Resisting, Refusing, Reclaiming, Reimagining: Charting Challenges to Narratives of AI Inevitability

Resisting, Refusing, Reclaiming, Reimagining: Charting Challenges to Narratives of AI Inevitability Authors: Tania Duarte *1, Ismael Kherroubi Garcia 1,2, Ramla Anshur 1,3, Harriet Humfress 1,4, Dylan Orchard 1,5, Steph Wright 1 *Corresponding author contact: [email protected] Affiliations: 1We and AI ror.org/021k7qh53; 2Kairoi; 3Royal College of Art; 4University of Oxford; 5Goldsmiths, University of London How to cite this paper: Duarte, T, Kherroubi Garcia, I, Anshur, R, Humfress, H, Orchard, D, Wright, S (2025) Resisting, Refusing, Reclaiming, Reimagining: Charting Challenges to Narratives of AI Inevitability, We and AI, DOI: 10.5281/zenodo.17343830 Abstract Increasingly, artificial intelligence (AI) adoption is widely framed as inevitable; a destiny which cannot be avoided. In the context of AI, “inevitability” is often used to argue for the need to adopt AI solutions, or to adapt and prepare for speculative AI developments. With soaring investments into AI technologies and infrastructure, unabating coverage of AI stories in the media, and growing enthusiasm from governments, it is unsurprising that AI is widely perceived as inevitable. However, this perception is shaped by narratives that diminish our agency as humans, as they present technology as advancing independently from human activity and interests. In this paper, we outline instances of individuals and collectives actively defying narratives about the inevitability of AI. We propose a taxonomy of such efforts, which are instances of Resisting, Refusing, Reimagining and/or Reclaiming AI. The taxonomy serves as a conceptual framework for those challenging the power dynamics underpinning AI technologies, and illustrates that there are many ways individuals and collectives can influence how technology is developed and used. The taxonomy is also a call to unity for a collective effort; a plurality of resistances which can define and build a future on terms not dictated by techno-determinist ideologies and agendas. Introduction Artificial intelligence (AI) technologies and discourse are everywhere. On our screens –big and small–, in adverts, on social media, in press conferences, in national strategies, in schools, on our phones. No wonder that AI proliferation and reliance are commonly seen as "inevitable” (Cross, 2025), “inexorable” (Allen, 2024) or “immutable” (Mattner et al., 2025). Hype around inevitable “AGI” (Altman, 2025; van Rooij & Guest, 2024) –which is often seen as a step We and AI Ltd is a Non-profit Company Limited by Guarantee Company No. 13376771.I [email protected] I www.weandai.org © 2025 by author(s) and We and AI Ltd This work is licensed under the Creative Commons Attribution International License DOI: 10.5281/zenodo.17343830 towards “ASI” (Mucci & Stryker, 2023)– fuels a narrative that “AI” will inevitably evolve into something “more intelligent than humans” (Taylor, 2023). Furthermore, when it does so, it will inevitably attempt to take control unless we stop it (FLI, 2023). In either case, whether we know how it will evolve or not, the only possible path forward is to adapt and adopt AI (Ragy, 2024). This is the inevitability narrative: that AI is a solution to so many of our problems that it must –and will– infiltrate each and every aspect of our lives; no matter what we do. But it isn’t that simple; the conditions of inevitability –of being “certain to happen” and impossible to prevent (McIntosh, 2013)– are far from being met. Firstly, “AI” is a contested term. Coined in 1955 to invite interest and funding (Lighthill et al., 1973) for a new field of research (McCarthy et al., 2006), “AI” now generally refers to a range of disparate technologies. Ask for a description of AI and you might hear different ideas: a “set of powerful technologies of centralisation and control” (Neff, 2025), a marketing term (Aten, 2023; ASA, 2024), even an ideology (Lanier, 2020). For the purpose of this text, we define “AI” as computer systems that use machine learning algorithms or sets of rules to identify patterns in large datasets. The pattern recognition can then result in predictions, content-generation, or recommendations (DSIT, 2025). These capabilities are present in diverse tools, such as large language models for the generation of texts or images, predictive algorithms for the identification of genetic mutations, and in computer applications for facial recognition and autonomous vehicles. Secondly, AI involves countless stakeholders, from investors and developers to policy-makers and adopters (RAIN, 2025a). The proliferation of AI technologies means that AI is present in many social contexts, and interpreted by many publics. Generally, we can speak of tensions between groups of actors: those investing, developing and deploying AI, and those using or being subjected to