Last year, the Ministry of IT and Telecom (MoITT) launched an Artificial Intelligence (AI) Policy meant for broader consultation with key stakeholders and the public at large. Although there is some confusion in the document regarding its status, whether this was a policy document already approved and ratified or merely a 'draft' for comments, the ministry's position (later) had been that it was the latter.
AI, today, is a fast-moving target that is rapidly advancing - shifting not only its speed but also its trajectory - and this requires not just a deep knowledge of the current states, opportunities, and challenges but also an adaptive approach to policy-making. Since November 2022, when OpenAI launched ChatGPT, AI has seen a rebirth and then burgeoning of interest not just from technologists, innovators, and entrepreneurs, but also from investors, and policymakers. Last year saw record fundraising and new venture activity focused on AI - both fundamental models and their applications - with global AI-focused venture investment exceeding any other sector and even many sectors combined.
Around the world, AI has been hailed as the "new electricity" - a once-in-a-generation (or century) foundational or platform technology that affects many, many others - that will lift the entire socio-economic landscape around it. Like electricity, AI is hypothesised to improve productivity across the board, not just in a single sector, leading to a multiplier economic effect on global and national economies.
AI policies and strategies around the world
In the last few years, a number of countries have also either launched their AI policies and strategies or re-calibrated past policies to account for the massive new interest in AI. India, for instance, had an AI strategy that was announced by NITI Ayog (the Indian Planning Commission) in 2018, which lays out that country's aspirations in the realm of AI. Similar policies exist for the United Kingdom (2021), Canada (2021 and 2024), Singapore (2019 and 2023), Saudi Arabia (2020) and many many other countries. A number of these policies define a very clear and unique pathway for these countries in the realm of AI. India, for instance, aspired to be the AI factory for the developing world (remaining 40% other than OECD and China, etc.). It doesn't aspire to be the world's leading country; it aspires to be the place where problems in the developing world can be solved using AI. Similarly, the United Kingdom aspires to be not the global leader in AI but the leader in Europe. Taking advantage of stringent AI regulation, the UK thinks it can attract AI talent who want to work in a relatively more relaxed regulatory environment than the EU. There is a certain degree of realism and precision in these visions and aspirations that is refreshing, and we will discuss this shortly.
Last year, Indian Prime Minister Narendra Modi announced a massive programme to take advantage of the global shortage of AI manpower in much the same way as it had capitalised on the shortage of manpower to deal with the Y2K bug. Working in concert with Microsoft and its Indian-origin CEO Satya Nandela, India has (once again) positioned itself as the go-to source of AI manpower for the global big tech. Singapore has also recently announced a programme to retrain its older workforce in using AI and other modern technologies, creating Masters degrees and other programmes to match older professionals (40+ years) and younger people so that the former can learn from the new skillset and mindset of the latter.
Enter MoITT's proposed AI policy
GOP/MoITT's attempt to draft an AI policy earlier last year was a step in the right direction. The ministry sought feedback from stakeholders and the general public. Many, including this author and colleagues, provided their feedback. However, instead of taking that feedback, engaging stakeholders, refining, and publishing the final draft, there has been an uncomfortable silence. All this while the fast-moving AI revolution continues to gain further pace and purpose affecting lives and economies around it.
Could an AI policy work to our advantage and help Pakistan do better to position us in this fast-moving revolution? It certainly can. However, a policy is better than no policy if it sets the course correctly, aligns everyone's interests, creates the rigorous foundation on which initiatives can be built, and it is carefully designed - with appropriate resources, metrics and milestones, checks and balances - to achieve the right impact.
Before we go further, let's look at the draft policy shared by the Ministry of IT and Telecom last year to see if we would have been heading in the right direction.
To start off, as stated earlier, MoITT's attempt to consult stakeholders and the broader ecosystem about a proposed AI policy is both laudable and welcome. However, its inability to take this feedback and execute the proposed policy even after over a year is akin to one step forward, two steps backwards. This is particularly more damaging because of the pace at which AI is moving and evolving in this very early stage of this current boom.
