# When AI Decides Who Survives: Climate Migration and the Equity Gap > Source: > Published: 2026-07-10 01:00:30+00:00 ## When AI Decides Who Survives: Climate Migration and the Equity Gap Somewhere in the Horn of Africa, a machine is trying to guess where you will go next. It does not know your name. It has never visited your village. It cannot tell you whether the rains will fail again, only that the historical pattern, the commodity prices, the conflict-event logs and the climate anomalies it has ingested suggest that a certain number of people in your region will move within the next month. The model produces a figure. The figure becomes a dashboard. The dashboard informs a contingency plan. And somewhere down that chain, a decision gets made about whether food, shelter and protection arrive before you do, or long after. This is not a thought experiment. Since 2017, the United Nations refugee agency UNHCR has run a predictive analytics platform called Project Jetson, a machine-learning system designed to forecast forced displacement in Somalia roughly a month in advance across eighteen regions. It pulls together data on conflict, fatalities, wages, commodity prices, climate anomalies and historical movement, and it turns that mixture into an early-warning signal that humanitarian planners can act on. The promise is seductive and genuinely humane: anticipate the wave before it breaks, and you can save lives instead of merely counting them afterwards. But there is a quieter question buried inside the dashboard, and it is the question that now haunts an entire field. If an algorithm is deciding, even partially, who gets help and when, who taught it what help looks like? And whose idea of a life worth protecting did it learn from? In the spring of 2026, a group of researchers writing in the Nature-published journal *Humanities and Social Sciences Communications* tried to answer that question head-on. Their comment, titled “Artificial intelligence and climate migration equity” and published on 28 March, lands as a warning shot. The datasets that underpin the AI tools used to predict migration flows, screen asylum applications and direct humanitarian resources, the authors argue, are systematically tilted towards the interests and information sources of wealthier nations. The communities facing the sharpest edge of climate disruption, those across the Global South, are the least represented in the training data and the most exposed to the consequences of what those systems decide. The piece, authored by Lawrence A. Palinkas, Mustafa F. Özbilgin, Miriam Aczel, Nathalie Ortar, Claire Monteleoni, Sarab Sethi, Eric Rice, Bistra Dilkina and Michàlle Mor Barak, does not read like a manifesto against technology. Several of its authors build AI systems for a living. It reads like a field issuing a correction to itself before the correction becomes an inquest. ## The Scale Nobody Has Fully Reckoned With To understand why the stakes are so high, you have to start with the numbers, and the numbers are staggering. A 2024 review published in *npj Climate Action* projected that roughly 143 million people across the Global South could face displacement by 2050. The distribution is brutally uneven: around 86 million in sub-Saharan Africa, 40 million across South Asia and the Pacific, and 17 million in Latin America and the Caribbean. These are the regions that contributed least to the emissions driving the crisis and stand to lose the most from it. The same review noted that a one-degree-Celsius rise in temperature correlates with a 1.9 per cent increase in global migration, and catalogued the immediate triggers in granular detail. The 2022 floods in Pakistan alone affected 33 million people and displaced 2.1 million. Flash floods in Bangladesh's Haor wetlands affected 4.2 million residents. The review was also unsparing about the moral geometry of the crisis. Sub-Saharan Africa, it noted, faces severe droughts and water scarcity despite contributing a negligible share of global greenhouse-gas emissions. Pacific island states confront sea-level rise that threatens habitability itself. Latin America contends with glacial melt and mega-droughts, South Asia with extreme flooding and temperatures that are beginning to break the agricultural systems on which hundreds of millions of people depend. The displacement these stressors produce, the authors stressed, rarely follows a clean line from disaster to departure. It moves through indirect pathways shaped by institutions, politics and economics, which is precisely the kind of messy, context-dependent causation that data-hungry models struggle to capture. These are not abstractions waiting somewhere in the future. They are people already moving, already navigating borders, already standing in registration queues, already waiting for an aid distribution that may or may not arrive. And increasingly, the systems that mediate those moments are algorithmic. That is the collision at the heart of this story. Tens of millions of the world's most climate-vulnerable people are encountering institutions that have begun, quietly and with the best of intentions, to