Temporary House of HomeTemporary House of Home

Tivon Rice's work _Reading Interiors_ brings together images of interiors and five voices generated by artificial intelligence (AI). These voices - the child, the home, the space, the worker and the other - examine our changing relationship with the home in the age of Covid-19. The work is part of the exhibition Temporary House of Home and consists of an installation in Het Nieuwe Instituut and a website where users can upload images of interiors themselves - and then choose a voice to 'read' the room.

A characteristic of Tivon Rice's artistic practice is an acceptance of the contradictions that exist in contemporary digital culture. He belongs to the first generation to have grown up with computers in the classroom. For Rice, playing video games and feeling as though you're in another world would later give way to a feeling of discomfort, caused by the memories of places and events that never really existed.

This resulted in him identifying a collection of opposites or polarities that exist on the border of physical and digital encounters: the direct and the mediated; presence and absence; the experience and being here versus being there in the frame of the screen. Later, Rice obtained a PhD in digital art and experimental media from Washington University, was a Fulbright Scholar (Korea 2012), and became one of the first individuals to collaborate with Google Artists + Machine Intelligence.

Writer Twan Eikelenboom talks to Tivon Rice about his own changing sense of home in the age of Covid-19, his early ideas and choices in the making process of Reading Interiors, and working with AI in artistic practice.

Twan Eikelenboom (TE): The broader theme of the exhibition poses the question, what makes a house a home? How have your own practice, home and feelings about home changed during the Covid-19 pandemic? What insights have you gained in these times that have proved valuable for your work and practice?

Tivon Rice (TR): For a long time, my practice has been concerned with the telematic dimensions of communication. As a digital artist, I remember having very optimistic visions of telepresence, virtual collaboration, and networked performance as far back as 15 to 20 years ago. But when Covid thrust these tools upon our work, our teaching, our learning, and even our socialising, I think we all experienced the shock of having our homes transformed into the primary site for these activities. So, along with the positive aspects of accessibility and collaboration without the need for travel, we have also seen negative impacts from bringing work home virtually, including the expectations of constant availability.

TE: How has your relationship with technology changed in this period? And how do you ensure that, given the greater emphasis on the use of technology, you find a balance?

TR: These times have actually reinforced my need to work physically, in order to balance or counteract my time spent interfacing with technology. For the previous three or four years, my work has been very screen-based, but this year I have returned to many of the sculptural and spatial practices that supported my earlier work. This also forces me out of my house, to bike to my studio, which is attached to my university's maker space. So, this has created a healthy - I think - balance between my work with technology at home, and physical making at my studio.

TE: How does your work for Temporary House of Home fit into your broader work, interests and projects?

TR: My past few projects have sought to create narrative encounters in different environments, specifically using AI generated texts. For example, I explored several large architectural environments through the AI 'voice' of science fiction author J.G. Ballard. I also recently created a series of films focused on environments at the boundaries of nature and human intervention, and these were similarly narrated by a chorus of AI voices. So, the theme of what makes a house a home resonated with my work in the sense that home has become a site of technological intervention in recent years, for better or for worse. Perhaps more interestingly, the home is an archetypal space for stories to be told in, from small personal moments to large life-changing events.

TE: _Reading Interiors_ uses images from the archive of Het Nieuwe Instituut and asks users to share their own images of interiors on the website. How do these two sources relate to each other? Do you see the exhibition and user contributions on the website as different iterations of the work?

TR: I can say that the 250 narratives from the HNI archives are a completed work in the sense that they present a polyphony of voices, and a comparative study of the different forms of language we may use to describe domestic spaces. That being said, the HNI archive images represent a very particular set of interiors, let's say upper-class Dutch architecture from the 20th century.

With this in mind, the other archive, the stories about interiors displayed on the website, stands in contrast to the HNI archive. They present other images of technologically mediated interiors: unsecured surveillance cameras, zoom backgrounds, and wildlife webcams. As these scenes propose other definitions of human and non-human interiors, so too does the website ask the public to contribute images that define their sense of home as it has shifted in Covid times. In this sense, the website dimension of the project is in a constant state of becoming and may even trace our shifting relationships with home as we transition from lockdown and back to normal (or more likely to a new normal that incorporates functions of both pre and post-Covid domesticity).

TE: I'm curious about the methodologies and resources you use to create your work, and how you come up with ideas for a work. Not many others bring AI voices and visuals together the way you do. In Reading Interiors, this combination provides "additional windows to look at our 'private' spaces." How did you arrive at your choice for the five AI voices? What were your first ideas - and which ones didn't make it?

TR: For this project, my very first thoughts focused on the different voices that would 'read' the interiors. I knew I wanted these voices to reflect the different types of language used around the home, especially in its emerging capacity as a site of labour and education, in addition to domesticity. By training AI voices on the works of Karl Marx, Roald Dahl, and Georges Perec, I feel that the voices of the worker, the child, and the home were well-represented from a very early stage.

I also had an idea for the voice of the architecture, trained on an encyclopedia of architectural graphic standards, but that AI spoke in very dry technical terms about the construction of space, so it was cut from the project.

After the failure of the architecture AI, I still wanted some non-fiction voices that reflected different perspectives on space. Through dialogue with Klaas Kuitenbrouwer, we identified a hybrid anthropology/philosophy voice that focuses on social space. 'The space' is a mash-up of texts from Marc Augé, Michel de Certeau, and Michel Foucault. We also formulated the voice of 'the other' which we see as a perspective on biological and non-human space. This is also a hybrid model including texts from Donna Haraway and Jakob von Uexküll.

TE: What are the first practical steps when starting a project?

