt and through a simple string of words and their order, the AI system generates images that are sometimes unexpected, sometimes weird, sometimes ugly, and quite often stunning. You then can keep on iterating on the images or versioning the ones that are most successful, as well as tweak the text prompt and learn how the AI reacts to subtle or drastic shifts.

Do you foresee a practical use for this method, or is it just novelty?
Due to its name, Artificial Intelligence is suggestive of a utopian reality where machines have taken over. Pockets of the design community embraced AI and are actively trying to figure out its role within our field, while others rejected it vehemently, just like photography was initially rejected by painters who had mastered their craft.

Today’s AI is a tool, and albeit a powerful one, it is limited, and our roles as designers are far from obsolete. AI necessitates a point of origin, a thought, an idea. It necessitates an editor, a curator, someone who can guide it. It necessitates an end point, too—a use for the output, a reason for its imagery. At the moment, it needs us more than we need it. Overall, in design, AI could become integral at two levels: (1) make any procedural task easier, faster and more accurate, and (2) expand the possibilities through its quick but deeply broad stylistic output. 

With any design, human control is essential. Where or how in this process have you retained creative control?
With a tool this powerful, where a large number of possible scenarios are quickly and accurately visualized, there is a shift in our role as designers. The time from idea to execution is cut exponentially, and the design control is now purely on concept and communication—how does one arrive at a precise articulation of an idea? What is its sharpest verbal articulation that will arrive for a successful set of images? What visual-verbal references and what historical references should be activated? How does the idea connect with the brief? 

Beyond image-making, working on the book pushed our role even further into pure curation, looking at several dozens of AI-generated images to arrive at the one selected in the book, and in total we generated well over 500 images, which ultimately resulted in 52 picks.

The book contains a multitude of tensions: established voices in the history of art reimagined by an amorphous digital tool that can reshape any media in convincing and unexpected ways. The decision to make it a book was to lean into these tensions by generating in a physical form a cutting-edge digital process—a conversation between the old and the new, analog and digital. 

A book is also something permanent, an object-form which we’re using to mark time for this specific era of design, capturing something quite volatile like the always-changing AI space.

How long does it take to create a letter?
We just wrote a prompt and hit “enter”—54 seconds. The prompt we ran is “Letters DH sculpted in stone by Isamu Noguchi.”