AI must learn to speak the language of the people

My recent engagements with the global AI community, most recently in Seoul, South Korea, have left me reflecting on a fundamental question: as artificial intelligence advances at extraordinary speed, are we making it accessible to the people who stand to benefit from it most?

At the Global AI & Digital Summit, hosted by the World Bank Group alongside the Government of the Republic of Korea, I had the opportunity to engage with policymakers, technology leaders and development partners on the future of AI and its potential to transform economies, governments and public services.

South Korea offers important lessons for countries like Pakistan. Its transformation into a global technology leader demonstrates what sustained investment in digital infrastructure, human capital and innovation can achieve. But beyond the technological advances, my biggest takeaway from Seoul was the importance of ensuring that this new wave of innovation reaches everyone, including those who have historically been left behind.

Artificial intelligence may be the most transformative technology of our generation. But for billions of people, there is still a fundamental barrier between its promise and its usefulness: AI does not yet adequately understand the languages they speak.

Think about a farmer in Punjab asking for agricultural advice in Punjabi. A mother describing her child’s symptoms in her local dialect. Or a student trying to learn a difficult concept in the language spoken at home.

If AI cannot understand their words, accents and local context accurately, how inclusive is it really?

The world is beginning to ask the right questions

This question is increasingly central to global conversations about the future of AI.

At the United Nations General Assembly, governments adopted the Global Digital Compact, calling for digital technologies and AI systems that protect linguistic and cultural diversity and reach people currently excluded from the digital economy.

In September, a global coalition of 60 organisations announced an ambitious five-year goal: to enable an estimated 3.4 billion people who speak languages underrepresented in today’s AI systems to use AI in their own language and voice.

The message emerging from these global conversations is clear: AI cannot be considered successful if its benefits remain concentrated among those who already have access to technology, education and opportunity.

The next frontier of AI is not just making it more intelligent. It is making that intelligence accessible to everyone.

Language is the missing infrastructure of AI

We spend enormous amounts of time discussing computing power, advanced models, data centres and investment. All are important. But we often overlook something equally fundamental: language.

Estimates suggest that more than 90 percent of the data used to train early large language models came from English-language sources. This matters because language is not simply a collection of words that can be translated. It carries culture, context, local knowledge and the way people understand and describe their world.

An AI system may speak grammatically correct Urdu or Punjabi and still misunderstand a health question, a local agricultural problem or a regional expression.

Translation is not the same as understanding. And accessibility is not the same as inclusion.

Making AI truly multilingual requires more than adding languages to a platform. It requires local language datasets, models tested against regional dialects, and the involvement of researchers, linguists and communities who understand the people these systems are meant to serve.

It also requires accuracy, privacy and accountability. Especially in healthcare, education and public services, a fluent but inaccurate answer can do more harm than good.

What this means for Punjab

Under the leadership of Chief Minister Maryam Nawaz Sharif, our ambition is to make Punjab the most AI-enabled province in South Asia by 2029.

But our success cannot be measured simply by the number of AI applications we launch or the sophistication of the technologies we adopt.

It must be measured by how many lives we improve.

For a province of more than 130 million people, language is central to that ambition.

At the Office of AI Punjab, we are already putting this thinking into practice.

In partnership with LUMS and with support from the Gates Foundation, our Office of AI Delivery Unit is piloting Awaaz-e-Sehat, an AI-powered maternal health voice assistant designed to help women access maternal health information and guidance through conversations in their own languages and local dialects.

The ambition goes well beyond translating medical advice into Urdu or Punjabi. A woman in a rural community may describe her symptoms using expressions, colloquialisms or traditional terms that are very different from formal medical terminology. The same symptom may be described differently across districts, dialects and communities.

For AI to be genuinely useful, it must understand not just what a woman is saying, but what she actually means. That is precisely what we are seeking to test through Awaaz-e-Sehat: whether AI can understand local speech patterns, interpret the medical meaning behind everyday expressions, and provide reliable, context-sensitive maternal health guidance, while recognising when professional medical attention is necessary.

This is particularly important for women who may have limited literacy, limited access to doctors or little experience navigating digital services. If we can make AI work for these women, in the language and context they understand, we can begin to demonstrate what genuinely inclusive AI looks like.

Our broader work through the Office of AI and its AI-enabled Delivery Unit is built around this same principle: technology must solve real problems for real people.

Language inclusion also represents an economic opportunity. Pakistan should not simply become a consumer of AI developed elsewhere. We have an opportunity to contribute to developing AI that reflects our languages, knowledge and local realities, while creating opportunities for our researchers, universities, technology companies and entrepreneurs.

An innovation designed to solve a problem in Punjab could offer lessons and solutions for millions of people across other developing countries facing similar challenges.

The real test of inclusive AI

My time in Seoul reinforced that countries like Pakistan have much to learn from the technological progress of nations like South Korea. But we also have an opportunity to contribute something important to the global AI conversation: how to make AI work for the diverse linguistic, social and economic realities of the developing world.

The global conversation is shifting from what AI can do to who AI can do it for.

That distinction matters.

Governments, technology companies and international development partners must work together to ensure that AI reaches people who have historically been excluded from technological progress.

Making this ambition a reality will require investing in local language datasets, supporting Pakistani innovators, and bringing universities, researchers, industry and government together. Our engagement with countries like South Korea and international development partners provides an opportunity to learn from global experience while building our own capabilities. The diversity of our languages and communities should not be seen as a barrier to innovation, but as an opportunity to develop solutions that could eventually serve millions of people facing similar challenges around the world.

But inclusion must also be built on trust. An AI system that speaks someone’s language but misunderstands their needs, provides inaccurate advice or mishandles personal information is not truly inclusive. As governments introduce AI into essential services, we must ensure these systems are tested across local dialects, protect people’s privacy, acknowledge their limitations and allow access to human professionals when needed. Trust is not built through technology launches or impressive demonstrations. It is earned when technology consistently proves useful, reliable and worthy of people’s confidence.

For Punjab, our responsibility is to turn that ambition into real applications that people can access, understand and trust.

We cannot allow the AI revolution to create another divide between those who speak the language of technology and those who do not. The true measure of AI progress will not be how powerful our models become, but how many people they empower.

And for that to happen, AI must learn to speak the language of the people, not the other way around.

Copyright Business Recorder, 2026