the use of AI technologies. On the one hand, we find AI developers, which tend to be big tech companies (such as Google, IBM, Microsoft and Amazon) and the AI-focused companies they heavily invest in (such as OpenAI and Anthropic); AI investors such as venture capital firms; and AI deployers, which are other organisations deciding that AI tools and systems will help them and their beneficiaries or customers. On the other hand, AI users are people who use AI tools, either because an AI deployer made it part of their offering or because we use the tools of AI developers directly. We may also be more or less aware of our use of AI tools. For example, our use is more obvious when prompting an AI chatbot to reword a sentence (knowing it’s an AI tool), than using a web browser (which less obviously ranks results algorithmically). Alongside AI users stand AI subjects, who may be any individual or group of people whose lives are shaped –witting or unwittingly– by AI. Impacts on AI subjects include algorithmically determined decisions about social benefits (Amnesty International, 2024), bank loans (Holli, 2023), or court cases (Yanchur, 2024), as well as harms caused by AI use, such as being the subject of a deepfake. Thus, AI becomes a sort of entangled mesh between the different actors; it becomes a medium for the reinforcement of power relations. AI developers can continue to excite investors, deployers and users with their latest innovations; users and subjects become locked into AI technologies that proliferate because they are imposed by deployers; and tech startups can be forced to make decisions about how they develop AI in order to meet the demands of investors. AI is used by developers and deployers to influence and dictate the lives of AI users and subjects. At their best, AI tools are developed to help people navigate the world and their lives, and make good decisions more quickly. At their worst, AI tools are weaponised by their investors and developers –or by nefarious actors (Sahota, 2024)– to spread worldviews that further their own agendas. Thirdly, as discourse about AI has entered the mainstream, diverse publics have learned of the problems caused by deploying and adopting AI tools with little regard to how they actually work. Among other things, there is now greater awareness about generative AI tools being built on copyrighted materials (McMahon & Vallance, 2025), requiring disproportionate amounts of energy and water (Zewe, 2025), and causing mental health disorders (Laestadius et al., 2024; Hill, 2025). These three challenges to the conditions required for AI adoption to be ‘certain to happen and unable to prevent’, all feed into the emergence of a “plurality of resistances,” as Foucault (1978) proposes and as Furze (2025) reminds us. The contested nature of the term “AI” means that we don’t even know what it is that we think is inevitable. The entangled mesh of power dynamics and dependencies that AI forms between actors means that all actors need to maintain the system in order for it to continue without being disrupted; and also means that threads from the mesh can be pulled to unravel it. Finally, the growing awareness about the limitations of AI technologies and research 2 Duarte et al. (2025) practices means that there are growing challenges to the current trajectory and practices of AI use. For example, some responses to AI come from companies seeking to protect their proprietary work from staff misusing AI tools (Kherroubi Garcia, 2023), whilst other responses come from creatives seeking to protect their livelihoods and intellectual property rights from companies building AI tools on their work without permission nor compensation (Milmo, 2025). With so many actors, forms, locations and limitations of AI, it is natural that there are many different types of responses, and a multitude of tactics employed to critique and challenge occurrences of AI, as Reynolds states “Resistance is not a monolith” (Library of Babel Group, 2025). There is an array of different activities and movements from the individual to the collective, from the unconscious to the organised. They are based on different implicit or explicit strategies or theories of change, or just a gut reaction, and could be categorised in a myriad of ways. In this paper, we suggest a taxonomy of responses to inevitability narratives about AI: Resistance, Refusal, Reclaiming and Reimagining. (N.b.: we use the terms with an upper case “R” where the category of responses is being referred to, and will otherwise use a lower case “r.”) The responses that different publics have devised before AI are what we capture in this paper, which results from a collaboration among the We and AI community. In early 2025, We and AI community members were invited to suggest case studies –real or imagined– for each proposed loose category of responses to the inevitability narrative. The