Project summary

This project would bring together several issues and ideas that we have been investigating and implementing in different areas of the work at Fingerprint Content. In short we are putting into practice as we grow bringing the more-than-human voices and umwelts into popular film and TV, implementing a Protocol of Care across the work streams from agreements to how we interact with AI and moving from that to explore if some of its principles can be made ‘legible interoperable between agents, without giving those agents independent authority’. 

Like so many things that perhaps haven’t been thought through properly in the race to dominance, animals and the more than human are not going to be encountered by AI systems neutrally. The human, largely western coded categories that do not take in the understandings and lived experience of tribal, indigenous and others, such as pet/pest/livestock/wildlife/nature/commodity/resource/sentience that is embodied into a hierarchy of worth and value. These categories are embedded across the cultural and informational material on which models are trained. We want this project to investigate whether, where and under what conditions, those inherited culturally coded categories produce specific, tangible or different reasoning about non-human life. 

The Protocol of Care for More-than-Human + AI would investigate those hidden assumptions and translate the findings into practical interventions for AI development. They would form part of the wider work of the Protocol of Care that we are currently leading the development of. 

This project would combine cultural and narrative research. These would be explored alongside a structured interrogation of AI systems. We would work alongside partners currently working in the arenas of more-than-human rights, animal-welfare and of course technical AI expertise. Together we would identify reoccurring cultural tropes and how the are translated into frames that the systems are trained on. We would then test where they produce materially different reasoning about animals and the more-than-human worlds. From this we would develop an initial addition to the Protocol of Care: a practical set of principles, test cases and language that could inform model constitutions, system prompts/cards and animal-welfare evaluations. 

The current Protocol of Care is doing two things. It is proposing ‘Care Signals’ as structured, non-authoritive records that something may require attention. Secondly it has suggested testing a ‘Deliberative Hold’ that allows for any material unresolved concern to remain, to persist rather than simply disappearing when the next confident instruction arrives. We are interested in explicitly preserving dissent rather than the tendency to manufacture consensus or produce ‘authority washing’. In these systems, care is attention and pause. It allows for the other. Agents increasingly are able to communicate, influence and pass between them their representations of the world with no human oversight, they are in short beginning to tell stories to each other. What sort of stories do we want them to tell? And how do we enable those to be stories that have the possibility of including a pause, to care?  Similarly, the aim of this project is not another statement of AI ethics. As storytellers, as welcomers and celebrants of other ways of seeing and understanding the world, we are paying attention and working to understand where human stories about animals currently enter machine reasoning – and where and how changing those stories could change outcomes. 

What are this project's goals? How will you achieve them?

·  Map: First we would identify the dominant assumptions, hierarchies of value and narrative frames through which selected AI systems encounter different categories of animals and more-than-human life. This would then allow us to trace where familiar categories such as pet, pest, livestock, wildlife, commodity, resource, appear to shape those responses. We would ensure that at as part of this mapping we would include in indigenous, tribal and other knowledge systems from around the world that understand and experience the human and more-than-human relationships differently.  

·  Test: Second we would create two types of test that investigate what happens when we change frames. For the AI systems we would need to design repeatable scenarios to test if different frames produce materially different reasoning, recommendations and solutions. In our work as storytellers, we would set up narrative experiments and audience listening to explore this in reverse: when we change the story do people notice, value or understand about more-than-human lives. 

·  Intervene: Finally we would turn what we lean into things that can be used. This should be split into developing and testing an initial More-than-Human Protocol of Care as part of the current experimentation work. This would include the principles, the language, prompts, test cases and evaluation criteria for AI developers and, crucially, evaluators. Secondly, we would develop narrative tools and create a series of provocations for storytellers seeking to bring the more-than-human into popular culture, building and strengthening the work we are doing with James Graham, Barbara Kingsolver and Amitav Ghosh. We would then test the interventions against the original test cases. What changes and where? What doesn’t? What and where are the interactive frames hardest to shift? 

The experiment as a pitch in an elevator with the unholy love child of Spielberg and Altman?  Can we identify an inherited cultural frame → demonstrate that it has consequences → introduce a different way of attending to the more-than-human → demonstrate whether reasoning, perception or response changes?

https://github.com/JessHines360/protocol-of-care-for-agents

How will this funding be used?

We are seeking approximately $110,000 for an initial five-to-six-month research and development phase.

