In BRICS economies, greater AI adoption is linked to measurable poverty reduction: a 1% rise in AI use corresponds to roughly 0.14% lower poverty in the short run and 0.39% in the long run. Gains are larger over time and depend on human development, clean‑energy access and sound governance, while rising inequality offsets benefits.
AI for poverty reduction (SDG 1): driving inclusive economic growth in BRICS countries
Using BRICS country panel data (2008–2023) and CS‑ARDL cointegration methods, the study finds that higher AI adoption is associated with statistically significant reductions in poverty—about 0.14% in the short run and 0.39% in the long run per 1% increase in AI adoption—conditional on human development, clean fuel access, growth, and governance.
The study examines the impact of artificial intelligence (AI) adoption on poverty reduction and inclusive economic growth in BRICS economy in the 2008–2023 period through the lenses of the endogenous growth theory and the capability approach developed by Sen. The research adopts second-generation econometric methods including the Pesaran cross-sectional dependence test, CIPS panel unit root test, Westerlund cointegration and the Cross-Sectionally Augmented Autoregressions Distributed Lag (CS-ARDL) model. The empirical findings indicate a statistically significant and strong connection between AI adoption and poverty reduction, with 1% of AI adoption resulting in 0.14 and 0.39% poverty reduction in the short run and long run, respectively, indicating that the poverty-reducing benefits of AI increase over time. Poverty reduction is further reinforced by human development, access to clean cooking fuels, economic growth, and effective governance, but income inequality has the opposite effect. The error correction value is negative and highly significant, indicating the stability of the long-run equilibrium relationship. Robustness checks confirm the consistency of the findings on an alternative measure of poverty. In the course of the research, AI can support poverty alleviation as a general-purpose technology, provided robust governance systems and supplementary funding for human development are in place. The results provide significant policy implications for using AI to achieve SDG 1 in developing economies.
Summary
Main Finding
Adoption of AI in BRICS (2008–2023) is strongly associated with poverty reduction: a 1% increase in AI adoption is estimated to lower poverty by about 0.14% in the short run and about 0.39% in the long run. The poverty-reducing effect of AI grows over time and is reinforced by human development, access to clean cooking fuels, economic growth, and effective governance, while higher income inequality reduces the poverty-reducing impact. Results are robust to an alternative poverty measure and the long-run relationship is stable (negative, significant error-correction term).
Key Points
- Quantitative magnitudes: 1% rise in AI adoption → −0.14% (short run) and −0.39% (long run) poverty.
- AI behaves like a general-purpose technology whose benefits on poverty accumulate over time.
- Complementary factors that amplify AI’s poverty-reduction: human development, clean cooking fuel access, GDP growth, governance quality.
- Countervailing factor: higher income inequality undermines AI’s poverty-reducing effect.
- Long-run equilibrium appears stable (significant negative error-correction coefficient).
- Robustness: findings hold when using an alternative poverty indicator.
- Theoretical framing: endogenous growth theory and Amartya Sen’s capability approach (emphasizing capabilities/human development as complements to AI).
Data & Methods
- Sample: BRICS countries (Brazil, Russia, India, China, South Africa), period 2008–2023.
- Core variables: national poverty indicator(s), AI adoption (paper construct—see full text for exact measure/definition), controls including human development, access to clean cooking fuels, GDP growth, governance quality, and income inequality.
- Econometric strategy (second‑generation panel methods to address cross-sectional dependence and heterogeneity):
- Pesaran cross-sectional dependence test (to detect cross-section dependence).
- CIPS panel unit root test (Cross‑sectionally augmented IPS) for stationarity.
- Westerlund cointegration tests for long-run relationships.
- CS-ARDL (Cross‑Sectionally Augmented Autoregressive Distributed Lag) model used to estimate short- and long-run effects while accounting for cross-sectional dependence and heterogeneous slopes.
- Error-correction representation used to gauge speed/stability of adjustment.
- Robustness checks: alternative poverty measure(s) tested—results remain consistent.
- Note: the exact operationalization of the AI adoption variable (index/components) is detailed in the full paper.
Implications for AI Economics
- Dynamics and returns: AI’s poverty-reducing benefits are dynamic and larger in the long run—studies of AI economics should model dynamic accumulation and diffusion rather than only cross-sectional snapshots.
- Complementarity with human capital and public goods: empirical results emphasize complementarities—AI investment without parallel investments in human development and basic infrastructure (e.g., clean energy, education, health) will yield smaller social returns. Models and policy assessments should incorporate these complementarities.
- Governance matters: quality of institutions strongly conditions AI’s social payoff. Research should explicitly model governance as a moderator (interaction effects) and evaluate regulatory/policy design.
- Distributional considerations: income inequality weakens AI’s anti-poverty effect. AI economics must move beyond aggregate growth effects to distributional outcomes (who gains/loses), include heterogeneity across skill levels, sectors, and regions.
- Measurement and identification: the study uses an AI adoption indicator at the country level. Future work should refine AI measurement (patents, investments, model deployments, firm-level adoption, skill penetration) and pursue stronger causal identification (IVs, natural experiments, policy discontinuities) to address potential endogeneity.
- Sectoral channels: evidence (and the paper’s discussion) point to agriculture, finance (credit scoring/financial inclusion), health, and education as key channels—policy design and microeconomic studies should test and quantify these transmission channels.
