Are Countries Really on Track for Sustainable Development? AI Exposes Warning Signs

Some countries are improving on SDG scores while falling behind on human development. A deep-learning study says global sustainability rankings may be missing that split.

The global race to meet the Sustainable Development Goals (SDGs) is usually tracked through scorecards, rankings and annual reports. New research suggests those tools may miss countries that appear stable on paper while their deeper development path starts moving in an unexpected direction.

The study, titled Forecasting National Sustainability Trajectories with Deep Learning: Predictability, Surprise, and Early Predictive Signals and published in Sustainability, introduces an AI-based forecasting model that compares expected and actual national trajectories, identifying where countries are on track, where progress is exceeding expectations and where human development is slipping behind.

AI forecasting adds an early-warning layer to sustainability tracking

The study addresses a major gap in how global sustainability is monitored. Most systems show where countries stand today, but they do not clearly identify whether countries are likely to stay on course, accelerate or fall behind. The authors tested whether past development patterns can predict future sustainability outcomes and whether deviations from those forecasts can reveal early signs of instability or unexpected progress.

Using 749 World Development Indicators across 184 countries and regions from 2003 to 2022, they applied a Temporal Fusion Transformer, an interpretable deep learning model designed for time-series forecasting, to predict two widely used development measures: the Sustainable Development Goals Index (SDGI), which captures broad goal-level progress, and the Human Development Index (HDI), which captures the foundations of human development.

The SDGI measures performance across the 17 Sustainable Development Goals on a 0 to 100 scale. The HDI measures core human capabilities through health, education and income on a 0 to 1 scale. The model was trained and validated on data from 2003 to 2017 and tested on the 2018 to 2022 period. It achieved a mean absolute error of 1.10 for SDGI and 0.008 for HDI, outperforming linear trend and XGBoost baselines. The error reduction was at least 19 percent for SDGI and 60 percent for HDI, showing that deep learning can improve sustainability forecasting, especially for human development.

The findings also show that many national development paths are highly persistent. A large share of countries continued moving along expected trajectories, suggesting that institutions, investment patterns and policy choices create long-term momentum that is difficult to change quickly.

SDG progress and human development are diverging in some countries

The study found that 115 of 184 countries and regions, or 62 percent, were on track for both SDGI and HDI during the test period. These countries stayed within the model's expected range for both goal-level sustainability and foundational human development. Among countries that deviated from expectations, the pattern was uneven. Thirty-five countries and regions showed positive SDGI surprises, meaning they performed better than expected on the SDGI. Many of these cases were developing countries in Sub-Saharan Africa and South Asia, including Benin, Togo, Rwanda, China and Ethiopia.

On the other hand, twenty-three countries and regions underperformed their expected HDI paths, with negative deviations concentrated among countries affected by conflict, economic instability, health-system strain or wider disruption. Venezuela, Libya, Lebanon, Iran and Ukraine were among the major negative HDI surprises. This split suggests a decoupling between goal-level sustainability and capability-level sustainability. A country may improve on policy-sensitive SDG indicators while health, income or broader human development weakens.

The study found nine double-surprise countries, where both indices moved unexpectedly. Six showed positive SDGI surprises but negative HDI surprises. That pattern indicates that institutional reforms, environmental targets or other SDG-linked indicators can advance even as life expectancy, income or basic human welfare comes under pressure.

Ukraine demonstrates this divide. Its HDI underperformance coincided with war-related disruption, while SDGI performance may have been supported by institutional reforms and alignment with European governance standards. The United States showed a different version of the same pattern, with human development pressure linked to falling life expectancy during the pandemic period and the opioid crisis, while clean-energy policy may have supported SDGI performance.

The model also identified different predictive signals for the two indices. HDI forecasts were dominated by economic indicators, including fuel exports, poverty gap, petroleum rents, private consumption growth and lending interest rates. SDGI forecasting required a wider mix of economic, social, environmental and institutional indicators, including education access, drinking water services, foreign direct investment and control of corruption.

Why it matters for policy and global monitoring

The findings show that sustainability monitoring should not rely only on annual rankings or static dashboards. Forecasting can help governments and development agencies identify countries that are beginning to move away from expected paths before those shifts become entrenched.

The study also warns against treating SDG progress as proof of broad human development recovery. If a country's SDGI improves while HDI weakens, policymakers may be seeing progress in policy indicators but setbacks in people's lived conditions. Recovery assessments after shocks such as pandemics, wars, debt crises or energy disruptions should therefore track both goal-level progress and foundational human capabilities.

Governance, as identified in the study, is a major factor in predictability. Stronger governance was associated with lower forecast error for both SDGI and HDI, suggesting that countries with stronger institutions tend to follow more stable and predictable sustainability paths. Weaker governance, conflict exposure and instability make development trajectories harder to anticipate.

Countries with rising forecast errors may need closer monitoring because unpredictability itself can signal institutional stress. For development partners, this could help identify where support is needed before conventional indicators show a full deterioration.