result was 29 case studies from the We and AI community and engagement with wider audiences, such as the ACM 2025 FAccT conference (ACM, 2025). The project was also supported by conversations held with members of the Possible Futures Discord Community, and the Critical Infrastructure Lab. The responses we have found and imagined are varied. Like Foucault’s “plurality of resistances,” responses to AI may be “possible, necessary, improbable; others [...] spontaneous, savage, solitary, concerted, rampant, or violent; still others [...] quick to compromise, interested, or sacrificial” (Foucault, 1978). Figure 1. Four Ways to Counter Narratives of AI Inevitability. It is worth noting that the four responses to the inevitability narrative are by no means perfectly delineated. Rather, each item can overlap with others in unexpected ways. With this, the present taxonomy is just one way to categorise diverse actions before the inevitability narrative. What’s more, the way the four responses intersect invites community and grassroots collaboration; no effort needs to be siloed or competing. Ultimately, the taxonomy serves as a 3 DOI: 10.5281/zenodo.17343830 blueprint for collective action. Resisting, Refusing, Reclaiming and Reimagining are all actions that reestablish the agency of AI users and subjects in the complex mesh of social relations that otherwise pit them against powerful actors who control both AI narratives and innovations. In giving visibility to different ways individuals and communities are pushing back against the symptoms and harms of “turbocharged” AI investment (DSIT, 2025), it is possible to identify and connect with resonant approaches different situations, even by the same actors. Resisting AI Rejecting the ideologies of extraction, datafication and automation Resisting AI as a specific approach is a resistance not just of technologies, or even their design and applications, but of what is seen as the role of AI technology as a pivotal method of extraction and control by powerful systems, organisations and people. There is a range of beliefs as to how intrinsically harmful fundamental concepts of AI such as datafication and automation are, and as to whether they can ever be redeemed for public good rather than the economic good of the most wealthy and powerful. However, what unites this view is the belief that the current systems of production, and the agendas driving the adoption of AI are inherently predicated on increasing precarity and inequity for those already marginalised. Therefore making individual use cases of AI more equitable, or even finding use cases for AI as a means to increase equity, can do nothing to address the fundamental damage being inflicted by an ideology based on extraction of planet and people. As such, the embedding of AI in not some but all of our processes, decision-making, tools, services, education, art, culture, relationships, must be resisted in principle as a fundamentally oppressive, destructive, dehumanising or fascist force. Understanding how AI reinforces systemic inequality starts by locating AI technologies, research and narratives in a complex tapestry of social institutions and power dynamics. It is about understanding the systems and incentives that give rise to interest in AI, as well as hype (Duarte et al., 2024), undue fear or excitement, which impact on policy and economic decisions. It is about articulating the relationship between people and AI within their shared social structures. Having identified the paradigms of power and control enacted through AI expansion, Resistance strategies respond to deceptive, misleading or unrealistic narratives promoted by AI innovators and the powerful institutions that support their narratives. Many AI innovators have claimed without evidence that AI has inexplicable “emergent” properties that surpass our understanding (Xiang, 2023). Focusing on the artefacts of technology as the sources of innovation allows big tech companies to diminish the human labour involved in developing AI tools, from mining metals for superconductors and electronics (Jung, 2023), to labelling data for training machine learning (ML) models (Perrigo, 2023). The narratives we resist are those that disregard the role and exploitation of humans in building AI tools (Bender, 2024). Resistance strategies also view AI developments as an evolution –a new iteration– of harmful ideologies McQuillan, 2022). In this sense, AI builds on ideologies that condone sexism, racism, ableism and other such injustices. For example, AI resistance may be motivated by AI’s origins in statistics, a domain historically borne out of eugenics, a pseudo-science whereby people are scored and ranked by their appearance and characteristics, which are deemed more or less valuable for future populations (McIntosh, 2013). At their worst, eugenicists have inspired and perpetrated genocides. More