A Designer Tries to Generate Fiction

Hello, everyone!

Spring has always been a crazier semester than fall, and this year is no exception. Beyond teaching and coordinating graduate admissions, I’ve also been presenting and collaborative archive-building. This week began with the solar eclipse and culminated in days of rain.

A few days ago, I gave a presentation as part of my university library’s digital scholarship showcase. What follows is adapted from a transcript of this talk, about my exploration of AI to fictionalize historical design texts.

I’ve lightly edited this for length, clarity, and readability — such as removing multiple occurrences of “um” and “you know” because, um, you know… Enjoy.

—

Hi everyone, thanks for being here. Minor design texts tend to be overlooked and forgotten, yet they hold promise for understanding aspects of graphic design practice and theory. These include things like essays and comments on design websites or blogs, letters to editors, editorial introductions, and manifestos or visionary statements of some kind. This is design writing that's published and often highly opinionated. Many of them are written by designers, students, teachers, and others, named or anonymous. These texts often have popular appeal but low literary status. Many of the issues discussed such as education, ethics, creative freedom, and business remain relevant today.

My long-term project, The Designers Respond, is based on these minor texts. Throughout my time with this material, I've encountered three big challenges. The first is, for lack of a better term, findability. Generally, these texts are difficult to find. Some design websites and blogs have removed their archives entirely from public access, but they can usually be found using archive.org — but only if you've got a URL.

Secondly, managing and making sense of large volumes of text. Some design website blogs have hundreds of comments, and many debates spread out over time and across platforms. Finally, there are copyright issues. Magazine content can be quoted like any other literature, but website comments are in a gray zone because of each site’s terms of use. I've received a lot of advice not to quote comments directly because of this.

Today, I’m looking at graphic design websites or blogs, very early 21st century things that were being published around 2003–2004 until around 2012.

There are two main design debates that I've been focusing on. The first one, "Posters for Japan" is from 2011. It spanned over five websites in two months. Those are mostly blogs, the websites, the debate focused on posters and the ethical implications of posters — many self-promotional— that were produced/sold to help Japan in the days following the Fukushima Daiichi earthquake, tsunami, and nuclear disaster. The second is a design debate from the summer of 2004, which spanned two magazines and three websites over a year. This particular debate focused on the sex-themed magazine special issues that were published by Print and STEP, which were both popular industry graphic design magazines.

I've decided to fictionalize these historical dialogues. The study has been a new direction for me as it involves writing studies rather than visual textual artifacts. But, I suppose it’s fair to say I’m designing written dialogues.

So why fictionalize them? This is where I landed as I was looking for ways to interpret or share the things I'm finding in these blog comments. The first reason goes back to copyright issues. Besides terms of use, many design readers may not have anticipated their words coming to life again 20+ years later. In the early days of Web 2.0 commentary, people may not have anticipated that their words would never truly go away. I'm also doing this for future engagement with readers and viewers in the future. My goal is to exhibit or publish in some way this project, and this is a way to aggregate commentary, identify themes, and rewrite in a way that is engaging — and honors the original contents.

Alas, I'm not a writer of fiction. I read a lot of it, but I have very little experience doing any of this beyond stories that I wrote as a kid. I thought, hmm, maybe ChatGPT could be useful in some way…

The question really wasn't can ChatGPT help but how. Of course, the tool can churn out whatever based on prompts. This itself is not difficult. But the question kind of became, how to do this in a way that actually works? I started playing around with it, exploring commands and text entry and actually feeding in some of the design debate content, and there were some immediate red flags. First off, it was generating way too much text. In some cases, it was quadrupling the amount of text that I had originally entered. It also felt disrespectful to these original conversations. I decided I would not enter any more actual blog commentary (data) into the tool.

Instead of asking AI to help write this new content, I decided to use ChatGPT to help me explore narrative personas. For those of you who may not be familiar with personas, here is a very simplified version of what I'm talking about. A persona, typically, is a fictional character that you create based on research to represent persons or individuals in whatever kind of project you're working on. In design research, we focus on user or buyer personas, which take into account product or feature preferences, and demographics such as age, education, and so forth. Narrative personas are different. These are more connected with writing practice. They are developed to identify voices and manners of speaking, and individual characteristics. This is what I wanted to go with toward fictionalizing historic debates and dialogues.

Instead of putting actual blog comments into the tool, I wrote this very generic, very bland phrase: “This is an important conversation. I see your point, but I disagree with your statement.” This is the kind of stuff I see all the time in comment sections (though not always as civil). I developed three voices based on the personalities I observed. Voice 1 is impatient and controlling, Voice 2 is soft, spoken, and reserved, and Voice 3 is a curious novice. ChatGPT rewrote my generic/boring line in these three ways. The tool outputs phrases like “cut to the point” to Voice 1, “respectfully” to Voice 2, and “I'm eager to learn more” to Voice 3. The results were okay. Could we do better?

I tried this a second time, and using the same phrase, I tried expanding on the descriptions for Voices 1, 2, and 3, thinking that maybe the tool would actually give me higher-quality suggestions. To my surprise, ChatGPT actually outputs less text than before. Phrases like “hash this out” and “respectfully disagree” and “this is a big deal” are still there and ultimately not much different from the first go-round. The tool could not output a whole lot of creative writing, but it showed me things that these characters — narrative personas — might say.

One of the things that had been on my mind since the beginning of this particular exploration was what's more important: what is said or how it is said? Maybe they're both important. Maybe there is a different balance that I'm not seeing, but this resonated with me throughout the study. Secondly, AI is simply not effective in fictionalizing historical textual dialogue — at least, not without a lot more input from me. It provided some things as I mentioned, in my first trials, it gave me too much. Maybe there's a place to use this more in the future, though I’m unwilling to use more of my energy to hone the outputs.

Going forward, the process itself was more helpful for me than anything else. Working on this forced me to think differently about how I would construct the voices for these fictionalized debates. Instead of looking for ways to enlist AI to generate fiction, I’m developing an offline collection of narrative personas based on these studies. I don’t intend to use AI but rather, methodically work through these personas, their voices, and what they might say about design histories.

[end]