This story is part of Next Generation, a series in which we give young makers a platform to showcase their work. Would you like to see your work here? Get in touch and plot your coordinates as we navigate our future together.

Sofia Crespo is a Berlin-based digital artist and recent Applied Computer Science graduate. She uses machine learning techniques to create unnatural images of nature. With an interest in the workings of the human brain, Crespo looks at the similarities between techniques of AI image formation, and the way that humans express themselves creatively and cognitively recognize their world. This, combined with her fascination of shapes, forms and colors found in nature, results in a series of awe inspiring images that force our brains to engage with nature in entirely new ways.

Crespo's work brings into question the potential of AI in artistic practice and its ability to reshape our understandings of creativity. Indeed, AI may influence the things we make as much as we influence it. If an artists’ equipment has always been an extension of the artist, where does machine learning stand in this balance? 

We spoke with Crespo to hear more about her project, Neural Zoo.

What first sparked your interest in biology-inspired technologies?

Looking back it seems like a combination of well-aligned things that showed me this was the one thing for me. I had been for quite a while interested in algorithms that model/simulate events in the natural world since there are so many of them with all sorts of complexities (another example would be the Bees Algorithm which is inspired by the foraging behavior of honey bees sending scouts to search for flower patches and nowadays has various applications in Computer Science). Naturally, the more I learned about neural networks and computer vision, the more questions I began asking about their relationship to biology.

How are the images for Neural Zoo made?

Neural Zoo is technically shifting, however, to break down a little more clearly how the illustrated specimens have been generated: each specimen starts with a dataset, a collection of images of a certain species, type, etc. These are then fed into a Convolutional Neural Network. In some other cases, we've created large datasets, around 250,000 images, and fed them into a GAN. This last one is a very powerful technique because it allows us to create interpolations from the latent space inside the model, basically navigate through the representation of the learned features. 

Although there are different techniques, a productive way of thinking of CNNs and GANs is that they take the input (the dataset) and through various, layered processes develop a certain sort of understanding of what the "essence" of a particular dataset is. With Neural Networks this is visual data, what kind of patterns or constellations of pixels create the essence of a jellyfish?

 What does Neural Zoo say about creativity?

Initially, the fact that Neural Networks themselves are abstractly  inspired by the functioning of the visual cortex led me to think about the flow of information in the creative process, i.e. the way as a child I developed a phobia of jellyfish, which eventually became a fascination for the visual elements of the jellyfish and finally the focus of an artwork. I wondered if there's a 'dataset' of human experiences in our brains, that we constantly filter through and rearrange, and that 'rearranging' of the elements into novel ones is what we refer to as creativity. No matter how hard I try, I can't imagine a color that I haven't previously seen, but I can imagine combinations of the ones I have seen, similarly, a neural network can't create something out of the blue from a dataset that hasn't been fed to it but it can recombine elements from data that has been given to it.

How do you see the relationship between nature and technology in your work?

The fact that I'm using technology to learn about 'nature' in the broadest sense of the word, has an eerie quality to it, one I think will probably become less and less questioned by future generations. Maybe we're still in the phase where nature is 'the given world' and technologies are, well, 'the human-made adaptations to that given world', and we still make a distinction between them. Maybe in the future, there won't be any relevant point to be made about them in such a manner. For this reason, I find it interesting to recombine the objects observed that my brain classifies as 'natural' using technologies that are classified as 'artificial'. There's something about synthetic biology that fascinates me.

What can be gained from creating imagery that challenges our perceptions of reality?

I never tire quoting what Vera Molnár says about computer art, because it's so relatable. There's this element of randomness that we're able to explore in-depth since computers can calculate certain things faster than we do, that contributes to the artistic process in such a way  that we're able to know better what we want, what we're trying to visually express, because of using these tools. When I look at the variations that can be produced from a single dataset, sometimes I feel amazed by the fact that I never thought of combining two elements, say a coral and an anemone, in that specific way. As a consequence, the way I look at that coral and anemone is permanently altered, as I have now established a new potential visual combination between them that wasn't there before.

How do you see the future of AI and machine learning techniques in art? Will AI become the new artists?

Even if there were to be an autonomous machine learning artist, it would still have to have been made by a human or a team of humans. It is human beings who consume the art, who feel emotional about images, or sounds no matter if they are generated or handmade by others. Art is as human as our emotions are.

How can we make digital systems more creative? And why should we?

Well, we could first try to define what is considered as 'creative' by looking at the general consensus. Data analysis could be used to try to settle upon a definition or delimitation, and then have machine learning models trained upon that. There's no 'should' here, it's more like do we want to do that, and how or why? That's where the interesting part of data science begins. There are certainly people already working on this, most definitely for business applications, but hopefully for art applications too.

What non-human life form inspires you the most?

Oh... where to begin? It'd have to be marine invertebrates in general, it's so hard to choose just one of them since they're all so wonderful. I recently had the opportunity of becoming an open water diver, getting to dive next to a coral reef has been one of the most touching and educational experiences in my life... oceanic life makes me feel something unexplainable, maybe it's something that got transmitted at home since my father is a sea captain and my mother an environmental law researcher. On the other side, I can't help but feel a strong sense of urgency and concern regarding the oceans, it's important that we do what's necessary to ensure this biodiversity doesn't just vanish from the planet.

What message would you like people to take away from your work?

That the natural world is beautiful and deserves to be taken care of and cherished by us and that technology has 'wholesome' applications, that can be used as tools for self-expression and healing too.

Enjoying this story? Show it to us!

Likes

Share your thoughts and join the technology debate!

Be the first to comment