A closer look at MoITT's proposed AI policy draft also reveals several shortcomings and points of concern as a policy document. I will talk about at least five main categories here for reference, though one could take a fine-toothed comb to identify many more specific shortcomings and make recommendations that would be critical to making this a workable document.
A lack of focus and a clear direction
First and foremost, the proposed policy lacks a clear focus and direction. Unlike the Indian and UK policies cited above, Pakistan's proposed AI policy tries to do everything and be everything for everyone. The document, initially, clearly lays out two key strands of the policy, namely, the development agenda (aka embracing AI) and the responsibility agenda (aka appreciating human intelligence). But then it fails to properly elaborate upon this. It is possible that what the document meant by using these two themes was to say that Pakistan aspires to balance technology with regulation or use AI only for the benefit of humans, but nowhere in the document is this properly elaborated upon. The rest of the document follows a different structural logic and is organised around the pillars driving AI, not the two strands introduced earlier.
The proposed policies try to do too many things, including cultivating relationships with global entities and national governments to influence the direction of the AI revolution globally - quite an aspiration for a country that doesn't quite have even the basic infrastructure for AI. This clearly indicates a disconnect between where we actually stand and our exaggerated understanding of our self-worth. Even India - a many times more technologically advanced country than us - doesn't have the kind of aspirations to shape the global direction of AI even though they may have a much better chance of doing so. But that doesn't stop our policymakers from aspiring or penning a totally unrealistic expectation or aspiration in an official government policy document.
Our AI policy must take a very realistic account of what we are capable of and build a vision and aspiration of where can we get to within the foreseeable future. This aspiration must be clear, well-defined, and realistic; otherwise, it will become another one of those meaningful goals that we are used to setting for ourselves and not following through upon.
A public sector-driven policy posture
Second, the proposed AI policy takes a very public-sector-driven view of the AI opportunity for Pakistan. For someone who has experience seeing the capabilities of our public sector organisations as well as an appreciation of how (and where) AI is evolving, the reality of the AI opportunity cannot be further than that. I simply do not see any public sector organisation in Pakistan being able to show the initiative and speed to match how AI is progressing and what Pakistan needs to do to catch up. While development policy in Pakistan usually shows a bias against the private sector - as policymakers find it easier to trust the public sector entities - capitalising on AI opportunities will require a clear departure from this tendency towards creating an enabling environment for the private sector (and individual entrepreneurs and innovators) to succeed.
It is true that Pakistan does not have big tech, like the United States or even India, and our big IT companies hardly have the kind of research and development (R&D) capability that many of these companies have, but AI provides an opportunity for us to change this as it is also an enabler of sorts that reduces the technological distance that we have to catch up, particularly when it comes to building applications rather than foundational models.
Does this mean the public sector doesn't have a role to play in AI? Far from that. The public sector could provide the foundation on which a private sector AI capability could be built in three important ways:
- First, ensuring and providing access to public datasets, which is the lifeblood of the AI revolution.
- Second, IT infrastructure such as cloud and computing capabilities could be housed within the public sector, though it is equally important to incentivise the private sector to build these at a globally competitive level.
- Third, talent and human capacity building through universities and other public sector entities.
Above all, a national AI policy should aspire to incentivise and facilitate the private sector (and individuals) to create value while playing a facilitating role rather than entrusting the public sector to lead or do it all, which seems to be the primary focus of the current policy regime.
Challenging institutional arrangements
Third, the proposed AI policy puts bulk of the onus of funding AI initiatives onto Ignite - MoITT's Technology Fund - by setting aside a significant percentage of its budget for AI initiatives. That is a welcome step as it would provide continuity to funding AI in Pakistan. The remaining funding comes from the government-controlled Public Sector Development Programme (PSDP) via the Planning Commission, and this puts restrictions on the type of initiatives that can be funded, i.e. majorly for public sector organisations or universities. It's not an ideal situation but, perhaps, a good start if at least Ignite funding could be routed to the private sector.