subtly, eugenics has morphed into an ideology that pervades many discussions about technological innovation. AI Resistance strategies are grounded in facts about our social worlds and our convoluted histories. They often serve as a foundation for further action against the inevitability of AI. In this sense, many research initiatives and journalistic investigations are part of building and embodying Resistance strategies. Some well-known endeavours that have helped uncover the relationship between those affected by AI technologies and those peddling them are Bender, Gebru et al.’s research into the environmental costs of bigger and bigger ML models (Bender et al., 2021); and Perrigo’s uncovering of questionable employment practices by a data labelling company in Kenya making AI models 4 Duarte et al. (2025) possible for big tech companies in Silicon Valley. Conversely, activist organisations have emerged that resist extractive AI, such as the Data Workers Inquiry, and the AI Now Institute. One example of a Resistance approach can be seen in the the 2022 book Resisting AI: An anti-fascist approach to artificial intelligence by Dan McQuillan, senior lecturer in critical AI at Goldsmiths, University of London. The book provides both an introduction to the technical aspects of AI technologies, and an incisive analysis of the relationship between AI’s technical aspects and their socially determined implications. More recently, McQuillan has suggested “decomputing” as the path towards a future that prioritises social justice over technological advancement (McQuillan, 2025). As Resisting AI illustrates, McQuillan’s thinking is rooted in a deep understanding of the power struggles between those pushing for AI and diverse publics. The idea of decomputing builds on that of “degrowth,” which is a social movement challenging the capitalist drive for economic growth. The alternative proposed by degrowth activists is a world where production and consumption are reduced so that society may operate within planetary boundaries, where natural systems are protected and the negative effects of climate change mitigated. Following from the degrowth movement, decomputing is motivated by the wasteful amount of energy and water consumption required to develop and maintain AI technologies. What’s more, with the advent of mainstream generative AI tools, we have seen an unprecedented growth in the number of data centres around the world. Data centres are enormous warehouses with the hardware necessary to maintain a great part of the flow of information through the internet. As data centres are also crucial for the operation of AI technologies, the data centre industry is expected to more-than-double by 2032, both out of perceived necessity and a valuable financial investment (Alaamer, 2025). At present, data centres are estimated to account for 1-1.5% of the world’s energy consumption. Their proliferation is therefore expected to make that value rise sharply. This stands in stark contrast to the call for economic activity to fall within planetary boundaries, as increased energy consumption will only accelerate climate change. Ultimately, decomputing arises as a response to the pernicious impacts of financial and economic growth on the natural environment. But the spirit of decomputing goes beyond resisting AI. Decomputing also invites us to imagine new forms of technology. In this respect, decomputing points to the intersectional feature of the four challenges to narratives about AI. Indeed, the call to imagine new technologies overlaps with the challenge of Reimagining, which is introduced later in this paper. We can see how decomputing sits across both Resisting and Reimagining in the following excerpt: “Decomputing is the prefigurative decoupling of advanced computation from social goals. It's the reassertion of the need for convivial tools and the construction of forms of collective social power that can bring them into being.” Resistance can take the form of artistic expression, which serves as a powerful means of countering the dehumanising and anonymising nature of AI systems. After all, artistic expression recentres the creative spirit of humanity, unleashing us from the controlling assumptions embedded in AI technologies. One example of this is “Error 406 [Tech Facism] Not Acceptable,” a call for art to explore “powerful forms of resistance, refusal, and subversion: from instructional approaches and how-tos to interventions and collective practices that push back against tech fascism” (Error 417, 2025). Refusing AI A Targeted Opposition to AI Applications Refusal of AI can range from the profoundly subtle to the loudly overt, from everyday practices which embed themselves into our behaviours and identities without adhering to any clear-cut political or activist form to organised efforts to alter policy, perception and material ends. 5 DOI: 10.5281/zenodo.17343830 