Leeds workshop 1: broken

The progress bar inches forward. It’s 2 A.M. on the night of day two. Since our arrival, we have clocked up over fourteen hours of long nights and early mornings in an attempt to revive the software that we had used to create an AI image-generation model trained with our own images and made as part of a previous project. This was going to be the pivotal technical demonstration of what an image-generation prototype could look like. As we wait, we reflect on the creeping divide between our team and the cohort. Our pursuit of a working model is precluding our presence at mealtimes and in the evenings. We are breaking our own rules. The progress bar freezes at ninety per cent. The computer crashes for the fourteenth time. I fear more than our tech is broken.

The past is ahead of us (field note)

aruma says, ‘Because we can see the past, the past is laid out in front of us. Because the future is obscured, we cannot see into it, it lies behind us’. In English, to say someone is ‘living in the past’ means they are unable to engage with their immediate reality because they have retreated into memories. To be aware of the past, like being aware of the landscape before you, allows you to act with awareness and intelligence.

Commission for the prototype

Our task for the week is to agree on the design of a prototype, encompassing both the technology and the social responsibilities of custom AI creation. The prototype will need to allow users to generate images via a simple user interface in Spanish and Portuguese; train custom models using personal images, such as faces, clothing, nature and artworks; and be accessed remotely via diverse devices and methods.

Our approach to achieving this blends creative facilitation, neuroscience, information design and what we term ‘applied storytelling’. We are first and foremost artists, designers and scientists with backgrounds in human computer interaction, visual art and live improvised performance. Critically, we are still in pursuit of meaning in this frontier of AI. Although our work with AI began over a decade ago, we are equal to our guests in the exploration of sensemaking and knowledge exchange. Our histories are radically different, and the fact that we are three white males presenting the techno-colonialist, extractivist technology that is AI is not lost on us. It is our departure point.

Methodology: to hold a space

Our method of creative facilitation focuses on holding safe spaces for critical entanglement and risk-taking. We have decades of shared experience in designing arts-based research labs. These experiences have taught us to pay attention to the edges, and in the context of AIAI, our edges include:

Neutrality. The main encounter should be held in a neutral space that balances power dynamics across participants, ideally an artist- and/or community-run space with cooking facilities and definitely not university buildings.

In-between spaces. We would expect most of the journey or ‘work’ to happen not in the main meeting space but in the spaces in-between: markets, pubs, on the dance floor or in the kitchen, over a meal.

Building a home. We seek out locations where everyone can sleep, live and eat in the same space, ideally close to the main workshop space. These are more creative residencies than they are workshops with strict timetables and divisions between activities.

Critical eating. We like to find an artist or other creative who can cook and chat with the group, who can act as a mirror for the journey. A well-balanced meal feeds the participants’ capacity for deep engagement, especially across three languages in novel territories.

Embracing uncertainty. We start by not knowing; we aim to fail fast, hard and often; we run towards difficulty. We seek to nurture and hold space for tension in the open.

Reciprocity. We work to hold space for others to speak first. We try to keep introductions light and remain mindful of our cultural position and privilege.

Leeds workshop 2: risk and ritual

Day one progresses at a gentle pace. As a group, we share stories, blessings and objects, contributing them to a communal altar. There is a buzz of excitement and apprehension about the coming week. Day two commences outdoors, despite the cold. We form a circle for a ritual burning of herbs, and we sing songs, expanding space for togetherness amid the mundane backdrop of high-rise flats and the M621 city loop road. In stark contrast, our first formal session unpicks the questionable ethical underpinnings of AI image-generation models. We hold nothing back in sharing the realities of how artificial predictive models of intelligence are each a derivative of forms of public or private extraction without consent, often involving violence. They are built from foundations containing images from the dark corners of the internet, such as abuse of children, adults and peoples. Even the AI image models that purport to be ethical, such as Stable Diffusion, are subversive capitalist malpractices framed as research. The interpreters’ faces are grave as we share the exploitation implicit in the billions of images scraped, the human cost of data labelling (Perrigo 2023) and the lack of desire in the industry to acknowledge or change these practices.

