This Is What Happens When You Let A Neural Network Design Fonts

Neural networks are increasingly taking on jobs that used to be the preserve of the human brain. So Erik Bernhardsson decided to see what would happen if he threw 50,000 fonts at a neural network and left it to chew at them. The results, it turns out, are pretty interesting. To feed the fonts to the AI, he created grids of characters — each character fitting in a 64x64 pixel box — so they could then be directly compared with one another. In the process he got the neural net to create what he calls a "font vector" — a kind of abstract mathematical construct that singularly defines the font. You can read about the details of how he did that in a blog post here.

More interesting is what can be done with the resulting vector, as Bernhardsson explains:

Since every font is a vector, we can create arbitrary font vectors and generate output from it. We can... pick a font vector and generate new fonts from random perturbations... We can also generate completely new fonts. If we model the distribution of font vectors as a multivariate normal, we can sample random vectors from it and look at the fonts they generate.

In fact, that's what you can see in the gif above. It's pretty smart, too: It's learned, for instance, that many fonts use upper case characters for the lower case set, and it intelligently switches between the two depending on the type of font it's creating.

Neural nets might not take over the job of designers just yet, sure, but it's a pretty cool project that demonstrates just how versatile they can be. You can go and read all about it here.

This is interesting. He throws images of font letters at it and it generates it's own based on a shared set of pixel vectors. It will be genuinely useful once he throws a bunch of federation starship images at the software and allows it to create it's own ship designs.

It will still be up to the human "designer" to choose the aesthetic/parameters they desire with the use of a font. (True of any design task.)

Do we really need AI to inform us of what we like most? (change is human)

AI (neural network isn't actually AI...no-matter, who gets to define what...) will tend towards "efficiency", "artistic mores" will vary over time as human trends change (not necessarily for efficiency based reasons).

NB. Who designs the cost function which determines what is efficient or not? (AI can't actually set its own boundaries or limits, so the Arch-Itect will always be needed.)

An overabundance of (too many) parenthetical comments (in brackets) makes one's arrant nonsense (bullshit) appear to be thought provoking and insightful (pretentious bullshit).