outsource fragments of judgement to software. The software is fast, scalable and tireless. It is also, by the account of the researchers now scrutinising it, built on a foundation that was never designed to see them clearly. ## A Field Built on Borrowed Eyes The technical term for the problem is unglamorous: representativeness. The lived reality is anything but. Machine-learning systems learn the world from the data they are shown. If the data over-represents some places, languages, infrastructures and institutions while under-representing others, the resulting model does not merely have gaps. It has a worldview, and that worldview reflects whoever generated the most data and held the most power to label it. The *Humanities and Social Sciences Communications* authors put it plainly: lower-income regions remain underrepresented in AI-driven planning models, and without careful design, AI systems can reproduce structural inequities rather than redress them. Consider what generating high-quality data actually requires. Dense networks of sensors. Reliable electricity. Broadband connectivity. Administrative systems that record births, deaths, incomes, harvests and movements in machine-readable form. Research budgets that fund the painstaking work of collection, cleaning and labelling. These are precisely the assets concentrated in wealthier nations and scarce in the regions where climate displacement is most acute. The result is a perverse asymmetry. The places generating the cleanest, richest, most abundant data are often the places experiencing the least climate disruption, while the places experiencing the most disruption generate the thinnest, patchiest data trails. This imbalance is not confined to operational systems. A separate study in the same journal, examining the global landscape of migration research itself, found that knowledge production is profoundly unequal, with some countries and subregions remaining systematically underrepresented despite hosting significant migrant populations. African countries, Central Asia, and Latin America and the Caribbean were singled out as persistently understudied. If the scholarship that informs migration policy is itself skewed, then the data pipelines feeding operational AI inherit that skew at birth. The bias does not begin in the model. It begins in the entire apparatus of who gets studied, by whom, and with whose money. When a model trained predominantly on rich-world data is deployed to make decisions about the Global South, the mismatch is not a rounding error. It is structural. A flood-risk model calibrated on well-instrumented river basins in the Global North may misjudge the hydrology of a river that has never been gauged. A migration-prediction model that has learned the signatures of movement from contexts with formal labour markets and documented residency may be blind to the informal economies and undocumented mobility that define displacement across much of the Global South. The model is not lying. It is doing exactly what it was trained to do. The problem is what it was trained on. This is the deeper meaning of the equity warning. It is not simply that some communities are missing from a spreadsheet. It is that the absence becomes encoded, automated and scaled, and then sold back to those same communities as objective insight. ## The Women Who Vanished From the Data If the geography of data exclusion is stark, the demography of it is starker still, and it cuts along a fault line that has been ignored for so long it now reads as design. In February 2026, Columbia Climate School published an analysis by Pavi Selvakumar, a postdoctoral research scientist focused on the integration of AI and climate justice, and Marco Tedesco, a research professor at Columbia's Lamont-Doherty Earth Observatory and adjunct scientist at NASA's Goddard Institute for Space Studies. Their argument was uncomfortable and precise: women, who make up a disproportionate share of climate-displaced populations and shoulder the heaviest burden of climate adaptation at the community level, are almost entirely absent from the data and governance structures shaping the AI tools now being deployed for climate migration management. The supporting figures are difficult to read as anything other than an indictment. An analysis of 133 AI systems found that 44.2 per cent exhibited gender bias, and 25.7 per cent exhibited both gender and racial bias. The digital exclusion that feeds those biases is measurable too. There is a 21 per cent gender gap in global internet access, and in the least developed countries that gap widens to 52 per cent. The UN projects that 341 million women will lack electricity by 2030. Women globally spend an estimated 200 million hours every day collecting water. Even under conditions of equal access, the analysis noted, adoption of generative AI tools runs 20 to 25 per cent lower among women. Stack those numbers and a feedback loop comes into focus. Women are less connected, so they generate less data. They generate less data, so they are less visible to systems that learn from data. They are less visible to those systems, so the decisions