The motives for refusal can be just as varied and, at times, oblique. As AI is mobilised in more and more areas of our lives, the desire to reject certain mobilisations or logics becomes ever more entangled with wider social, cultural and political landscapes. For example, the refusal to co-operate with the extraction of our personal data can take myriad forms, often more easily framed as personal choice, or even failing, than intentional refusal. We can ‘switch off’ from contributing to social media because we ourselves feel obsessed with it, rather than because it seeks to constantly extract from us. Yet, in that desire to deny representations (no matter how vague) of our selves to be harvested for the ends of AI training, we still undertake an act of refusal, even if we remain unconscious of it. The act of refusal can even appear merely tangential to the question of critical engagements with AI. For example, contemporary ‘Butler lies’ (Hancock et al., 2009) –the act of lying to disengage from digital communication– are, at first sight, framed by the ‘always on’ nature of modern work and social media engagement. Declaring that messages were missed, or that poor connection meant you were unable to respond, can draw out agile evasions as we discover the red lines of our own desires for privacy and peace (Mannell, 2017). Within the context of a world of hyper-extractive data mining however, the innovation of social evasions breeds behaviours that strike deeply at AI’s most central drive – the need to surveil, observe and consume ever more information. It can also be far more direct. The 2023 Writer’s Guild of America strike was heavily based on worker’s refusal to cede creative space and agency to AI mobilisations, something loudly declared on the picket lines (Merchant, 2023). However, from the thoroughly disorganised through to formal union organising, the same desire to refuse can be recognised, finding spaces of agency where it can undertake the everyday tactics of refusal that manifest it. Another example might be deemed an instantiation of Resisting AI. In a city in Montana in the U.S., residents have successfully objected to the plans for a Data Centre in their local area. To make the case for refusing such a data centre, they cited the “unique and massive power and water demands that they impose on national infrastructures” (O’Donovan, 2024). This is a relational aspect of AI that places data centres within complex infrastructural and environmental ecosystems; the refusal results from an understanding of data centres as having tangible impacts on both a city and the wider natural environment. Between the extremes of subtle everyday practices and organised campaigns, there are a vast range of eclectic, contextual and unique tactics for refusal. A potential which places most people, consciously or unconsciously, within a space of refusal at some point. Even those who aren’t actively engaged with the discussion of technology at all, can find themselves looking for opportunities for refusal. Amidst myriad assertions of the artificial and algorithmic, there will always be that which crosses our own personal ‘red lines’ and –even where we embrace some implementations– our desire to refuse will still be activated elsewhere. Understanding the right to, potentials for and necessity of saying ‘No’ is a key resource to enable this refusal. One instance where we can see that clearly manifesting in relation to AI mobilisations is in the emerging tool of ‘tarpits’ (Belanger, 2025) designed to draw AI webcrawlers into a labyrinthine abyss from which they can’t escape, using websites to stop their extractivist activities. Here, we see the meeting point of (i) a refusal to be extracted from, commodified or treated as an inert resource for AI training, and (ii) the contextual skills of coders finding space to deny that access. Specialised in nature –most don’t have the tech savvy to create such tools– it still speaks to an agility of refusal. Refusal is not a prescriptive process; it doesn’t come from a central archive of tactics or ideas. As with Butler Lies, covering yourself for facial recognition, turning off cookies, lying on forms, refusing loyalty cards and any number of other practices, it’s a matter of finding out where your individual and communal capacities to refuse lie and, from there, asserting where your red lines of ‘No’ are to be drawn. Within that subjective agility of refusal, there also lay numerous intersections with other forms of resistance. There is no neat delineation, as refusal can easily segway into something like Reimagining, with the desire to say no triggering a need to create an alternative. To avoid algorithmic cultural management, for example, it may be useful to create new models of cultural networking. Rather than watching what YouTube wants you to watch, what new networks of human recommendation can be found? Here