However, this is where the good sense ends.
The proposed AI policy aspires to create a number of Centres of Excellence in AI and Allied Technologies (COE-AIs) across the country - three in the main cities and then another five in auxiliary cities - over the next three to four years. These COE-AI's will be designed as Public-Private Partnerships (PPPs) along the model of the National Incubation Centres and will be given a very wide mandate and responsibility including, but not limited to, becoming advocates for policy change and aggregators of national data. First of all, it would be a mistake to create additional infrastructure without evaluating how previous initiatives have worked. Has the government or the MoITT rigorously and independently evaluated the performance of the National Centres of AI, Robotics, Data and Cloud, and Cybersecurity, etc? Part of the mandate of these national centres was to create IP spin-offs from universities. Also, the performance of National Incubation Centres needs a thorough examination before creating more of the same. If these NCs/NICs seem to have worked, they should be tasked to take on AI as a special focus area rather than creating new centres from scratch. Why can't we make National Incubation Centres recruit 30% of their targeted startups in AI and achieve the above objectives quicker and much more cost-effectively?
Similarly, to expect these privately-run COE-AI's to secure data - most of which lies within the non-digitised government systems - is a non-starter and will be akin to pulling teeth out of an unwilling patient. Ministries usually treat data as instruments of control and one of the most carefully guarded secrets, which they are very hesitant to share, even with other government departments, not to speak of private entities like COE-AIs. It is going to be challenging even for Provincial IT Boards (PITBs) and the National IT Board (NITB) to aggregate data from across the government and the country at large. It may require an especially empowered entity with the full and complete backing of the PM Office to be able to do so.
Other institutional arrangements proposed in the AI policy also need a radical rethink. There is also a mention of creating a new AI regulator, which is not a good idea for three reasons.
- First, AI is a very nascent area and there is a lot of experimentation that will go on for a considerable period. Coming down too heavily on trying to regulate the technology will likely be counterproductive to this innovation.
- Second, there is a distinct strand of opinion worldwide (including in the US and UK) that we should adopt a "wait and see" approach to the regulation of this technology and regulate applications only when harm becomes obvious, and that too via existing regulatory arrangements.
- Third, it is hard to imagine a new regulatory body in Pakistan being created from scratch and developing the capacity required to properly regulate a fast-moving industry. Building capacity of existing regulators such as the State Bank of Pakistan (SBP), the Securities Exchange Commission of Pakistan (SECP), the Pakistan Telecommunication Authority (PTA), etc., while keeping regulation light and allowing experimentation is the right way to go.
Similarly, some other proposed institutions (such as a cloud and computational facility) are probably best suited for the private rather than the public sector.
The government should think very clearly and carefully before starting any new public sector institutions and only do so if no private alternative is possible. Instead of funding and creating inefficient perpetual bureaucracies, many of these contracts can be given against clearly defined KPIs and service levels and value can be realised.
Unrealistic and exaggerated targets galore
Fourth, the proposed policy makes an attempt to set a number of targets - some of these quantified as well - which is the right idea. However, a number of targets are unrealistic and exaggerated. Take AI manpower, for example. The policy aspires to have a million graduates in AI and allied skills within three years (2027) out of which one year has already passed. Further, this figure is several multiples of the current capacity of CS/IT departments of the country's higher education sector, and it's probably way more than the total stock of manpower of the IT industry. To produce a million graduates, the policy talks about creating 10,000 trainers - a number way out of reach in any sector of the economy, not to talk of a new and emerging area like AI.