those systems inform overlook their needs. And because women are over-represented among the climate-displaced, the oversight compounds at exactly the point of greatest vulnerability. The exclusion is not a single missing variable; it is a self-reinforcing cycle, and each turn of the cycle makes the next harder to see and correct. Selvakumar and Tedesco offer a concrete illustration of what this means on the ground. A disaster-response model built without women's input, they argue, might prioritise asset recovery while overlooking concerns that fall disproportionately on women: prolonged heat exposure, the safety and sanitation conditions of evacuation shelters, the continuity of medication during displacement, the loss of informal income. None of these are edge cases. They are central to whether displacement is survivable. They are also precisely the things a model trained on infrastructure and assets, rather than on the texture of women's daily survival, will never think to optimise for. The authors push the analysis further, arguing that intersecting factors of class, race, caste, migration status, geography and age compound the distortion, so that AI-driven climate governance ends up aligning technological power with existing social privilege and reinforcing the very inequalities it claims to address. Their prescription is a feminist approach to AI governance, one that refuses to treat any of this as incidental and instead asks the foundational questions: who produces the data, whose labour is recognised, who bears the environmental costs, and who actually participates in deciding any of it. Without those steps, they conclude bluntly, there can be no climate justice. ## What the Systems Actually Do It would be easy to treat all of this as speculative, a warning about machines that have not yet been built. They have been built. They are running now. The clearest examples come from the humanitarian sector itself, where the intentions are unambiguously good and the equity questions are therefore at their most instructive. The World Food Programme operates HungerMap LIVE, an AI-based platform that tracks and predicts food security in near real time across more than 95 countries, combining weather, population, conflict, hazard, nutrition and macroeconomic data to forecast conditions 30 to 90 days ahead. The WFP's Optimus tool helps decide which supplies go where, based on factors such as location and population size, and the organisation has reported serving 20 per cent more people on the same budget as a result. Its SKAI project, developed with Google Research, uses computer vision to compare satellite imagery before and after disasters and assess damage within 24 hours. The underlying prediction engines have grown formidable. The *Humanities and Social Sciences Communications* authors point to Google DeepMind's GenCast, a system that applies graph neural networks to meteorological data to produce weather forecasts that outperform established physics-based models, the kind of capability that turns a vague sense of “the rains might fail” into a probabilistic warning a planner can budget against. They note that machine-learning models such as long short-term memory networks are being used to predict disease outbreaks in the aftermath of floods, when stagnant water and disrupted sanitation can turn displacement camps into vectors for cholera and other waterborne illness. The authors map five dimensions across which these tools could, in principle, support more equitable interventions for climate migrants: disaster preparedness and response, health disparities, community sustainability, resettlement, and child development. Read in isolation, the list is almost utopian, a vision of anticipatory humanitarianism in which suffering is forecast and forestalled rather than merely tallied. The catch, the one the authors return to insistently, is that every one of those five dimensions depends on data, and the data is exactly where the equity problem lives. On the migration side, the IOM-Microsoft collaboration that the authors cite is a textbook case of the promise. In Ethiopia, the partnership analysed satellite imagery, population data, cropland maps and IOM office locations to identify communities at risk of flooding, and concluded that 700,000 people and 1.5 per cent of the country's croplands were exposed. In the south-east near Somalia, it found that 9 per cent of the population along the Shabelle River sat in flood-prone areas, against a national average of 0.5 per cent. That is genuinely actionable intelligence, the kind that lets an agency pre-position resources where they will matter most. Then there is the other face of algorithmic migration governance, the one less interested in helping people move than in deciding whether they may. The European Union's forthcoming European Travel Information and Authorisation System, ETIAS, will profile visa-exempt travellers using a screening-rules algorithm that cross-checks personal data against security databases to generate predictive risk scores. Legal scholars have warned that this amounts to a new form of what they call “algorithmic discretion,” an instrument of differential exclusion that could fall hardest on certain groups of travellers. The EU's earlier iBorderCtrl pilot, which attempted to detect deception through facial recognition and the measurement of micro-expressions described as “biomarkers of deceit,” drew sustained criticism for relying on racialised assumptions and for the thin scientific basis of emotion-recognition technology applied to individual behaviour. The point of laying these side by side is not to flatten them into a single villain. A flood-risk map that pre-positions aid and a border algorithm that assigns a suspicion score are doing very different moral work. But they share a lineage. Both translate messy, contested human realities into scores and categories. Both operate with limited transparency and uneven safeguards. And both, when their training data reflects the priorities of wealthier states and well-instrumented institutions, risk encoding a hierarchy of whose movement counts as a logistics problem to be solved and whose counts as a threat to be screened. There is also a structural reason the humanitarian and the border systems keep appearing in the same breath, and it is worth naming. Both rely on the same scarce raw material: information about people who are, almost by definition, hard to count. A displaced person may have crossed a border without documents, may have no fixed address, may speak a language poorly served by the natural-language tools doing the processing, may have every reason to distrust the institution collecting their data. The systems built to help them and the systems built to screen them are drinking from the same shallow, muddy well. When the well is shallow, the system fills the gaps with assumptions, and assumptions, in machine learning, are just biases that have not yet been measured. ## The Efficiency Trap Here is where the WIRED-era faith in optimisation meets its hardest test, because the systems are not failing on their own terms. They are succeeding. Serving 20 per cent more people on the same budget is a real achievement. Forecasting displacement a month ahead is a real capability. Identifying 700,000 people at flood risk is a real and useful insight. The machinery of humanitarian AI is, by the metrics it was built to maximise, working. And that is precisely the problem the equity literature is circling. A system optimised for efficiency will faithfully optimise for efficiency, which means it will reward whatever the data tells it produces the most measurable benefit per unit of cost. If the data systematically under-counts women, informal workers, undocumented movers and unmonitored regions, then the most “efficient” allocation will quietly route resources away from exactly those groups, not out of malice, but out of arithmetic. This is the discrimination-in-effect problem, and it is far more insidious than discrimination by intent. No engineer needs to harbour a bias for the outcome to be biased. The bias is upstream, baked into what was measured and what was ignored, and the optimiser simply carries it downstream at scale and at speed. A human caseworker who overlooks a displaced woman's need for medication continuity makes one error. A model that has learned to overlook it makes that error a million times, consistently, and calls it a result. Worse, it launders the error through the appearance of objectivity. A number on a dashboard carries an authority that a harried official's hunch never could, even when the number is wrong, and the harder a decision is to contest, the more that borrowed authority matters. The temptation, when confronted with this, is to reach for a technical fix: better data, debiasing algorithms, fairness constraints. Those tools matter, and the researchers calling for them are not naive about their value. But the deeper argument running through both the *Humanities and Social Sciences Communications* comment and the Columbia analysis is that the problem is not fundamentally technical. It is about power. It is about who gets to define what the system is optimising for in the first place, and a debiasing routine applied after the fact cannot answer a question that was never asked at the design stage. You can tune a model to allocate resources more evenly across the groups it can see. You cannot tune it to care about the groups it cannot see, because to the model they simply do not exist. ## Consent Was Never Requested There is a word that recurs in the equity literature, and it is the word that turns a technical critique into a moral one: consent. The communities whose lives are increasingly mediated by these systems were not, by and large, asked. They did not participate in defining the problem. They did not contribute the knowledge that shaped the models. They were not consulted on what counts as a good outcome. They cannot, in most cases, see the systems that sort them, let alone contest the results. The *Humanities and Social Sciences Communications* authors name this directly when they call for co-design and co-ownership of the AI design process with climate migration stakeholders, including vulnerable and affected communities. The phrase “co-ownership” is doing heavy lifting there. It is a long way from the standard humanitarian-tech posture, in which affected populations are sources of data and recipients of services but rarely architects of the systems that govern them. This is where the field's intellectual borrowing becomes telling. The authors ground their governance recommendations in human-centred design, community engagement, data feminism and decolonial theory. That is a deliberate set of references. Data feminism insists that data is never neutral and that the question of who is counted is inseparable from who holds power. Decolonial theory insists that knowledge produced about a place by outsiders, however well-intentioned, can reproduce the extractive relationships of empire under a new technical vocabulary. To invoke both in the context of AI for climate migration is to say, in effect, that the field risks building a digital infrastructure of governance over the Global South that mirrors the analogue injustices that came before it. The consent problem also exposes the limits of the “AI for good” framing that surrounds much of this work. Intentions are genuinely good. The IOM, the WFP and UNHCR are not malign actors; they are organisations trying to do an impossibly hard job with finite resources, and AI offers them real leverage. But good intentions are not the same as legitimate authority, and a system can be both well-meaning and illegitimate if it governs people who had no say in its creation and no recourse against its errors. The history of development is littered with interventions that were generous in spirit and disastrous in effect precisely because the people they were meant to help were treated as beneficiaries rather than authors. Algorithmic humanitarianism risks repeating that pattern at the speed and scale of software. ## So Who Is Accountable? This is the question the whole debate has been circling, and it is the one with the least satisfying answer, because accountability in algorithmic systems is engineered to be diffuse. Consider the chain. A government or agency decides to deploy a system. A private technology company builds it. The training data comes from a patchwork of sources, some public, some proprietary, collected by still other parties under still other conditions. The model is integrated into a workflow alongside human decision-makers who may or may not understand how it works and may or may not be free to override it. When the outcome is discriminatory in effect, every link in that chain can plausibly point to another. The agency says it relied on the vendor's tool in good faith. The vendor says it built to specification using available data. The data providers say they collected what they could. The caseworker says the system flagged the case. Responsibility evaporates into the gaps between institutions. That diffusion is sharpened by a particular feature of the partnerships now driving this work. When a UN agency teams up with one of the world's largest technology companies, the resulting system sits across a public-private boundary that complicates every line of accountability. The humanitarian body brings the mandate, the field presence and the moral authority; the corporation brings the compute, the models and, often, the proprietary infrastructure on which the whole thing runs. Each can credibly disclaim responsibility for the other's domain. And the displaced person standing at the receiving end of the system has a relationship with neither. They cannot file a complaint with a cloud platform. They frequently cannot even learn that an algorithm was involved in the decision that shaped their fate. The *Humanities and Social Sciences Communications* authors are clear that aspiration is not enough to fix this. They call for enforceable mechanisms rather than aspirational principles alone, and they enumerate them: mandatory algorithmic auditing, transparency requirements for public-sector AI procurement, clear appeal and redress mechanisms for affected populations, and participatory oversight that allows displaced communities to contest AI-supported decisions. Each of those is a deliberate attempt to nail responsibility to a specific actor at a specific point in the chain. An audit requirement makes someone accountable for testing the system. A procurement-transparency rule makes the purchasing institution accountable for what it buys. A redress mechanism gives the affected person a named door to knock on. Participatory oversight puts the governed in the room where the system is judged. What unites these proposals is a refusal to accept that algorithmic decisions are uniquely ungovernable. They are not. The diffusion of responsibility is a choice, embedded in how these systems are procured and deployed, and it can be reversed by a different set of choices. The reason it so often is not reversed is that doing so is slower, costlier and less efficient, and we are back, once again, at the efficiency trap. Accountability is friction, and friction is precisely what these systems were sold as eliminating. ## The Environmental Irony There is a final twist that the *Humanities and Social Sciences Communications* authors are careful not to let slide, and it sharpens the whole picture into something close to absurdity. The AI being deployed to manage the consequences of climate change has a climate cost of its own. The authors cite an estimate that a single training run for a large language model can emit as much carbon as five cars over their entire lifetimes. Their recommendation is to favour energy-efficient algorithms and hardware, and to recognise that task-specific AI models built for targeted climate applications are generally far more energy-efficient than sprawling general-purpose foundation models. This is not a peripheral concern. It speaks directly to the equity argument, because the carbon burden of training and running these systems lands, like the carbon burden of everything else, disproportionately on the regions least responsible for emissions and least equipped to adapt. So the spiral is complete. Wealthy institutions build carbon-intensive AI, trained on data that over-represents wealthy contexts, to manage the displacement of people in poor regions who are being displaced in part by the emissions that wealthy contexts produced, using systems those people did not consent to and cannot contest. Stated baldly, it sounds like a parody of itself. Stated in the measured prose of a peer-reviewed comment, it sounds like a field finally looking squarely at what it has been building. The choice between a lean, task-specific model and a vast foundation model is not, on this reading, merely an engineering preference. It is an equity decision, because the energy the larger model burns is borrowed against the same future the displaced are already paying for. ## What Equitable Deployment Would Actually Require It would be a failure of nerve to end on the diagnosis alone, because the researchers raising these alarms are not arguing for abandonment. They are arguing for a different way of building. And when you assemble their prescriptions, a fairly concrete blueprint emerges. It would start with the data. The *Humanities and Social Sciences Communications* authors call explicitly for public-private-academic collaboratives to collect and integrate high-resolution, localised, open-access datasets tailored to address existing disparities. The word “open-access” matters as much as “high-resolution.” Data locked inside a single vendor's proprietary system cannot be audited by the communities it describes, and data that only describes well-instrumented regions cannot correct the asymmetry. Closing the representativeness gap is not glamorous work. It is the slow, expensive, unfashionable labour of measuring the places and people that current systems do not see, and doing it in partnership with them rather than about them. It would require gender to move from afterthought to architecture. Selvakumar and Tedesco's prescriptions are specific: investment in women's digital and green skills, data infrastructure that accounts for the informal care economy, enforceable environmental accountability across AI supply chains, and women in leadership positions within the climate and technology institutions that build these tools. The throughline is that you cannot debias a system into seeing what it was never built to look for. Gender has to be present at the design stage, in the room, in the data schema, in the definition of what a good outcome is. It would demand that co-design and co-ownership become operational realities rather than slogans. That means displaced communities helping to define the problem, contributing the local knowledge that no satellite can capture, and retaining a stake in the systems that govern them. It means treating consent as an ongoing relationship rather than a box ticked once. There is a hard-headed argument for this beyond the ethical one: a model that incorporates local knowledge is simply a better model. The herder who knows which floodplain becomes impassable first, the midwife who knows which shelters women will refuse to enter, the community elder who knows the routes people actually take when they flee, all of them hold information no satellite captures and no commodity-price index encodes. Excluding them is not only unjust. It is bad engineering. And it would require the enforceable accountability mechanisms to actually be enforced: audits with teeth, procurement rules with consequences, redress channels that real people can use, and oversight bodies in which the governed have genuine standing. None of this is technically impossible. All of it is institutionally inconvenient, which is a different and more honest kind of obstacle. The authors of the *Humanities and Social Sciences Communications* comment are unsentimental about the ceiling on all of this. AI alone, they write, cannot be expected to achieve climate migration equity. What it might do, if it is aligned with human-centred values and global justice, is help shift climate mobility policy away from perpetual crisis response and towards something more like resilience-building. That is a modest claim, and its modesty is the point. The danger has never been that