Why Billions of People Are Being Excluded From AI’s Benefits

Artificial intelligence now powers much of the modern economy. It verifies our identities, assesses loans, detects fraud and helps doctors read scans. All of it depends on one input: data. Large, reliable, representative data. In many parts of the world, that foundation is thin or missing.

For billions of people, the data needed to build effective AI systems is incomplete, fragmented or skewed. This isn’t just a technical challenge; it is a structural barrier that risks deepening inequality and determining who benefits from the AI economy and who is left behind.

Related: AI to Boost Global GDP by USD 15.7 Trillion, But Divide Widens

The data divide beneath everyday AI

Every AI system— whether it scans for fraudulent transactions, analyzes speech or predicts disease — depends on high-quality datasets, such as transaction histories, voice recordings and medical images. In advanced economies with decades of digitized records, these datasets are abundant.

Elsewhere, the picture is stark. According to a 2025 IMF analysis, economies with mature data ecosystems are capturing AI-driven productivity gains at roughly triple the pace of those without. This gap is widening, and if left unaddressed, it will accelerate the global economic divide.

The risks are not abstract. When models trained on data-rich environments are deployed in data-constrained ones, they often fail. A credit model designed for Singapore can reject qualified borrowers in Central Asia, not because of overly high repayment risk, but because of data drift. A diagnostic system trained on U.S. hospital records may miss critical signals in clinics where patient files are incomplete. In each case, people are excluded from opportunities not because of their capability, but because the data to represent them simply does not exist.

From scarcity to ingenuity

Waiting for perfect datasets is not an option. The more urgent task is to design AI systems that perform under conditions of scarcity.

A key emerging approach at the core is synthetic data generation. A subfield of AI established a decade ago, synthetics consist of realistic, AI-generated datasets designed to complement existing datasets, however limited those may be. Synthetic data generation embeds an advantage of enabling the training of models on rare or extreme scenarios — overcoming challenges with underrepresented data in historical records.

Another path is adaptive learning. Through techniques like transfer learning, models built in data-rich environments can be fine-tuned for new markets with far less local data. This makes it possible for institutions in emerging economies to benefit from global advances in AI without having to wait for decades of digitized records to accumulate.

Federated learning offers a third solution. Instead of pooling sensitive information into a single central repository, institutions can train models locally and share the insights, thereby preserving privacy while bridging fragmented data landscapes.

Together, these approaches make AI resilient even when the data is thin — and crucially, they ensure that progress is not confined to economies already ahead.

Building guardrails

Technology alone is not enough. Governments and industry must accelerate the digitization and standardization of core records, from credit histories to health files.

Equally important is designing systems that acknowledge their own limits. AI models should be engineered to quantify uncertainty and signal when human oversight is required. In finance, for example, an automated system should flag instances where its confidence is low, allowing a credit analyst to step in. In healthcare, a diagnostic model should highlight when incomplete inputs could undermine reliability.

Handled this way, scarcity can shift from being a barrier to becoming a catalyst for innovation.

Related: AI and the Global Economy: Growth or a New Divide?

The stakes

The global opportunity is enormous. But unless the data gap is addressed, billions of people risk being excluded from these gains.

We face a choice. Either we build AI models that serve only the data-rich, or we develop systems that extend opportunities for all. The second path will require creativity, investment and collaboration across borders. It will also require courage — to innovate beyond the current limits — and to design AI systems that are robust even in the absence of perfect information.

If we get this right, AI will not just accelerate economies that already work. It will extend access to finance, healthcare and economic opportunity to the billions still excluded. The technology shaping our future should be a force for inclusion. By rethinking how we build and deploy it in a world where data is often missing, we can turn scarcity into opportunity and ensure that AI delivers on its promise of equitable progress.