The proposed policy also talks about creating an R&D culture in AI and Machine Learning. To do this, it puts forth a target of jump-starting "at least 1,000 AI-led R&D initiatives funded in academia and the private sector via fiscal and non-fiscal support" by 2026. Again, notwithstanding the fact that a year of this very ambitious target has already passed, the idea that 1,000 AI-led R&D Initiatives can (or should) be funded to address the gaps in our R&D culture is quite a non-sensical idea and reflects plainly how ignorant and out of touch the writer of the document is of the current state of national AI research capability and how quickly can it be built unless, of course, the objective is to just throw a big number in the air out of nowhere and never have the intention of following up on it or producing something of quality in the first place. Another similar number is the aspiration to file 2,000 patents in AI and related technologies by 2026.
Other similar targets include contributing 50 new AI models to open AI platforms annually (by 2024) and helping academia to register, apply, and train 150 unique AI models annually (by 2025). Clearly these are not just unrealistic and exaggerated numbers but also meaningless targets unless they could be justified with proper reasons and rationale for why we should do this in the first place.
Lack of evidence and performance orientation
Finally, most of the goals and targets identified in the document lack any evidence or performance orientation. Most "targets" are actually input measures achieved once money is disbursed and spent or straightforward "outputs," i.e., things that this money can buy. There is a clear dearth of "outcome" and "impact" metrics that require a lot more planning and effort to achieve than merely spending taxpayers' money and buying stuff from it.
Most of the "targets" identified in the AI policy are of this nature: a million graduates, 10,000 trainers, 1,000 AI projects, etc. and thus are susceptible to a mindless race to spend money rather than demonstrate the impact.
Also, take, for example, the three areas where the proposed AI policy puts a clear onus on AI to create a social impact. Here, there is a lot of vague language that does not clearly identify a goal and/or target for which the policy can be held accountable. In health services, for example, the policy talks about AI helping analyse clinical data, support primary care physicians, enable diagnosis, and make it easier for patients to self-manage. In education and learning, it talks about personalised learning and AI-enabled adaptive assessment. In agriculture, it talks about predictive analytics for the agricultural supply chain, weather prediction, soil monitoring, etc. None of these are specific enough for the policy itself to be held to account.
A bold, ambitious, and performance-oriented policy could try to ensure that every Pakistani has access to AI-enabled diagnostics in 10 years so that the dearth of specialist radiologists, pathologists, or oncologists in semi-urban and rural areas becomes a thing of the past. It could also make literacy available to everyone using a device and an AI-enabled tutor in 10 years, thus making out-of-school children a thing of the past. It could also improve agricultural productivity and reduce waste by 25%. Such targets could then become drivers of the change that the policy seeks to bring about and would mobilise all involved to demonstrate the impact of AI on the country and society at large.
Working backwards from the desired results toward the precise actions needed to achieve these will require disciplined thinking and a deep understanding of how things translate. It will require considerable effort, but will also bring accountability and a performance orientation to the policy exercise.
A policy or no policy?
The policy has a way of signalling a particular intent, guiding action in a particular way or direction, and focusing the energies of a wide array of stakeholders towards a consciously identified strategy. A badly designed policy does exactly the opposite and is way more wasteful and time-consuming than no policy at all. Not only does it not put the resources where they're most needed, but it also puts them (and the energies of all those who follow these resources) where they are not needed thus creating senseless initiatives, projects, and institutions that do not contribute to the desired course of action.
The national AI policy needs to be minimalist, reflect meaningful changes in the status quo or course of action, and leave the implementing ministry or agency to fill out the details. The policy must also be adaptive—particularly for such a fast-moving area as AI—and built on solid intellectual foundations and a set of assumptions that can be questioned, tested, and revised. If possible, the crisp and minimalist policy document must accompany the more comprehensive assessment and thought piece on which it rests.
An over-specified policy that is a hodgepodge of unrealistic and often meaningless numbers is likely to do more harm than good for Pakistan's aspirations to establish a foothold in the age of AI.
Any effort to try to create a sensible AI policy will certainly benefit from learning from the shortcomings of the previous draft.