Since the November 2022 launch of ChatGPT, AI frenzy has taken the world by storm. We have seen AI improving month after month with the launch of newer models (ChatGPT 3.5, 4.0, 4o, etc) as well as similar language – and multimodal – models by other providers such as Google, Meta and Anthropic, among others. There is clearly a race and not just between big tech players around the world who see their traditionally secure revenue streams threatened by upstarts in this new technology.

Two major developments in the field have defied earlier predictions and will probably define how this revolution will proceed.

First, while most tech giants have launched models with low compute, designed for lower energy intensity and edge-computing, such as OpenAI’s 4o (Mini) and later versions of Llama etc, a Chinese AI company has recently launched DeepSeek – a model built upon ChatGPT – reportedly trained at an astonishingly low cost of $5 to $6 million, against hundreds of million dollars in development cost for the larger advanced LLMs.

Second, OpenAI has claimed that it has figured out the path to Artificial General Intelligence and has set its eye on Super Intelligence. If true, this has proved wrong experts and industry pundits who had predicted how fast the AI could grow, what it could deliver and how disruptive it could be for society.

What does this mean for the application of AI and its societal impact? What technologies – and applications – will be built and how deep will their impact be on the future of work and society at large? How will they impact countries such as Pakistan? Building AI models is still a really expensive business out of the reach of most developing countries; even many well-funded companies in the developed world, though the barrier is coming down.

The field of AI applications, however, is considerably wide. At least until a year ago, it had had a significantly level playing field. In around mid-2023, when I started looking at applying AI for education, I was pleasantly surprised to find most of the world at almost the same level of development. This is how it should have been. We were asking the same questions about potential applications and impact in Pakistan that the rest of the world was. The last time something like that could have happened was at the leading edge of the last technological revolution – the advent of the internet and ecommerce. It didn’t feel like this then.

A year and a half later, questions from the developed world are considerably more refined while ours remain stagnant. The last year or so has been quite disappointing with little or no impetus to properly adopting AI in Pakistan both in terms of government policy and corporate strategy.

Pakistan must focus on developing (and applying) AI capability in the domains where we find the greatest need to augment human capability.

On the domestic front, the priority areas can include education (shortage of skilled teachers, 25 million children out of school) and healthcare (shortage of doctors and specialists). In areas where there is surplus labour, an effort should be made to apply AI selectively to enhance the productivity of that labour rather than replacing it (in short-to-medium run).

On the export front, we need to undertake a massive re-skilling exercise to enable our export-focused workforce to be most effective and productive in the world where AI will augment and multiply skills and capabilities of humans. If we are unable to do so, we run the risk of massive disruption to the growing number of people associated with knowledge and IT exports, particularly freelancers active in activities such as content development, graphic design, social media, blogging, coding and the like.

In order to gain a foothold in this fast-moving AI revolution, Pakistan needs to do a number of things.

The use of AI by a big-tech or knowledge worker sitting in a corporate office or a teacher or a pupil in the United States cannot be the same as by one sitting in Islamabad or New Delhi.

First, it needs to systematically train its educators and trainers to upgrade the capacity of its educational and skill system to bring a larger number of people up to speed with this revolution. This needs to happen in multiple domains and at multiple levels, namely:

1) The number of AI PhDs and entrepreneurs who will provide thought and technical leadership to the entire AI enterprise from both an academic (research) and entrepreneurial (innovation) perspective, currently around 200, needs to grow around 500 in five years.

2) The number of mid-level software architects and ML/ AI experts who can quickly deploy cutting edge AI technologies and tools in operations and processes across a wide range of organisations, currently at 100, needs to grow to around 500 in five years, and around 5,000 in 10-15 years.

3) The number of low-level AI talent (programmers, coders) who will build local applications for AI use in the society and to catchup with how the rest of the world will do computer science, coding and technology development, currently around 1,000, needs to be pushed to 10,000 in five years, and 50,000 in 10-15 years.

4) Hundreds of thousands of users of systems and processes and creators of value within the society – doctors, lawyers, teachers, corporate employees, etc – need to upgrade their skills to use AI tools.

This will also require developing a knowledgebase around AI applications – through rapid experimentation, evidence generation and dissemination – particularly, around what works and what doesn’t specifically in our context to enable the people, organisations and the society to make informed choices that are likely to lead to desirable outcomes. Knowledge of what works and what doesn’t, in AI is quite rudimentary, at best. It will continue to develop but it will be context specific. The use of – and benefits from – AI by a big-tech or knowledge worker sitting in a corporate office or a teacher or a pupil in the United States (or Europe) cannot be the same as one sitting in Islamabad or New Delhi. The knowledge will have been acquired – at some cost – and then disseminated and applied across the board for countries to gain advantage from AI.

Pakistan might benefit from a massive AI awareness and skilling campaign to train tens of millions of people to become AI-aware citizens. In 2018, Finland launched a similar exercise to get hundreds of thousands to gather basic AI literacy. A 6-week (50 hour) online course called Elements of AI was launched and completed by more than a million people over a couple of years. Similar courses are now available online, free of cost. It may be possible to convert some of those into our local languages. Advanced options can be created and made available for those interested in going further.

Given the current state of affairs, the country doesn’t have much to offer. It must focus on intelligent adoption of AI to enhance productivity of the workforce, limited applications and solutions development for the export markets, and the export of AI literate workforce to the world. Each of these segments will require significant investments.

The AI revolution may well be the first and the last major technological revolution that could be within Pakistan’s reach, only if we choose to do the rights things in the right order, with the right emphasis, incentives and organisational structure in place.

The writer is a former member of the Planning Commission. He is currently, the managing partner of INNOVentures Global (Pvt) Ltd and co-founder of PakGPT, a platform to promote AI skills and education. He is accessible at athar.osama@gmail.com