AI for climate: from technological solutions to relational accountability Authors: Sharon Stein Published: 2026 ABSTRACT Despite the promise of “AI-for-climate” to catalyze climate research and action, dominant approaches risk reproducing the same cultural patterns that have fueled ecological breakdown and its unevenly distributed harms. These patterns include human exceptionalism, technosolutionism, and epistemic universalism. By tracing these patterns, this comment situates AI-for-climate within a broader web of extractive planetary relations, examining its potential to either reinforce those relations or contribute to their repair. In doing so, it invites reflection on the ethical and epistemological challenges of co-developing AI systems that are accountable to the diverse human communities, species, and generations who share our entangled Earth. Introduction Artificial Intelligence (AI) is increasingly framed as a key enabler of climate monitoring, modelling, adaptation, and mitigation. In this comment, I use “AI-for-climate” to refer to AI systems designed for these purposes. From optimizing energy use and improving predictive capacity to scaling adaptation planning and offering decision support, AI-for-climate holds considerable promise1–4. Yet AI is not neutral. Like humans, it is shaped by the political, economic, and epistemological systems from which it emerges, and in which it is embedded. AI-for-climate thus risks reinforcing the dominance of Western societies, economies, and knowledges and reproducing the extractive relational dynamics that are driving climate destabilization in the first place5. Many contemporary AI systems also rely materially on infrastructures of extraction, including global energy grids, rare earth mining, and precarious labour. At the same time, some researchers and communities are beginning to develop AI systems that are more aligned with our condition of planetary entanglement, guided by intergenerational and interspecies responsibilities, and responsive to local knowledge systems and priorities. This comment addresses some of the epistemological and ethical complexities of designing more relationally accountable AI-for-climate systems. I write as a white settler researcher who is shaped by and implicated in the colonial systems this piece examines. The reflections offered here are informed by long-standing critiques from Indigenous, decolonial, and climate justice literatures that trace how extractive dynamics are reproduced in digital, ecological, and political infrastructures5–11, including in the development and governance of AI12–16. I draw on this scholarship to trace three recurrent and interrelated cultural patterns shaping AI-for-climate: (1) human exceptionalism, (2) technosolutionism and optimization, and (3) epistemic universalism. Collectively, these patterns narrow the scope of whose knowledge is treated as real and important, what kinds of intelligence are accounted for, and which climate futures are deemed possible and desirable. These patterns are grounded in the same ontological paradigm that drives climate change: one that rationalizes the dispossession of land, “natural resources,” and knowledge in the name of progress, growth, and development. This paradigm has also led to the paternalistic imposition of Western-designed climate solutions onto non-Western communities who are disproportionately impacted by ecological destabilization5, thereby devaluing those communities’ own climate knowledges and bypassing their governing authority6–9. Naming these patterns is not an argument for rejecting AI altogether, but an invitation to situate AI-for-climate systems within practices of accountability that recognize their entanglement in human and more-than-human relations12,13,16. Cultural patterns in AI-for-climate The cultural patterns that shape modern Western societies reflect their specific histories, political-economic systems, and underlying ontological assumptions. I trace how three of these patterns operate within AI-for-climate efforts, while also gesturing towards emerging efforts to design AI systems that interrupt these patterns and invite a paradigmatic shift in how human communities relate to each other and to the living systems we not only depend on, but are also part of. Pattern 1: human exceptionalism Many AI-for-climate efforts operate within an inherited hierarchy that positions humans as masters of nature rather than participants within it. This pattern naturalizes human exceptionalism, erases more-than-human agency, and renders ecological systems into external objects for measurement and management17,18. AI systems that translate rivers, forests, or soils into abstract data mirror deeper cultural logics that treat them as resources to be optimized for human benefit. Increasingly, AI-enabled ecosystem monitoring draws on remote sensing infrastructures and large-scale modelling to predict floods, droughts, and extreme weather, assess watershed health, and track forest degradation. This can yield valuable insights for climate adaptation and mitigation. Yet when ecosystems are approached primarily as data inputs, without engaging the territorial, cultural, and relational contexts from which those data emerge, it reinforces a framing of the more-than-human world as an object of top-down technical oversight, rather than a multi-species community of living beings19,20. In this way, even well-intentioned AI-for-climate systems can extend forms of surveillance and administrative control over territories and the human communities who inhabit and steward them, deepening structural vulnerabilities and obscuring long-standing interspecies responsibilities that are central to many Indigenous and other non-Western worldviews6,10. Addressing this pattern would require embedding relational accountability into AI-for-climate systems by designing and governing them not only to generate data about ecosystems, but also to support the renewal of respectful and reciprocal relations between humans and more-than-human beings. What this looks like in practice will depend on the specific context. It may involve pairing remote sensing with community-led ecological monitoring that upholds territorial and data sovereignty and treats consent and co-design as conditions of AI development and deployment. Here, AI’s role would not be to replace community expertise but to dialogue with it in ways that strengthen rather than supersede local governance and self-determination. Such collaborations should also include provisions to cease data collection if participating communities determine that the use of AI is leading to increased risks of territorial enclosure, intensified extraction, or cultural harm. In other contexts, AI-for-climate systems might be designed to invite climate researchers and policymakers into reflexive practices that surface inherited assumptions about human authority and more-than-human agency. For instance, they could prompt reflection on what shifts when an ecologically threatened river is approached not as a resource, but as a living being with its own forms of agency, intelligence, and responsibility. These AI systems could also support researchers to engage respectfully with the knowledge traditions that embody this more relational ontological orientation, including many Indigenous epistemologies and the rights of nature movement21,22. AI-for-climate systems could also invite a shift in how intelligence is commonly understood, reframing it as a dynamic, relational