- Policy design: policymakers aiming to use AI for SDG‑1 should combine AI-supportive investment with targeted human-development funding and governance reforms; redistribution/ social protection may be needed to counteract displacement risks and inequality amplification.
- Research extensions: cross-country heterogeneity within BRICS (and beyond) deserves closer study; microdata linking AI adoption at firm/household level to welfare outcomes would clarify mechanisms and incidence.
For the exact AI adoption metric, variable definitions, and coefficient tables, see the full paper (Garg et al., 2026, Quality & Quantity; DOI: 10.1007/s11135-026-02892-x).
Assessment
Paper Typecorrelational
Evidence Strengthmedium — The study applies appropriate second‑generation panel methods that handle cross‑sectional dependence and long‑run dynamics and reports robustness checks, which strengthen confidence in the estimated associations; however, the analysis remains observational with a very small cross‑section (BRICS, N=5), potential measurement error in AI adoption, and limited protection against omitted variable bias or reverse causality, limiting causal claims.
Methods Rigormedium — The author(s) use up‑to‑date panel unit‑root and cointegration tests and the CS‑ARDL framework (including an error‑correction term) and conduct robustness checks, indicating solid time‑series/panel practice; nevertheless, the small number of countries, lack of exogenous identification strategy (IV/natural experiment), and limited discussion of measurement and functional‑form sensitivity constrain methodological rigor.
SampleAnnual country‑level panel of BRICS economies (Brazil, Russia, India, China, South Africa) for 2008–2023, using an aggregate AI adoption indicator, poverty measures (primary and alternative), and macro controls such as human development, access to clean cooking fuels, GDP growth, governance indicators, and income inequality.
Themesadoption inequality governance innovation
IdentificationUses panel time-series econometric techniques (Pesaran cross-sectional dependence test, CIPS panel unit root tests, Westerlund cointegration, and Cross-Sectionally Augmented ARDL with an error-correction term) to estimate long-run and short-run associations between an aggregate AI-adoption indicator and poverty measures; robustness checks use an alternative poverty measure. No quasi-experimental source of exogenous variation (IV, diff-in-diff with plausibly exogenous timing, or randomized intervention) is reported.
GeneralizabilitySmall cross-section (only five BRICS countries) limits external validity to other developing or developed countries, National‑level aggregate data hide within‑country heterogeneity (regions, sectors, households), AI adoption measure likely coarse or composite and may not capture sectoral differences in AI use or quality, Results may not generalize outside the 2008–2023 period or to non‑BRICS institutional contexts, Observational design limits causal extrapolation to policy interventions without complementary evidence
Claims (14)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| A 1% increase in AI adoption results in a 0.14% reduction in poverty in the short run. Social Protection | positive | poverty reduction (short run) |
Reading fidelity
high
Study strength
medium
|
0.14% poverty reduction per 1% AI adoption (short run)
|
| A 1% increase in AI adoption results in a 0.39% reduction in poverty in the long run. Social Protection | positive | poverty reduction (long run) |
Reading fidelity
high
Study strength
medium
|
0.39% poverty reduction per 1% AI adoption (long run)
|
| The poverty-reducing benefits of AI increase over time (long-run effect larger than short-run effect). Social Protection | positive | poverty reduction (temporal comparison) |
Reading fidelity
high
Study strength
medium
|
|
| Human development reinforces poverty reduction. Social Protection | positive | poverty reduction |
Reading fidelity
high
Study strength
medium
|
|
| Access to clean cooking fuels reinforces poverty reduction. Social Protection | positive | poverty reduction |
Reading fidelity
high
Study strength
medium
|
|
| Economic growth reinforces poverty reduction. Social Protection | positive | poverty reduction |
Reading fidelity
high
Study strength
medium
|
|
| Effective governance reinforces poverty reduction. Social Protection | positive | poverty reduction |
Reading fidelity
high
Study strength
medium
|
|
| Income inequality has the opposite effect (it works against poverty reduction). Social Protection | negative | poverty reduction |
Reading fidelity
high
Study strength
medium
|
|
| The error-correction term is negative and highly significant, indicating the stability of the long-run equilibrium relationship. Other | positive | long-run equilibrium stability (error-correction) |
Reading fidelity
high
Study strength
medium
|
|
| Robustness checks using an alternative measure of poverty confirm the consistency of the findings. Social Protection | positive | poverty reduction (robustness across measures) |
Reading fidelity
high
Study strength
medium
|
|
| The study analyzes AI adoption and poverty outcomes in the BRICS economies over the period 2008–2023. Other | null_result | study sample and period |
Reading fidelity
high
Study strength
high
|
|
| The empirical approach uses second-generation panel econometric methods: Pesaran cross-sectional dependence test, CIPS panel unit root test, Westerlund cointegration test, and the CS-ARDL model. Other | null_result | methodological approach |
Reading fidelity
high
Study strength
high
|
|
| AI can support poverty alleviation as a general-purpose technology provided robust governance systems and supplementary funding for human development are in place. Social Protection | positive | poverty alleviation (policy conditional claim) |
Reading fidelity
high
Study strength
speculative
|
|
| The results have significant policy implications for using AI to achieve Sustainable Development Goal 1 (no poverty) in developing economies. Social Protection | positive | policy relevance for SDG 1 |
Reading fidelity
high
Study strength
speculative
|