AI and the type industry, article.

Jamie Clarke

Hi All, 

Last year I heard a talk about AI and type by @Filip Paldia, who's PHD subject is "Making Aesthetically Acceptable Fonts by Deep Learning Methods". It made me step back a while to consider what on earth was going to happen to our industry and even whether I should keep going or pivot into something else (type design is already my third design career, and I love it).

After reading the insightful posts on TypeDrawers, especially this thread:  https://typedrawers.com/discussion/comment/64337/ , and chatting with type friends, I eventually came to the conclusion that AI will certainly disrupt the industry but not destroy it. In fact it will play a critical role in its evolution beyond 'static type' much like @Ray Larabie suggested. 

I reached out to Filip and we co-authored this article. It's aimed more at graphic designers, but would be interested to hear your thoughts, even if it's a load of old codswallop*.

https://www.smashingmagazine.com/2024/01/ai-dilemma-graphic-design-typography/

*yes, 'Codswallop' the font name is already taken, like everything else, I just checked.

Miles Newlyn

I look at the speed and breadth of Variable fonts' transformation the font landscape and think, 'OK, business as usual'.

Jamie Clarke

It's true, the take-up is glacially slow. (I've taken a poll of how many people have used a variable font, in a room full of self-confessed type lovers, and only a few raise their hands). Having said that, VFs are a step on the path to more contextual fonts, and that path will likely be faster or slower depending on adjacent technology.

John Hudson

In the article, you say

The next evolutionary step will be dynamic, context-sensitive typefaces. These would provide more nuanced and precise forms of communication, tailoring text to specific contexts and user needs.

I would have more cautiously stated this as something that dynamic, context-sensitive typefaces may be something that happens (I would also have avoided the term ‘evolutionary’). I think the notion of tailoring text to specific contexts and user needs requires a lot more unpicking of assumptions if the results are not going to languish at the level of superficial and banal. Predictive AI, unsurprisingly, tends towards and reinforces cliché.

GPT-4 Turbo did, however, make this attractive but totally random and meaningless image when Adam Twardoch tried to use it to help me find a test word for Sinhala shaping behaviours.

https://twitter.com/adamtwar/status/1750531216938541385

Which makes me think that deep language models might not actually recognise the boundaries of language. 

Dave Crossland

John, you seem to be missing the forest - the Latin texts, where the typeface was generated and typeset by the AI - for the tree of the nonsense sinhala shapes....

Jamie Clarke

Thanks @John Hudson, That's interesting. You don't think type will evolve to break free of its 400+ year old static state? I feel pretty confident that barrier will be broken eventually. It feels like the next natural step, though currently way beyond our capabilities to create or reproduce.
Love that GPT image!

John Hudson

You don't think type will evolve to break free of its 400+ year old static state? I feel pretty confident that barrier will be broken eventually.

That barrier has already been broken in several ways by digital typography, so that is not what I am saying—and not what I wrote. I said that I prefer ‘may be’ to ‘will be’ when talking about specific future directions in text, especially when accompanied by unexamined notions. For example, what does it mean to ‘tailor text to specific contexts and user needs’? And what ways can that question be answered that are not the most obvious responses as would occur to a first year design student?

As I have written elsewhere, I also have concerns about the directions of current AI with regard to non-Latin text. Yes, as Dave has pointed out, the handling of English text in AI generated graphics is getting better pretty rapidly (but still inserts unneeded letters, is often badly spaced, sometimes inconsistent, or simply degraded—all seen in the labels in that GPT image), but there is a pattern in North American-driven tech developments of ignoring the needs of the text needs of rest of the world until long after English text has been solved, and of solving English text needs in ways that actively disadvantage users of other languages and writing systems. As one of my Indian colleagues put it with regard to OpenType, every time US tech giants introduce a new technology, ‘the clock gets reset’ for everyone else.

[The current AI hype also reminds me too much of the dotcom bubble in the 1990s, so I think there is a greater than 50% chance of this generation of the technology becoming an investment sinkhole that we’ll have to ride out before we end up with something useful.]

Jamie Clarke

Thanks John. That makes sense.