process that emerges through interactions between humans, more-than-human beings, and machines, rather than as a fixed property or possession of individuals. Exploring this shift might involve collaborations among frontline communities, climate researchers, artists, and AI designers to consider whether and how the intelligence and agency of non-human animals and plants might be engaged through AI systems without anthropomorphizing them or reducing them to human (and human-centric) constructs23. Such efforts would need to proceed cautiously, recognizing the limits of interspecies translation and the risk of reinscribing human (and specifically Western) interpretive dominance. Pattern 2: technosolutionism and optimization A pervasive technosolutionist orientation in climate research and action casts climate disruption primarily as a technical problem to be solved with more data, scientific innovation, and sophisticated modelling24. This pattern also orients many AI-for-climate innovations designed to optimize existing systems (e.g., transportation, energy, agriculture, resource management25). AI itself is materially entangled with these same systems: both the training and use of large-scale AI models require considerable water, electricity, and critical minerals26. While increasing the efficiency of existing systems can yield important benefits, rebound effects may lead to increased overall emissions and material extraction, a dynamic known as Jevons' paradox 27. More broadly, optimizing current systems often leaves the structural drivers of ecological breakdown intact. These include imperatives for economic growth, disconnection from and domination over nature, and the prioritization of short-term individual advantage over long-term collective flourishing28. In this way, technosolutionism mistakes a relational problem for a technical one. This pattern can lead AI-for-climate systems to sidestep deeper questions, such as: Whose futures are being encoded into algorithmic systems? What relations of power are being reproduced or interrupted29? In some contexts, AI systems could be designed to accompany broader efforts to transform unsustainable systems. For example, they might invite users to surface and examine the assumptions that underpin extractive practices that damage relations between humans and ecosystems, and consider how technical interventions might be paired with the more difficult work of relational repair. AI systems can also be co-designed by and with the communities they are intended to serve, ensuring that the technology aligns with place-based priorities, knowledges, and ways of being. Yet even in these cases, there is no guarantee that AI will interrupt the entrenched economic and institutional incentives that continue to reward optimization, scale, and growth. It is therefore essential not to treat AI as an objective source of climate solutions, but as a situated actor in the ongoing, collective work of responding to both the symptoms and root causes of ecological breakdown. Pattern 3: epistemic universalism Many mainstream AI-for-climate interventions privilege Western scientific and computational knowledge as the primary basis for understanding and addressing ecological challenges. Large-scale models are typically trained on datasets dominated by Western contexts, languages, and reasoning patterns, encoding a Eurocentric worldview that marginalizes other ways of knowing, being, and relating15. Even the field of AI ethics is largely shaped by Western perspectives30. Simply adding more knowledges and languages into AI models does little to shift this epistemic centralization if the systems’ underlying reasoning structures and ontological assumptions remain unchanged16,31. In such cases, non-Western epistemologies are often subsumed within a Western framework that limits what is recognized as real, valuable, or intelligible11. The result is that distinct approaches to intelligence, relationality, and technology are flattened or erased. Moreover, incorporating communities’ languages, stories, or other forms of “data” without their free, prior, and informed consent constitutes a continuation of colonial data extraction in digital form32. The prioritization of predictive accuracy as a core metric of AI system success also reinforces the pattern of epistemic universalism. It minimizes the role of uncertainty, complexity, and contextual specificity in shaping ecological disruption and responses to it. This then elevates quantifiable foresight over and above local knowledge and relevance, emergent co-constructed interventions, and the value of holding multiple truths, priorities, and potential futures without imposing consensus. AI-for-climate systems intended to counter universalism would therefore need to do more than grapple with biased training data. They would need to reflexively foreground their own partiality, with humility, while also recognizing that certain types of knowledge may remain intentionally opaque or inaccessible out of respect for communities’ epistemic sovereignty. With the consent of all participating communities, AI systems could be co-designed to facilitate encounters across distinct intelligences and knowledge systems without presuming commensurability or full mutual intelligibility. Such efforts would require acknowledging the long histories of extractive research and epistemic violence that many communities have experienced, honouring their rights to decline participation without penalty, to protect sacred or sensitive knowledges, and to define terms of engagement that may not align with conventional AI development timelines or workflows32. For the communities that do participate, upholding data sovereignty also requires ensuring they retain governance over the digital infrastructures through which their knowledge moves, including servers, storage, and model deployment33. While some communities assert sovereignty by refusing AI, and others engage AI through carefully negotiated collaboration, some are creating their own initiatives. Projects like Abundant Intelligences and Earth Friendly Computation are developing AI systems that centre Indigenous community knowledges and ontologies and decentre Western computational logics, challenging the presumed inevitability of centralized, ecologically costly, and universalizing AI13,14. Conclusion This comment has traced how many current approaches to AI-for-climate are grounded in the same cultural patterns that have contributed to ecological breakdown: separating humans from the rest of nature, casting climate disruption as a problem to be solved through top-down technological innovation, and universalizing particular ways of knowing, being, and relating. However, there are emerging efforts to develop and deploy AI-for-climate systems whose architecture and governance are more accountable to the human communities they affect and the wider planetary systems in which they are embedded. Among the possibilities considered here is approaching AI as a participant within a wider ecology of intelligences, and grounding its reasoning on ontological foundations of interdependence, reciprocity, and plurality, rather than separation, extraction, and universality16. Ultimately, if climate researchers and AI developers want to support this reorientation, we will need to go beyond technical redesign and undertake the far more difficult work of unlearning the patterns of extraction and exceptionalism that shape our own relationships with the Earth and one another.