also emerges the potential for both individual and collective actions. While refusal may begin as a personal expression, it may by necessity extend to communal actions. If the identity and craft of the artist are threatened, then the subtle sense of opposition can easily be transferred to a collaborative and 6 Duarte et al. (2025) overt cultural front. The desires of the ‘I’ to refuse can, depending on context, be best expressed by the actions of the ‘We’. For individuals, refusing AI may mean shaping the image algorithms build of them by deidentifying themselves online (Fabrino Mendonça et al., 2023), wearing clothes that hide one’s presence from facial recognition-enabled cameras (Cole, 2019), or resolving to not use AI tools in your work (Denial, 2025). For organisations, refusing AI may take the form of policies formally banning the use of certain tools Cahal, 2023). Refusal may also take the form of organised, collective efforts, including worker walk-outs (Von Struensee, 2021), discipline-specific consortia (Sano-Franchini et al., 2024), and sector-wide recommendations (Dusseau, 2024). Reclaiming AI Giving communities control of technology design, development and use Reclaiming AI acknowledges the concentration of wealth and power that results from how AI has developed and been governed. Reclaiming AI confronts the voices of many people, collectives and institutions before the few voices that constitute elitist groups of big tech leaders. Reclaiming AI strategies include restoring democratic and pluralistic values in decisions and approaches to AI without privileging “AI expertise” (Gourlet et al., 2024) and valuing publics’ perspectives on questions of AI governance (Attard-Frost et al., 2025). In its aim to remove power from big tech, reclaiming AI may strive to distribute decision-making powers to different parties, from scientific communities (Kherroubi Garcia, 2025) to civil society (RAIN, 2025b). Reclaiming AI means going beyond critique. It’s not just about resisting extractive or harmful systems, it is about taking the tools, infrastructures, and techniques already in circulation and using them differently. Around the world, communities are reworking the building blocks of AI –data, models, licenses and governance structures– to serve their own languages, values, and priorities. This is not inclusion on someone else’s terms; it is a shift in ownership, direction, and purpose. At its core, reclaiming AI acknowledges the concentration of power and wealth that has resulted from how AI has developed and been governed to date. It pushes back against the dominance of a narrow, self-reinforcing elite, the handful of big tech firms and high-profile leaders who shape both the tools and the narrative. Instead, it centres the many voices of communities, activists, researchers, and cultural workers who have long called for more just, democratic, and diverse visions of technology. Reclaiming AI is not only about building different tools; it is about redistributing power. It is a challenge to the assumption that only those with technical expertise should shape the future of AI. It seeks to restore democratic legitimacy and public agency in decisions about how AI is designed, governed, and deployed and to value community knowledge, lived experience, and collective ethics as much as engineering skill. In doing so, it opens space for shared decision-making across scientific communities, civil society, Indigenous nations, local governments, and other publics whose futures are shaped by these technologies. A powerful example of this is the Te Reo Irirangi o Te Hiku o Te Ika (Te Hiku Media) Papa Reo project in New Zealand. When Māori language data was being scraped by large companies to train commercial models, Te Hiku chose a different path: they built their own Māori language corpus and trained their own models, using data collected by and for Māori. Crucially, they retained full control through a culturally grounded license based on guardianship instead of ownership, designed to protect Indigenous data, the Kaitiakitanga License, which explicitly limits access and use to Māori communities and purposes (Coffey, 2021). This act of technological sovereignty is not an anomaly. Around the world, communities are reclaiming AI as a site of self-determination and care. For example, let us look at Masakhane, a grassroots organisation whose mission is to strengthen and spur NLP research in African languages, for Africans, by Africans. Masakhane roughly translates to 7 DOI: 10.5281/zenodo.17343830 “We build together” in isiZulu and their goal is for Africans to shape and own these technological advances towards human dignity, well-being and equity, through inclusive community building, open participatory research and multi-disciplinarity (Masakhane, n.d.). In Brazil and across Latin America, feminist tech collectives like Coding Rights challenge