Why did you choose to take part in Re:Humanism?
A friend, Carola, sent me the open call as she knew I was working on an AI-driven research & project. To be honest we were really lucky with timings, since I was finishing Adversarial Feelings when I decided to participate…
“Adversarial Feelings”, the title reveals a tight relation between two subjects. Could you tell us a bit more about this link between the capacity of humans to feel and of machines to learn emotions and interactions?
I am interested in employing neural networks as performative tools. I am fascinated by the idea to drive machines to perform human emotions… it is an occasion to observe human behaviors from the outside, or at least to observe what machines perceive as significant about our feelings. Which is sometimes really surprising and uncanny. Especially when working with text and music, Neural Networks offer the possibility to deal with the unconscious in many ways.
Your work is audiovisual, encompassing a range of disciplines that could better define the meaning of your piece of art. Do you think think this has opened new paths to eventually explore with your art?
I think artificial intelligence is an effective tool to melt languages, and to translate data into different combined media. At the moment I’m trying to start from text, to create AI generated narrative artworks, where audiovisual events follow a generated “neural” plot.
Your work deepens the field of latent neural networks. What did you intend to bring up with it?
That of latent space is a really representative concept in the world we live in. A latent space is a multidimensional space (a matrix) where all the possible features of a NN lies. It means that (for instance in the famous CelebGAN I used for the video of ‘3402 Selves’) it’s the space where all the faces the NN could generate lies. Well, this multidimensional space, just like the truth in post-digital age, is LATENT: it Can’t be observed directly, but we can “explore” it through interpolations and other operational functions. And LSs don’t deal with time directly. Every single timebased narration we produce by exploring a certain path within the latent space is a representation of the latent space itself, but it never resolves in any way the whole content, which is by definition non representable.