Human + A.I. poetry.
Generated by a computer. Edited by a human.
05.2017 - 05.2018. One book a month.
A limited edition boxset of 12 poetry books written in one year by digital poet David Jhave Johnston with neural net augmentation. ReRites is accompanied by a book of 8 essays written about the project. Published by Anteism Books, Montreal (2019).
ReRites exhibit format also includes over 120 hours of neural-net text-generation videos, 15 hours of editing videos, and often includes a participatory performance component called ReadingRites.
Human brains are capable of nuanced insight, but our inspirations are ephemeral, and our energy, finite. The core claim of ReRites is that using neural nets empowers and nourishes limited human creativity: the digital pattern-devouring process of assistive technology not only operates at a far faster rate than humans, but the silicon neural nets’ capacity to ingest data is vast, and their energy (capacity & consumption), relentless.
Yet ReRites also clearly establishes that contemporary—and as such, disembodied and unemotional—neural nets will never produce coherent, contextually-sensitive poetry. These nets are limited by lack of life-experience: they neither eat, sleep, love, dream, shit, or bleed. To create, neural nets need to first learn from human experience; ecstatic agility arises only when these very different modes — embodied human and disembodied algorithm — conjoin. In the case of ReRites, that human experience came in the shape of data (human literature) and human editing.
A block of A.I.-generated text, massive and incomprehensible, can exude the presence of solid stone. Here, the cursor exists
chisel; I called this human-editing part of the process, carving.
It is 6 am. It’s silent. The internet is off. Mind is hammer. I carve.
In the year it took to create ReRites, many of the poems I carved had the sense of remote dreams or warped aphorisms, collaged fragments or cryptic morsels. Most did not speak in a direct way to my life or my thoughts; rather, the poems emerged as talismans, oracles, incantations, and mirrors. And each hinted at a future of writers burrowing into digitally-digested archives where apparent chaos reflects self to self and culture to self and language in and as being.
Poets & audience members read poetry written by artificial intelligence at the rate that the machine writes, often at inhumanly rapid speeds, and often bewildering and mystifying. Consider what the poet of the future looks like as you witness human performers take on an evocative, infinite deep-learning muse.
2018-2019: Printed Matters’ LA Artbook Fair; Barbican (Hub Space); Beinecke Library, Yale University; Interrupt Festival; Elektra; ELO2019; Creativity & Cognition, San Diego; Leonard & Bina Ellen Gallery; Espace4; Printed Matter’s New York Artbook Fair; In Fiction Now, Brown University; Rhode Island School of Design (RISD); Kanada Koncrete (The Canadian Literature Symposium); Mind the Gap (Electronic Literature Organization Conference); International Poetry Nights Hong Kong (2019)
The ReRites neural network code is adapted from three corporate github-hosted machine-learning libraries: Tensorflow (Google),
PyTorch (Facebook), and AWSD (SalesForce). Huge corporations, for profit, mine language for strategic advantage, using their results to
sculpt idioms and ideologies and impulses.
ReRites is a poetic intervention to demonstrate a cultural, altruistic, playful use of A.I..
ReRites continues to investigate deep learning algorithms as they evolve. Below is the 2019 OpenAI GPT-2 algorithm 345mb model fine-tuned on the ReRites custom contemporary poetry corpus generating poems in realtime.
60 pages of poems selected from the over 4500 ReRites poems; some Raw Output generated by the computer; and 8 Response essays.
Essay authors include: Allison Parrish, Johanna Drucker, Kyle Booten, Stephanie Strickland, John Cayley, Lai-Tze Fan, Nick Montfort, Mairéad Byrne, Chris Funkhouser, and Jhave.
ReRites acknowledges the generous support of CALQ: The Conseil des
arts et des lettres du Québec,
and an NVIDIA TitanX GPU donated through the NVIDIA Academic Grant Program.