Algorithms parsing, evaluating, generating and playing with strings of text is my jam. One way and another I’ve been involved in SEO for over 10 years. And a significant part of SEO is computers understanding text at scale.
But a subtle implication is that I’ve been involved many times over the years in teaching, training and working with human writers who are trying to write as humans. Because if computers think they’re computers then no bueno for the Google gods.
Anyway - you would not be surprised to know that I’m also a big fan of digital art and using neural nets to generate art.
So along walks Robin Sloan, a more talented programmer, thinker and writer than I and… he just nails it. First came writing with the machine where Robin experimented with plugging a RNN trained on sci-fi writing into his text editor:
If you haven’t read that piece I highly encourage you to go read it.
But then I stumbled across his Eyeo talk from earlier this year and it goes way beyond that blog post. Robin’s commissioning custom machine learning algorithms, experimenting with a writing interface that doesn’t involve text input and so much more. I can’t recommend this more highly:
It also includes the magical phrase “I want to play this one just as it came out of the neural network…”
Anyway. Turns out the link Robin mentions as turning him onto machine learning is easily accessible. Check it here: The Unreasonable Effectiveness of Recurrent Neural Networks.
So I loaded up all of the mathematical shapes found here: https://en.wikipedia.org/wiki/List_of_mathematical_shapes. And we’re off the races. My basic RNN generated some beauties. A handful of my faves:
None of those have ever been seen by Google before… For the maths-curious I uploaded a sample of 6000 or so outputs from my RNN. Feel free to do whatever you like with it!
Update: this video from the Cybernetics.social conference by Allison Parrish is wonderful and totally on-theme for algorithms and text (and poetry!) so I’m including it here: