Can AI help reduce poverty without reinforcing inequality?

AI is now being embedded in governance systems that identify vulnerable populations, allocate welfare resources, monitor service delivery and support decisions about who receives assistance. A new systematic review finds that AI can improve poverty-related decision-making whilst warning that biased data, weak infrastructure and limited oversight could reinforce the very inequalities such systems are meant to reduce.

The review, The Role of Artificial Intelligence in Poverty Governance: A Systematic Literature Review of Innovations and Implementation Challenges, published in Administrative Sciences, reviews recent research on how AI is being used in poverty governance, including social protection, agriculture, health, education, financial services and public administration.

The authors argue that AI should be treated not as a standalone anti-poverty solution, but as a socio-technical system whose impact depends on institutions, data quality, regulation and inclusion.

AI is moving into the machinery of poverty governance

According to the review, AI can strengthen poverty governance by processing large and varied datasets more quickly than traditional systems. Machine learning tools can combine satellite imagery, mobile phone data, administrative records and digital transactions to produce more detailed poverty maps and detect emerging vulnerabilities. These tools may help governments and aid agencies move from delayed responses toward more timely interventions.

Traditional poverty measurement often relies on household surveys and administrative records that may be incomplete, outdated or collected at long intervals. AI-based approaches offer a possible way to update poverty information more frequently, especially in places where conventional data collection is difficult. In principle, this can help target cash transfers, food aid, subsidies and emergency support more accurately.

AI can also improve administrative efficiency. Automated systems can support eligibility checks, fraud detection, payment scheduling and programme monitoring. In public sectors with limited capacity, these tools could reduce delays and help agencies coordinate services more effectively.

The strongest evidence of AI's value appears in areas where the task is predictive, classificatory or administrative. Poverty mapping, beneficiary targeting, early warning systems and resource allocation are among the clearest use cases. AI is also being applied to financial inclusion by using alternative data for credit scoring, helping people without formal credit histories access loans, savings, insurance and other financial services.

In agriculture, AI can support rural livelihoods through precision farming, crop yield forecasting, pest detection, soil monitoring and climate-smart advisory systems. In health and education, AI can assist with diagnostics, telemedicine, adaptive learning and service optimisation. Across these sectors, the common promise is better information, faster coordination and more targeted support.

Bias, exclusion and weak institutions remain major barriers

AI can amplify both capability and inequality, the review warns. Systems trained on incomplete or biased datasets may fail to identify the poorest or most marginalised groups. People in informal settlements, remote rural areas or digitally disconnected communities may be poorly represented in the data used to build poverty models. If those models are used to allocate benefits, the result could be exclusion from support.

This risk is concerning when AI systems are used for eligibility decisions in social protection, credit access or public service delivery. An error in a poverty prediction model is not merely technical. It can determine whether a household receives food aid, cash assistance, healthcare support or access to credit.

The study also raises concerns about digital exclusion. AI systems depend on data infrastructure, connectivity, electricity, computing resources and technical expertise. In low-resource settings, weak infrastructure can limit both the accuracy and sustainability of AI deployment. If digital systems are introduced without ensuring access for poorer communities, they may widen gaps between those who are visible to the state and those who remain outside formal systems.

The review points to concerns around privacy, surveillance, algorithmic accountability and data justice. In poverty governance, data are often collected from vulnerable populations. Without strong rules on consent, purpose limitation, auditability and appeal mechanisms, AI systems could expose people to monitoring or automated decisions they cannot understand or challenge.

AI shouldn't be treated as a neutral technical fix, the authors warn. Poverty is shaped by income, health, education, housing, food security, public services, political voice and exposure to shocks. AI can help detect patterns across these dimensions, but it cannot replace policy choices or address structural causes of deprivation on its own.

The authors call attention to the risk of data colonialism, where AI infrastructure, cloud systems, proprietary models and data-processing tools are controlled outside the communities or countries where poverty interventions are implemented. This can limit local control over development systems and deepen dependence on external platforms.

Another gap is the limited evidence on long-term outcomes. Many studies report technical performance, but fewer examine whether AI systems actually reduce poverty, improve fairness or strengthen public accountability over time. The review calls for more impact evaluations, fairness-aware AI, participatory design and methods suited to low-resource environments.

Why it matters for policy and development

AI can support poverty reduction only if it is built into accountable, inclusive and locally grounded governance systems. Predictive accuracy alone is not enough. Governments and development agencies must also ask who controls the data, whose needs are represented, who can contest decisions and whether affected communities have a voice in system design.

Policymakers must prioritise data governance. AI systems used in poverty programmes need clear rules on privacy, data protection, transparency, auditability and human oversight. Communities affected by automated decisions should have meaningful ways to question or appeal outcomes.

Then comes inclusive infrastructure. Investments in AI for poverty governance must be matched by investments in connectivity, digital literacy, public-sector capacity and locally relevant data systems. Without these foundations, AI tools may work best for populations that are already visible, connected and easier to measure.

The third priority is institutional capacity. Public officials, civil society groups and local organisations need the skills to interpret AI outputs, monitor bias and ensure that automated tools remain aligned with social justice goals. Human-in-the-loop oversight should not be treated as optional when systems affect access to essential services.

The findings also matter for development agencies and technology providers. Tools designed in high-income or data-rich contexts may not transfer smoothly to low-resource environments. Context-sensitive design, participatory consultation and independent auditing are essential before AI systems are used in welfare targeting, credit scoring or service allocation.