metahaven

What Finnegans Wake Teaches Us about AI As AI becomes ever more interwoven in daily life, new research from Nina Beguš, PhD ’20, and Gašper Beguš, PhD '18, is beginning to map out machine learning’ "latent space," the hidden mathematical ar...

Harvard's magazine reporting on our paper 'Latent Spacecraft'
with @metahaven.bsky.social and @begus.bsky.social
gsas.harvard.edu/news/what-fi...

Very few artworks have the actual model running in real time and generating outputs on the go.

Below is a GAN model trained on Finnegans Wake audio, FinneGAN.

Explore the latent spaces with Latent Spacecraft:
latentspacecraft.antikythera.org

@metahaven.bsky.social @begus.bsky.social

Latent Spacecraft pertains to latent space as an interiority of AI which can be made navigable. The piece tests this idea for fiwGAN (for Featural InfoWaveGAN), a speech-generating artificial neural network by @begus.bsky.social et al - see latentspacecraft.antikythera.org cc @ninabegus.bsky.social

"Latent Spacecraft: Brains, GANs, Finnegans" (DOI 10.1162/ANTI.5KP1) creates a craft to explore the interior space of a GAN. By @ninabegus.bsky.social @metahaven.bsky.social & @begus.bsky.social Interface(s) by @metahaven.bsky.social & Riccardo Petrini - latentspacecraft.antikythera.org

- AI has a physical interior space.
- By exploring language’s uncharted cognitive terrain both in humans and machines, we bring attention to the interiority of speech in brains and in computational models.

pre-speech edge, further into entropic interiority. Generated outputs portray limits of intelligibility and make the model’s interior layers a site for interpretative exploration.

- Language is a final linearizing layer over a richer prelinguistic space.

AI is creating a new kind of space, the latent space. We pair art and internal interpretability techniques to explore philosophical implications of these new spaces.

FinneGAN generates speech, based on the text of the novel, and pushes the language of Finnegans Wake toward the