We don’t want to treat this as either a cultural research exercise or a technical AI evaluation. They must work and function collaboratively for this to have meaningful real-world impact. This funding would enable us to build a small interdisciplinary team from across the work that we are doing. It would support the project leadership, the cultural/narrative research; a technical AI researcher, an evaluation partner, expertise in the nascent more-than-human areas of research; general research assistance and then into the structured testing across existing AI systems followed by a small programme of expert interviews, workshops and narrative experiments. 

As explained this work moves through three stages: mapping the assumptions and narrative frames that represent different categories of animals and more-than-human life and how the are experienced, and interacted with; testing what happens when those frames change in both human attention and response and then more widely in machine reasoning. We will then translate these into interventions that can be used in both AI development and storytelling.  

Like all good storytelling the outputs will be tailored to the audiences. The principle audience will get the Protocol of Care for More Than Human + AI that will include test-cases, candidate language for model constitutions or system instructions and recommendations for animal-welfare evaluation. The second audience is story practitioners and cultural commissioners. These would consist of created and narrative provocations and would explore how more-than-human lives and their umwelt and voice can be encountered and integrated into popular culture.  The first version of the protocol will be made openly available and free so that the work can be used, challenged, tested, and developed by others – rather than dying a slow digital death as a research report.  

We are deliberately proposing this as a substantial first phase rather than a small proof of concept because the real world development of AI and its wide ranging impact on that real world is happening considerably faster than the traditional ‘small proof of concept’ ‘scaling and developing’ framework. We need to do this fast and we need to get people working across silos and in as interdisciplinary fashion as possible. The central question is not simple whether inherited cultural assumptions about animals can be identified in AI systems, it is if by changing the frame we care able to change what becomes visible, valuable or possible to machine learning and its subsequent framing of what should be recommended or is possible but also in the stories that humans tell and receive and therefore frame all our futures. 

If the work does not demonstrate meaningful change or leverage, we would stop. If it does, Phase One and the existence of the Protocol of Care would leave us with evidence, collaborators and a tested framework from which a larger programme could grow.

Who is on your team? What's your track record on similar projects?

The project will be led by Jess Hines, founder of Fingerprint Content and originator of the Protocol of Care, working with a small interdisciplinary team spanning cultural and narrative research, AI research and evaluation, animal welfare and sentience, more-than-human thinking and responsible technology.

We have already identified potential collaborators and advisers within my existing professional network with senior and internationally recognised expertise across AI research, multi-agent systems and cooperation, technology governance, animal welfare and sentience, more-than-human thinking, cultural systems and narrative. Several have already engaged with the thinking behind the Protocol of Care or closely adjacent work.

The project is still being scoped and we do not want to name individuals we have spoken to before they have formally agreed to participate. If the proposal progresses, we would be very happy to share proposed names, roles and relevant credentials with the Falcon Fund before an award is made.

The core team would bring together Fingerprint's cultural and narrative practice, responsible for project leadership, the research hypothesis, cultural analysis, storytelling experiments and development of the Protocol; technical AI research and evaluation expertise, responsible for experimental design, model interrogation and establishing whether observed effects are meaningful rather than superficial differences in language; and animal-welfare and sentience expertise, responsible for identifying consequential cases, grounding the work in current evidence and challenging our assumptions about what meaningful outcomes might look like. We would also draw on a small group of advisers across more-than-human thought, different cultural and knowledge systems, AI governance and, where appropriate, qualitative or audience research.

Jess Hines own track record spans over three decades and many continents and sits at the intersection of storytelling, culture, research and systemic change. Fingerprint develops film and television for mainstream audiences internationally, while also working with foundations, researchers and cultural organisations on the role stories play in shaping how people encounter complex social and environmental questions.

Fingerprint's current film and television slate is already exploring many of the questions that have led to this research. Flight Behavior, based on Barbara Kingsolver's novel, explores climate change through a story of self-actualisation in a poor white farming community in the Appalachians, and the collision between human lives and the more-than-human world. Gun Island, based on Amitav Ghosh's novel, is an uncanny mystery story that moves through climate, modern and ancient migration, myth, folk legends, animals and different ways of understanding the relationship of humans to the rest of the living world. The Hum, being developed with James Graham, is about the men who created the world of lies, doubt and denial we live in today. It is about the way power manipulates people through stories — and increasingly through AI. These are mainstream film and television projects, not research exercises: our work is to get these questions into popular culture and make them compelling enough that people want to watch them.

Alongside that slate, we have been developing the Protocol of Care, including its recent experimental extension into AI-agent interaction. For us these strands are connected. We are interested in the stories humans tell about the world, what those stories allow us to see or fail to see, and what happens as the assumptions embedded within them begin to travel through AI systems as well as through culture. That work is being developed publicly on GitHub, with an experimental protocol, Care Signals, test cases and an initial research programme, and is beginning to be challenged and developed through conversations with researchers and people working in multi-agent systems, AI governance and standards.

