We trained a generative adversarial network (GAN) on historic newspaper scans to produce newspapers which never existed and contain no coherent information. A literate human viewer is burdened with the conceptual understanding that newspapers contain words which are meant to be read and thus has no choice but to read them even subliminally, but this GAN trained only on image data (pixels and pixels, black or white or grey) is subject to no such compulsion. In this way, the GAN functions as a man-made way to divorce ourselves completely from our unshakeable man-made understanding of what a newspaper is. This project contrasts the antiquity of print newspaper design aesthetics with the technical modernity of GANs, and unites these two seemingly disjoint things to create a visual result that is simultaneously meaningful and nonsensical, familiar and foreign, absurd and mundane.
Read the more detailed report (the "Introduction" section functions as an artist's statement).
Explore the digital zine to see fake pictures of people reading fake news next to the fake news they're reading. (This digital zine is in the process of becoming a physical zine, printed on newsprint, to complete the circular life cycle of these historic newspapers from print to scan to GAN output back to print.)
Alternately, browse the gallery.
This project was done by Miranda Li (when she used the pronoun "we" earlier on this page and all throughout the report, she was lying; "W" makes a more aesthetic drop cap) for CS230. She likes computer science, art, parentheticals, and newspapers more than a lot of things.
Speaking of print design, there are a few somewhat meaningless Easter eggs in the design of this page: the specs for font size and the width of this text block were stolen from The New York Times' website; the drop cap was highly intentional; and all of the writing, to the author's knowledge, follows AP Style except that it uses the Oxford comma.