In a week-long lab, the rapid, open navigation of tension can enable early cohort cohesion. Leading with radical openness focused on the AI industry – a provocation that positions every generated image as complicit in these histories – risks disengagement. However, the greater risk would have been to let the technology act as a cover and knowingly co-create a prototype with a deeply unethical backstory without acknowledging this.

Holding your words in my mouth (field note)

In my own work, when I collaborate with someone, I listen carefully to them and try to repeat the meaning of their words back to them. Then they tell me where I am right and wrong. A similar act of verbal repetition appears very clearly in the discussions here because of the three languages, but it also has another function. Aside from the three professional interpreters in the room, many others choose to step in to translate at different moments. It becomes an act of advocacy. I witness a building of trust, understanding and consensus, as the words are repeated from different mouths, and heard again in different languages.

Leeds workshop 3: What do you mean by ‘intelligence’?

‘What do you mean by “intelligence?”’ Loreto asks.²

Christophe³ responds, ‘In the West, we have a very narrow understanding of intelligence. It can only see what it can empirically measure. It has little respect for the intelligence of plants, animals or the intelligence of a community’.

A break is called. When we return, discussion animates the room.

Tadeu postulates, ‘AI is here. We can acknowledge what and who has come before and still create what comes next. When we have the option of a model that is ethical, we can work with it’.

Mariela adds, ‘We need to know what this means for our young people. This journey is in part about how we shape the way this unfolds’.

Our risk in sharing the dubious ethical backstory has started to bring the group together. But people still seem unsure. The energy of our first encounter seems like a distant memory. We embrace the uncertainty.

Listening through my eyes (field note)

When I don’t understand the language, the level of my attention is still the same, but the voices become a kind of melody. I am listening only through my eyes. To compensate, my eyes are going around the group catching gestures, expressions and other responses I would usually miss. This is exhausting, rich and nuanced communication.

Leeds workshop 4: What are your dreams?

On day three, in the light of having shared AI’s dirty backstory with the group, Sebas suggests leading the morning with a session on dreams. One of the academic team members suggests framing dreams around the cohort’s desires for how they might use AI image technology, ‘What do you dream it can do for you and the people you represent?’, so it can act as a before and after form of evaluation.

Sebas fiercely responds, ‘No fucking way. We can’t ask that of people’s dreams. We don’t want to measure a fucking dream’.

The cohort arrives in subdued silence. Sebas sits, eyes closed, poised, ‘I ask you right now, at this very moment, what are your dreams?’

As I contemplate my dream, I reflect on the isolation of ‘crunch culture’, work patterns widespread across the technology and creative industries that value high pressure and burnout productivity over human health and connection. I see this mirrored across academic cultures, although, whereas the ethos in the technology and creative industries is to ‘fail fast, hard and often’, in academia failure is generally ignored or only processed after the fact, buried in methodologies. A realisation suddenly dawns that we are here to speak our failures, not to account for them later. We are here to prioritise our human synergies over technical proficiencies. I write down my dream celebrating my failure. My voice shaking, kneeling on the floor, I read it out loud. It’s something like:

‘I dream of letting go of the technology we are trying to make work for you. Each mealtime and evening, we are attempting to revive a custom AI image model to demonstrate what it can do. Instead, I dream of sitting and eating with you at mealtimes’.

Sebas grabs the paper and reads it aloud. The translations bounce around the room, bodies open, smiles meet my eyes as connection is re-established. We are reconnected with our primary objective, which isn’t simply to build a technical prototype; we’re first and foremost prototyping the immaterial and the intangible – trust, vulnerability and the delicate connections that hold our group together.