The trajectory matters to this proposal. 

The More-than-Human + AI project is not Fingerprint moving away from storytelling into technology. It is us refusing to be siloed into any one area of thought, and understanding the power that there is in understanding and acting on: if not us then who? To stand behind our convictions and following questions already present in our work where they take us. Of being sure that our role as storytellers is not bound by one format or by one set of skills. Just as we in film production we would bring in experts to run different departments, areas that we understand but cannot do ourselves such as cinematography or costume or lighting we bring in expertise and delight in the new ideas brought to us by those who know more about   It is us following a question already present throughout our work: what stories and inherited assumptions shape what we notice, value and act upon — and what happens when those stories begin to travel through machines as well as people?

This project therefore does not depend on Fingerprint claiming expertise we do not have. Our role is the thing we do know how to do: investigate culture and narrative, make and test different stories and frames, connect ideas across disciplines, and translate them into work capable of reaching the wider world. Technical AI, animal-welfare, sentience and other specialist expertise will be built into the project precisely so that those propositions can be challenged and tested seriously rather than merely asserted.

What are the most likely causes and outcomes if this project fails?

One of the things that we’ve noticed working in cultural campaigning, storytelling, film making and other areas of culture is that funders want culture to do everything but fear it does nothing. It is not as easily measurable as say, drilling wells = clean water for villagers or shifting buses from diesel to electric = cutting carbon emissions in an area. And yet we know that storytelling underpins cultural constructs, and cultural constructs underpin how we behave and what futures we build. This project has an element of the measurable and an element. So, the greatest risk is not that the research disproves our starting assumptions; that is part of what the research is for. The greater risk is that we identify interesting cultural patterns without being able to demonstrate that they have meaningful consequences either in how AI systems reason about animals and more-than-human life and what human audiences notice. I think that this will most likely be a risk from our complicated, ornery human audiences and not so much from the AI systems – but I could well be wrong. 

The risk is one we understand: the relationship between culture, story and machine reasoning is much messier than we are framing here. In fact, we would be disappointed if it wasn’t. Another problem is that animals are not a coherent category, especially across cultures and knowledge systems, and also adding in the new but necessary layer of the more-than-human to the already complicated mishmash of conflicting categories such as farmed animals, companion animals, wildlife – all of whom occupy radically different positions within human culture, economics and language. Rather than forcing those differences into a universal framework, we would treat them as findings. The result might be a more contextual or plural Protocol than the one we imagine at the outset.

A further risk is translation. We may identify something important but fail to turn it into interventions that are useful to storytellers, AI researchers, developers or evaluators. That is why technical, animal-welfare and cultural expertise will be involved throughout the project, and why we will test the interventions rather than simply publish principles at the end.

For us, failure would be producing an elegant framework that nobody can demonstrate makes a difference. 

Success does not require proving our original assumptions right. It requires learning whether changing the stories and frames through which humans and machines encounter more-than-human life can change what is noticed, how it is reasoned about, and ultimately what becomes possible. If the evidence tells us there is no meaningful leverage there, we stop. If it shows us that the leverage lies somewhere different from where we expected, we follow it.

How much money have you raised in the last 12 months, and from where?

The Protocol of Care for More-than-Human + AI is a new research project and has not previously raised external funding.

This proposal grows from the wider Protocol of Care which, to date, Fingerprint Content has been developing and funding through its own work. This began as a practical framework for our film, television, cultural work as well as our internal organisational functioning. It addresses power, consent, extraction, authorship, ecology and the more-than-human. This has recently expanded into an experimental Protocol of Care for Agents. This work is now in public development through GitHub where we are working with experts to test how care-relevant concerns such as uncertainty, dissent, provenance and unresolved harm might remain visible between AI agents, whilst not acquiring any authority of their own. 

Fingerprint Content is an independent film, television and cultural strategy company with experience raising, structuring and working with project finance internationally across commercial production, philanthropy and public-purpose partnerships. As such our wider slate includes projects at very different scales. One current feature film, which explores the more-than-human voice within popular culture, is presently closing financing at approximately $20 million. Other work has brought together philanthropic, cultural and commercial partners around climate, storytelling and public engagement.

The Falcon Fund would be the first external funding that is raised specifically for the Protocol of Care and its more-than-human AI research. We would continue to develop the wider questions internally, but it would allow us to take this work that we have incubated and create a properly resourced interdisciplinary development project around these key issues.