AI will do too little for climate migrants. The danger is that it will do a great deal, efficiently and at scale, on terms set entirely by those who already hold the power, and that the people it sorts will discover too late that the system optimising their fate was never taught to see them at all. The machine in the Horn of Africa is still running. It is still guessing where the next wave of people will go. The question the field is now forcing itself to confront is not whether the guess is accurate. It is whether the people being guessed about will ever have a say in the guessing, and who answers for it when the guess goes wrong. Those are not engineering questions. They are questions about justice, dressed in the language of code, and they will not be optimised away. ## References & Sources Palinkas, L. A., Özbilgin, M. F., Aczel, M., Ortar, N., Monteleoni, C., Sethi, S., Rice, E., Dilkina, B., & Mor Barak, M. (2026). “Artificial intelligence and climate migration equity.” *Humanities and Social Sciences Communications*, Volume 13, Article 374. Published 28 March 2026.[https://www.nature.com/articles/s41599-026-07087-1](https://www.nature.com/articles/s41599-026-07087-1)Selvakumar, P., & Tedesco, M. (2026). “How Can AI Address Climate Justice When Women's Voices Are Silenced?” *State of the Planet*, Columbia Climate School. Published 27 February 2026.[https://news.climate.columbia.edu/2026/02/27/how-can-ai-address-climate-justice-when-womens-voices-are-silenced/](https://news.climate.columbia.edu/2026/02/27/how-can-ai-address-climate-justice-when-womens-voices-are-silenced/)Almulhim, A. I., Nagle Alverio, G., Sharifi, A., Shaw, R., Huq, S., et al. (2024). “Climate-induced migration in the Global South: an in depth analysis.” *npj Climate Action*. Published 14 June 2024.[https://www.nature.com/articles/s44168-024-00133-1](https://www.nature.com/articles/s44168-024-00133-1)“Prioritizing global equity in migration research.” *Humanities and Social Sciences Communications*.[https://www.nature.com/articles/s41599-025-06308-3](https://www.nature.com/articles/s41599-025-06308-3)UNHCR Innovation Service. “Project Jetson.” United Nations High Commissioner for Refugees. [https://www.unhcr.org/innovation/project-jetson/](https://www.unhcr.org/innovation/project-jetson/)UN Global Pulse. “Using Artificial Intelligence to Model Displacement in Somalia.” [https://www.unglobalpulse.org/project/using-artificial-intelligence-to-model-displacement-in-somalia/](https://www.unglobalpulse.org/project/using-artificial-intelligence-to-model-displacement-in-somalia/)World Food Programme. “HungerMap LIVE.” [https://hungermap.wfp.org/](https://hungermap.wfp.org/)WFP Innovation. “SKAI.” World Food Programme Innovation Accelerator. [https://innovation.wfp.org/project/SKAI](https://innovation.wfp.org/project/SKAI)International Organization for Migration. “IOM and Microsoft Collaborate to Address Climate-Driven Displacement.” [https://www.iom.int/news/iom-and-microsoft-collaborate-address-climate-driven-displacement](https://www.iom.int/news/iom-and-microsoft-collaborate-address-climate-driven-displacement)IOM Environmental Migration Portal. “Vulnerable Communities in Ethiopia at Risk of Flooding.” November 2024. [https://environmentalmigration.iom.int/sites/g/files/tmzbdl1411/files/documents/2024-11/vulnerable-communities-ethiopia-communities-at-risk-of-flooding-final.pdf](https://environmentalmigration.iom.int/sites/g/files/tmzbdl1411/files/documents/2024-11/vulnerable-communities-ethiopia-communities-at-risk-of-flooding-final.pdf)European Parliamentary Research Service. “Artificial intelligence in asylum procedures in the EU.” 2025. [https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/775861/EPRS_BRI(2025)775861_EN.pdf](https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/775861/EPRS_BRI(2025)775861_EN.pdf)Hertie School Centre for Fundamental Rights. “Algorithmic Risk in EU Migration and Asylum Governance.” [https://www.hertie-school.org/en/news/detail/content/algorithmic-risk-in-eu-migration-asylum-governance-reconciling-the-eu-ai-act-and-the-council-of-europe-framework-convention](https://www.hertie-school.org/en/news/detail/content/algorithmic-risk-in-eu-migration-asylum-governance-reconciling-the-eu-ai-act-and-the-council-of-europe-framework-convention)Brill. “Rule of Law Challenges of 'Algorithmic Discretion' & Automation in EU Border Control.” *European Journal of Migration and Law*, Volume 25, Issue 3 (2023).[https://brill.com/view/journals/emil/25/3/article-p249_1.xml?language=en](https://brill.com/view/journals/emil/25/3/article-p249_1.xml?language=en) **Tim Green** *UK-based Systems Theorist & Independent Technology Writer* Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at [smarterarticles.co.uk](https://smarterarticles.co.uk), challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship. His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities. **ORCID:** [0009-0002-0156-9795](https://orcid.org/0009-0002-0156-9795) **Email:** [tim@smarterarticles.co.uk](mailto:tim@smarterarticles.co.uk) Listen to the free weekly [SmarterArticles Podcast](https://www.smarterarticles.fm)