Electricity did not end poverty. It amplified the fortunes of those already positioned to benefit while leaving others even further behind. Artificial intelligence (AI) may follow the same logic.

A recent report by the United Nations Development Programme (UNDP), titled The Next Great Divergence, draws exactly this historical parallel. Just as the Industrial Revolution lifted the West while much of the world lagged, AI now offers a new inflection point. It can widen opportunities or deepen the divides that already run through Malaysia’s development geography.

At the heart of UNDP’s argument is a governance question: who does AI serve, and who does it leave behind? As Malaysia mid-sprints towards an AI-driven economy by 2030, it is a question we must sit with. AI can worsen inequality through three mechanisms already visible in the Malaysian context.

The first is the data desert. AI models are trained mainly on people most visible in digital records, typically urban, young, male, educated and connected groups. Those absent from the datasets are the likeliest to receive skewed, unfair outcomes. In Malaysia, these fault lines are easy to predict.

If models are trained primarily on Bahasa Malaysia and English datasets, they will misread speakers of Iban in Miri and the Bajau language in Tawau. A rural patient may be overlooked in healthcare triage. A local entrepreneur may be denied credit based on flawed scoring. A community may be erased from the mapping of economic vulnerabilities.

To be invisible to an algorithm is a new form of deprivation. It is not unlike lacking an identity document in a bureaucratic system. Malaysia has one of the most sophisticated identity frameworks in the world, anchored by the MyKad. It is a powerful key. But a key only works where the lock was built to accept it.

The second mechanism is infrastructure asymmetry. Communities that lack computing capacity, reliable connectivity and technical talent cannot build AI that reflects their own priorities. They import models shaped by other realities, calibrated to conditions not their own and, hence, do not see them clearly.

The third mechanism is techno-solutionism. Top-down AI deployment tends to prioritise efficiency metrics over inclusion outcomes. In the process, it often sidelines the very communities it is meant to serve. Worse, it frequently externalises its costs onto water resources and fragile ecosystems.

While efficiency, inclusion and sustainability are not always at odds, they are rarely aligned by default. Without deliberate design, the system tends to serve the most connected and leave the rest behind.

These three mechanisms converge. AI can exclude people and strain natural resources at the same time, which means governance must address inclusion and sustainability together. For decades, environmental thinkers have used the IPAT equation (Impact = Population × Affluence × Technology) to make a basic point: technology is never neutral. Its environmental and social outcomes are dictated entirely by how it is governed.

If AI primarily rewards urban, connected elites while making marginal communities “harder to see” in public services, and if it relies on energy-hungry data centres in a water-stressed state, the system is not delivering progress. It is delivering failure on multiple fronts simultaneously.

AI that widens inequality is, by definition, unsustainable. When the social contract fractures, the political economy of reform becomes far harder.

None of this argues against Malaysia’s state-led AI architecture. The Sovereign AI Cloud, the MED4IRN council and the National AI Roadmap represent genuine strategic foresight. The government has correctly identified AI sovereignty as a national imperative. The question is whether infrastructure sovereignty and inclusive design are being pursued with equal urgency. They are related but distinct, and treating one as a proxy for the other is a costly mistake.

The parallel that comes to mind is Huzhou, a Chinese city that has drawn attention for building a practical green finance system. It did not succeed just because the state built infrastructure. It succeeded because it has an inclusive digital infrastructure that allows more small and medium enterprises to be seen, assessed and financed by capital. Malaysia’s sovereign AI push matters for the same reason. Capacity helps, but inclusion has to be designed.

So what would inclusive AI governance require in Malaysia? The technical answers are within reach. The harder part is political will and implementation. Meanwhile, UNDP’s engagement with the government is emerging and is already pressing the right questions about who AI will see and what rules should govern testing and deployment. The next step is to turn those questions into standards.

First, national AI training datasets must include minority languages and rural contexts. Recent UNDP initiatives in the Philippines and Cambodia show what becomes achievable when survey and geospatial data are combined to pinpoint multidimensional poverty and service gaps. That kind of diagnostic work should now inform procurement specifications in Malaysia. It must identify where the algorithmic blind spots fall, so systems do not take connectivity at face value.

Second, social protection for gig workers should not come after AI adoption. It should come before it. Ongoing efforts in Malaysia are already raising ethical and AI governance questions, including who bears the transition costs and how early the safety net must be put in place. UNDP’s role in digital innovation and social protection can help connect technological change to fairness in the real economy, not just efficiency on paper.

Finally, the Sovereign AI Cloud must adhere to strict sustainability standards. Given Malaysia’s water constraints, energy efficiency is not an environmental concession. It is a condition for the long-term viability of data centres. UNDP’s work on development and climate in this region makes the same point from a different angle. Infrastructure that ignores resource limits does not become more sustainable over time. It becomes more fragile.

The next great divergence is not inevitable. It is a choice, and our standards will shape what inequalities get baked in. The original divide of the Industrial Revolution was the accumulated outcome of thousands of policy, investment and governance decisions made over many decades, including many that were never made at all. Similarly, the decisions being made now about datasets, community consultations and resource constraints will compound in ways that are hard to reverse.

The readers of this column are the investors deciding where to deploy capital in the AI stack, the executives choosing which AI vendors to trust, and the policymakers writing the specifications for national roadmaps. The divergence will not arrive out of the blue. It is built here, from decisions made in boardrooms and ministries. Writing an algorithm that sees everyone is a choice Malaysia can still make.

Originally published on: The Edge Malaysia.