surveillance capitalism through public storytelling, counter-design, and creative infrastructure, centring transparency, care, and resistance. Their project Chupadados –the “Datasucker”– tells the tale of our abusive relationship with surveillance technologies (Coding Rights, n.d.) In Hawaii, Indigenous technologists are producing design principles and protocols for AI rooted in relational worldviews, arguing that machines must be accountable to land, community, and kin, not profit. The researchers propose a starting place for those who want to design and create AI from an ethical position that centres Indigenous concerns (Abdilla et al., 2020) Elsewhere in the United States, Data for Black Lives confronts algorithmic racism head-on, advocating for abolitionist approaches to technology and reparative data governance (Milner & Traub, 2021). These projects are not asking permission to participate in a pre-ordained technological future. They are building new futures from the ground up, using AI tools while refusing the extractive practices that dominate their development. To reclaim AI is to take seriously the political and material conditions of its design and to challenge the idea that AI should be governed only by engineers, CEOs, or think tanks. It is to insist that the future of AI must be collective, participatory, and accountable. And it is to recognise that this transformation doesn’t require waiting for benevolence from big tech or inclusion on someone else’s terms and that it is already happening, through practices of community-led stewardship, co-design, refusal, and care. Instead of asking how to regulate or reform dominant systems, reclaiming AI asks what if communities owned and governed their own data infrastructures, not just as subjects of extraction, but as stewards of knowledge? Or what if machine learning models were trained to reflect care, kinship, and resistance, rather than scale, efficiency, or dominance? And what if AI systems were designed from the ground up to be answerable to people and not profit, and are rooted in place, memory, and meaning? One thing is clear, Reclaiming won’t happen at scale unless more of us step in to organise, to challenge, to build, and to protect. Reimagining AI Collectively reshaping tech narratives through an environmental, decolonial and ancestral lens Reimagining AI moves beyond dismantling current systems and hegemonic ideologies shaping AI. Reclaiming AI strategies call for radical imagination, building alternative narratives and imaginaries as ways to “imagine and craft the world we cannot live without, just as we dismantle the ones we cannot live within.’ as Ruha Benjamin (2024) says. Beyond popular dystopian and “doom-ist”, and tech-utopian imaginaries (Strickland, 2022), reimagining AI –often speculative and experimental– develops new approaches to challenge oppressive systems. In opposition to capitalistic, anthropocentric ideologies, experiments drawing on degrowth and climate justice emerge. Central to many reimagining AI is the expansion of epistemological frameworks beyond dominant Western rationalism, or the “One World World” (Escobar, 2018). Adopting a pluriversal approach, reimagining AI discussions draw on other forms of knowing and intelligence, such as recollections of ancestral technologies, afroand indigenous futurism, feminism, and decoloniality. Through diverse lenses, such discussions dream up regenerative futures and centre relationality and collective responsibility (van Norren, 2020). 8 Duarte et al. (2025) Re-imagining AI situates computation practice within the histories, principles, and politics that have scaffolded it, and allows us to challenge the fantasy of “neutral” data and outputs, and intelligence that is beyond accountability. The “AI Decolonial Manyfesto” argues that AI should be defined by communities to “reshape reality on their own terms.” Authored by technologists and human rights activists, it advocates for AI to “move beyond Western-centric biases in isolation,” providing frameworks for “decolonial governance” of AI and, through this, recentres AI as a site of contestation and community input, rather than a finished and external artefact. We can also reimagine our interactions with AI by making the politics that underpin AI’s fantasy of neutrality legible at the point of use. The Dataset Nutrition project proposes a standardised and clear label that foregrounds embedded biases, absences, and intended uses of datasets. Instead of treating data as a neutral fuel for analysis and use in AI, these labels help recognise that the data reflects the wider structure of inequality by naming who made the dataset, for whom it was curated, and whom it is missing. Once the ‘ingredients’ are visible, the myth of data’s view-from-nowhere collapses, allowing outputs to be seen as situated claims rather than universal truths. Once we are able to situate AI within