A model and a dog

Even though we have failed at reviving our custom image-generation prototype, which would have enabled the group to generate and use their own image models during the course of the week, we know that we can still show people a proxy for this process, based on showing them some of the images that it had generated at earlier stages of the training process, before it crashed. This will suffice as validation. Our aim here is to use a live demo of the image-generation process to critically address matters of ownership, authorship and permission. We need around eight images for model training, and our choice of images is important. Each image needs to feature the same subject, such as a pattern or person, but be taken from different angles and at different shot sizes. The images need to be taken fresh, validated by our shared experience of our week thus far, and they need to feature a neutral subject. This is the process that we refer to as applied storytelling.

We decided to use Thea’s dog, Ed, for this purpose. He had met and made friends with multiple group members on the first evening in the hostel lobby. It was thus that his likenesses were used as a provocation that led to discussion of ethical questions about the responsibilities of model-sharing within the group and about how permission and attribution – in relation to both human and non-human subjects – should be handled with AI-generated images and, in particular, in relation to Indigenous epistemologies.

Prisoners of science (field note)

Tadeu talks about how the sacred objects of his people have been stolen and scattered across the continent and the world. These precious objects are not only ‘from the past’ but still necessary, still working, still ‘switched on’. To know this and know they are hidden from their people seems like a kind of torture. Tadeu also talks about his people, who were taken, murdered and dissected in labs. There are museum archives that contain remains belonging to his people. Because the spirits of Tadeu’s people are not gone, he considers them to be prisoners – prisoners of science.

Leeds city market and the global circulation of images of Indigeneity

There is a downward shift in mood after a visit to Leeds city market. Images on T-shirts and other ephemera show stereotypical, inaccurate depictions of Indigenous people. This not only mirrors the wider challenges of cultural bias inherent in the models; it also demonstrates the lack of agency the group of Indigenous artists and writers has across territories and technologies. To address this imbalance, we need to not only give them complete control of the prototype we are designing; we need to make an art-object that can one day be repatriated.

Indigenous data security

Due to the potentially culturally sensitive nature of the group’s images that will be used to train custom image-generation models and the desire to protect data, our only option for hosting the prototype is to use a physical computer instead of faster cloud-based alternatives, where the location, permissions and wider use of the input and output images are unknown. (The upside of all these decisions is that they also lower the overall carbon footprint of what we’re doing.) Our initial plan is to host this on our studio machine and allow access with a login. However, after the debates around stewardship of cultural property, we realise that the prototype needs to be hosted on a laptop, which will be a kind of art-object. Another positive aspect of hosting the prototype on the laptop is that, if desired, the deletion of the prototype and all its data would be final and physically demonstrable via a video of the computer being smashed to pieces.

I head off to purchase a laptop for this purpose straightway. Ongoing discussions focus on how the group will manage and access their space on the machine and on permissions to share models. We name the prototype IndigenIA. As I arrive with the new machine, Loreto asks to hold it. I open the packaging, and she comes alive at the sight of the dragon logo etched onto the laptop case. ‘The dragon is our friend, it is a good omen’, she says. I respond with stories of the dragon as a powerful symbol of hope, resistance and guardianship; a being of great power, existing outside of good and evil. The machine quickly becomes known as ‘El Dragón/O Dragão/The Dragon’. To demonstrate the model working, we generated and shared a series of images of Ed the Dog meeting a dragon in the wilds of West Yorkshire (Fig 15.1). As part of the learning process, users went on to generate images depicting their model subject meeting a dragon in some way (Fig 15.2).

Final thoughts: AI and Indigenous epistemologies

‘Everything we create with this technology is alive, it has a soul,’ says Mariela, as we prepare to share the final prototype proposal with the group.

This statement has changed my approach to AI to being one of nurture and of care; a slow resistance to the crunch of Western productivity, the hungry data centres and empirical progress at whatever human cost. If we see our AI creations as living beings, then how many would we create? Who would we call their kin? What would be our approach to collaboration?

Notes

1. For more information on Dragon Dreaming, see https://dragondreaming.org/.

2. Loreto Millalén, one of the Indigenous team members. All speech is given in English only, because it is based on notes taken during the workshops rather than audio recordings of the events.

3. Christophe de Bézenac is a neuroscientist and another member of the Immersive Networks collective.