TR: Before doing any programming or deep machine learning, I feel it is very important to have a strong relationship with my chosen dataset(s). In the case of generating stories that emulate particular authors or genres, the datasets are literally the raw texts of those author's works. So, collecting and 'scrubbing' (or cleaning up the formatting) of these texts is actually a very important step, as it gets me closer to their work. It forces me to either re-read or scan over the work, which shows me the patterns and use of language that I should expect to see the AI recreate once the model is trained.

TE: The voices for Reading Interiors are really distinctive and the structure of the paragraphs is very interesting: a combination of almost sterile descriptions (for example, "A photo of a white sheep"), additional context and story layers ("Just to the side of the picture frame, a shepherdess holds a sheepdog in her arms. Its paws are stretched out, with its chin resting on her right shoulder"). In developing and training the voices, which part is formed by your influence and your determining the source material, and which part is formed by the AI?

TR: The first step in generating a story involves some computer-vision/image-captioning software. This creates the very basic, sterile image descriptions. Then the fine-tuned GPT-2 model creates a longer paragraph about that very basic 'prompt'. That created a very interesting process when showing the same image to all of the voices. They all started with the same prompts but described them in the very different styles of the five models.

Because I tell the system to generate five texts, then choose one to be included in the final project, I've literally read thousands of these short stories. And it has been interesting to be in that position of a reader, and decide what is 'good' or 'strange' or 'accurate' or 'believable.' This becomes a very particular task when reading something that is supposed to be fiction (in the case of Dahl or Perec). There seems to be a freedom to let the voice write a poem or an absurd story. Whereas with Marx or the more philosophical voices, I have to ask the question: does this make sense? Is this an economic or anthropological philosophy that I would believe? I suppose we should ask this about the original texts as well. But at times, I just let the jargon, or the very specific vocabulary of those philosophical texts, tell their story, regardless of the 'veracity' of their statements.

TE: In this creative process, how do you ensure that elements such as image, language, narrative and technology continue to form a coherent whole?

TR: That's a good question, because I believe that some of the stories and image/story pairs are actually quite incoherent. They can ramble on, repeat or contradict themselves, or get lost in the convolutions of language. So it is important for me to step back and look at the bigger project. Thinking about the ways technology has impacted our sense of home over the past 16 months, I feel that the chaos or the absurdity of this dynamic needs to be reflected in Reading Interiors. Ultimately, I wanted to create a system that has the capacity for creative re-imaginings of interiors spaces, and I expected those stories to cover the spectrum from simple, believable narratives, to the feverish hallucinations of the AI voices.

TE: How do you deal with the tension between shaping and controlling the narrative as an author on the one hand, and facilitating the creation of narratives through technology on the other? What role or position do you take as an artist in this co-creation process?

TR: I think there was a tension in my earlier work, when I wanted the AI voice to create perfect, complete narratives on its own. But I've come to the position that AI, in many ways, is a romantic façade. There has been an image built up around AI that it can be completely autonomous and replace the work we do in many fields. Having worked with machine learning in an artistic capacity, I understand that AI does not have creative intelligence, but rather it can present possibilities and provoke ideas for the artist to execute or further develop.

TE: About provoking new ideas - can an AI voice surprise you in an unexpected way, for example?

TR: One interesting and charming surprise that I often see from these AI authors is their gender fluidity and switching of pronouns in the middle of a story. At first, I found this frustrating, that it couldn't keep a character's pronouns consistent, but now I see this as an honest feature of AI logic, that it doesn't really care. It will imagine different possibilities for characters as it goes.

I also appreciate the semantic fluidity that can emerge through these AIs. For example, in one story by the child, the voice of Dahl is speaking about magic and reality, and at some point, it questions "that which is real." After formulating this idea about reality for a while, it comes to the conclusion that "that witch is real," speaking about a woman with magical powers.

TE: What insights from this project can you use in new projects? Which projects are you currently working on - and in which direction do you want to develop your artistic practice?

TR: I enjoyed the process of creating an archive of many AI generated narratives and presenting them in a non-linear manner. The order of the 250 stories will never repeat itself, thus the experience in the institute's gallery will never be the same. This is different from my recent films, which are fixed in duration. I'd like to apply this idea to an upcoming sound-sculpture in which several concrete objects (some containing audio speakers) attempt to describe the architecture of the gallery or the site-specific space of the installation. I like the idea of turning away from screen-based images for this project and letting the objects, the space, and the spoken narratives evoke images in the audience's minds.

TE: In addition to your artistic practice, you also work as a teacher. What themes and projects are you currently working on with students? How does the teaching profession nurture your artistic practice?

TR: I get so many ideas and so much inspiration from my students. I think this is partly because they also become readers when working with AI and natural language processing systems for the first time. I really enjoy witnessing what they find to be exciting, or strange, or provocative in AI-generated text. With this in mind, one of my main research/teaching projects is to make machine learning accessible to non-programmers. What will poets, writers, musicians, or performers do with these tools? What datasets will they train? What questions will they ask of it? So, in my role as an educator, I aim to make tools that simplify the coding process and let students focus on collecting their own datasets and exploring their creative ideas with generative text.

Another dimension of AI accessibility recognises that most large AI language models are English-language, so I have started a research group focused on Creative Resistance to Anglocentric AI. This concern actually began when working with workshop participants in Amsterdam who wanted to create a Dutch-language AI, to speak in the voice of the North Sea. I some ways, I think _all _artistic work with AI is a form of creative resistance to its corporate and institutional implementations, but it is very important to recognise where bias and exclusion exist within widely available machine learning models. Having students from diverse backgrounds, speaking many languages, keeps me acutely aware of, and motivated to challenge, the problematic dimensions of these tools.

Text_: Twan Eikelenboom_