the human political structures that it has been created for, we can start to reimagine the future of AI as post-anthropocentric, where we recognise the rights and intelligences of not only humans, but also other species. Superflux’s “Nobody Told me Rivers Dream” project, exhibited at the Design Museum in London, showcased the building of a kind of ecological intelligence. By placing sensors along the River Thames to capture environmental phenomena beyond ordinary human perception (birdsong before storms, wind on water, the tidal pulse), each with an AI model embedded inside which interprets the data and builds a complex network of ecological knowledge. These projects reframe ‘innovation’ as a negotiated co-existence, where the environment can provide frameworks for intelligence which augment human capabilities for understanding. An example of reimagining AI through more-than-human intelligence is BlueMarble; a speculative project born out of a collaboration between the UNDP Strategic Innovation Unit and the UN Futures Lab (n.d.) set in the near future, in April 2030, where planet-multispecies collaboration is realised for planetary health through employing more than human intelligence. They imagine this collaboration through two technologies, the Mycelium Network Interface (MNI), and the Multi-Language Interspecies Interpreter (MLII). The MNI, introduced in 2038, reimagines artificial intelligence through mimicking the connectivity and symbiotic properties of “nature’s internet” , the mycelial thread, to act as an environmental monitor and sensor to connect multiple plant species and carry signals. The MNI gathers this data and interprets it to understand forest health and ecosystem dynamics. The MLII, introduced in 2036, leverages machine learning to facilitate inter-species communications through interpreting various “data”, such as behavioural patterns, vocalisations, and vibrations, into a language “all species can understand”. This interpretation is imagined to facilitate collective decision making for planetary governance, decentering humans as the single intelligent species. This speculative project brings forth artefacts and provocations to challenge systems predicated on extraction, optimisation and control, inherited from industrial capitalism and technocratic rationality (Marcucci, 2025). The ideas demonstrated in Blue Marble build on various indigenous and African epistemologies, which centre principles of relationality, degrowth and plurality in relation to human and non-human connection. Relational philosophies –like those of the Anishinaabe (Wilson, 2008) and Ubuntu– maintain a world view of “All My Relations”, and “I am because we are” (Mhlambi, 2020) respectively, affirming knowledge and intelligence created through reciprocity, kinship and connection through relationships within communities, with land and non-human ‘relatives’. Embedding these principles in AI design and development calls for a shift in paradigm, from extraction and centralisation, to distributed, accountable and to enhance connectivity and cultural relationships. Equally, the concept of plurality, as explored in Escobar’s Designs for the Pluriverse (2018), proposes AI developed and governed through diverse ways of knowing, situated in multiple local contexts as opposed to universal models that generalise solutions under a guise of rationality, respecting and bringing forth the wisdom and interconnectedness of many human cultural alongside planetary ones. 9 DOI: 10.5281/zenodo.17343830 Yanchur, A. (2024) ‘“All Rise For the Honorable AI”: Algorithmic Management in Polish Electronic Courts’, AlgorithmWatch, 27 May. Available at: https://algorithmwatch.org/en/polish-electronic-courts/ (Accessed: 4 October 2025). Zewe, A. (2025) ‘Explained: Generative AI’s environmental impact’, MIT News, 17 January. Available at: https://news.mit.edu/2025/explained-generative-ai-environmental-impact-0117 (Accessed: 4 October 2025). Contributions Conceptualisation Project Administration Tania Duarte Ismael Kherroubi Garcia Ismael Kherroubi Garcia Tania Duarte Supervision Visualisation Tania Duarte Tania Duarte Ramla Anshur Writing – original draft Writing – review & editing Ramla Anshur Ramla Anshur Tania Duarte Tania Duarte Harriet Humfress Harriet Humfress Ismael Kherroubi Garcia Ismael Kherroubi Garcia Dylan Orchard Dylan Orchard Steph Wright Steph Wright Acknowledgments We are grateful to everybody who submitted a case study. Each case study was crucial in refining our thinking. Thank you to PJ Annand, Molly Crockett, Stef Garasto, Patricia Gestoso, Georgia Iacovou, Laura Martinez Agudelo, Kailyn Slater, Eleanor Taylor. Thank you to Janet Turra for creating inspiration artwork to illustrate a vision of Resist, Refuse, Reimagine, Reclaim for Tania Duarte’s “Imagined Futures” presentation at AI UK in March 2025. Thank you to Leon Furze for helping to give shape to ideas. 16