# CROSSLUCID Art

- Channel: [Edge City](https://streameth.org/edge_city)
- Date: 2024-10-19
- Duration: 09:52
- Watch: https://streameth.org/watch/671326fc8f864ede03420ced
- Download: https://vod-cdn.lp-playback.studio/raw/jxf4iblf6wlsyor6526t4tcmtmqa/catalyst-vod-com/hls/e6b8h8wyzp2gh01a/720p0.mp4

## Transcript

We would like to show one use case from what was the training set and how it sort of developed all the way until a gallery use case where there has been a parameter-limited part of a trained alliance that has been made possible to interact with within the gallery and also purchase prints from within the scope of the gallery? So it all started with a photographical book. Our background is in photography, film and architecture. So these were a series of portraits and assemblages that we created over the course of two years. It was a variety of different individuals that we photographed. It was a variety of different individuals that we photographed, and this book became a data set for the current Elias that we are presenting. We have multiple Eliases on the platform, but this one is very specific because we really played with the parameters and we made the parameters very constrained, so the image is always becoming, instead of being very concrete in that sense. This was an iteration that we did, it was 5,000 covers that were created with guns, adversarial networks for a magazine that're traveling all around the world. And this is the actual outcomes from our allies. We have never been human, that you can actually test yourself. It was shown in a gallery context and was opened yesterday for everyone to interact with. And I think we have here a QR code. So you can directly jump into the page and prompt if you would like to interact with the model. Yeah, and explore how that interaction looks like from a visitor or user side, as well as training your own alliance eventually. And what has been very interesting within this scope is that we never really use generative AI to create representations of humans, because there are certain ways of pulling in negative prompts. And then still, there is all this bias that comes out. And this sort of like bias that comes out and this sort of specification through this Laura truly allowed to create representations of humans or anthropomorphic figures that are constantly created as a possibility. It was quite of a challenge to create a model where you can just prompt the word human and you won't get a white person, white blonde person, but actually very often very diverse being that comes to life. And now we jump directly maybe to our work, like very, very quick overview of what we do. So we, as I said before, we are kind of technology agnostic. We work with technology with prototype futures and in our practice always trying to find ways of, new ways of being and seeing the world. This was the collaboration with Ursula Le Guin Estate and Berggruen Institute in which we reacted to her story that's not very visual. We created with LLMs a language of the forest and this was a story that was driven by the language that was created by LLM to create the narrative of the story. So it was basically this very, very specific forest that is within the scope of work of Ursula Le Guin that created language that is promptable back and forth. So sort of it's pretty cohesive on different types of prompts. And we let all of this language, which for us is unreadable, to drive the visual outcomes of the whole story, together with narration from the son of Ursula Le Guin. This was a recent commission for Serpentine Gallery for actually a really incredible future art ecosystem book that talks or compiles different ideas around AI. So if you ever want to read a cohesive body of work around what AI is, around models, and around sustainability of those models, it's a very good address to actually read about it. And there's an online version and a physical version that just came out also. And then this is a project that started off actually from a quite negative experience with text-to-image generation because we have been so much working with StyleGAN and and Gans and sort of like found that found that space very exciting but the moment that we went over to text to image and started to visit these these different possibilities where imagination can take you and of course like putting in different different functions to words suddenly we were catapulted to extremely violent content and we started to deal with this sort of traumatic or extremely violent specifications that exist within those datasets and also the irrepresentability of certain concepts that are not part of those data sets to enough extent that they would resurface. And we utilized some mechanics from actual hypnotherapy protocols, or like approaches where a sort of like traumatic image, image narrative is alterated and pushed over into a more sustainable, more healthy narrative. And we did this on the latent spaces of multiple models with functions to explore the continuum that exists between different meanings that language imposes. And it's a multichannel piece. Also has multiple evolving presentations and changing scenes. And what we are working currently on to create this narrative ecology where people, visitors, can actually plug into that by interactions with LLMs and sharing. Because there seems to be quite a strong potency of LLMs to actually create alternative metaphors for experiences, but still more tied inside of those. So these are some installations view for recent exhibitions that we did. And we now very quickly come to the project we are working on during Edge City. We started the residency in Lisbon, and it's a project around marine protected areas. We learned about them also recently, which means not many people know about them. These are sanctuaries in oceans that don't have borders, which means regeneration can spill over the area that they are assigned to over 600%. The issue is that there are not that many of them, and they are mostly also not very sustainable. So through our project, we are creating actually a physical experience in which the user or the participant of the experience through haptic, liquid haptics, can navigate an environment that's live AI generated around those topics. So it's within the oceanic bed, so to say. And it's an extremely data-rich environment that is also proposing an experience of interaction instead of a direct interaction. So you are having multiple sort of like data inputs, real world data, as well as data from conversations with different agents and LLMs. And it's all of this, and there is an inscribed memory that is re-bringing that environment. So it's sort of like an evolutionary sort of space that is evolving on the go and is based on the different types of interactions that will happen there and also with real-world data of MPAs and the communities living around them. And during Edge, we'll be working on welding, first and foremost, creating almost like a representative that will introduce us to the artwork. But also we already had a meeting with Matt from Radical Exchange, how we can actually steward artwork like that, how we can have multiple actors that are good parents for the artwork to live to evolve and to do the most incredible work it can do beyond ourselves thank you