_akhbar.jpg)
Picture a section officer in a provincial finance department on a Monday morning. Her minister wants to know, by noon, how many pensioners in three flood-hit districts are still waiting for payments, and what it would cost to clear the backlog. Today that means letters, phone calls, spreadsheets in six formats and a week of reconciliation.
Now imagine her opening a government AI workspace, typing the question in plain Urdu and choosing a ‘Pensions Agent’. The agent already knows the pension rules and the latest notifications. It asks each district’s own system for current figures, combines them, and within minutes produces a draft briefing with every number traceable to its source. She checks it and signs it off. In this case, the machine does the chasing while the human makes the call.
Capable AI models already exist. What Pakistan has lacked is a safe, standard way to access authoritative government data without copying, leaking or misusing records. That is WASL, Pakistan’s National Data Exchange Layer, best understood as the nervous system for Pakistan’s AI.
An AI needs two kinds of knowing. The first is what the rules say: laws, gazette notifications, service manuals and the open data the state already publishes. A Pakistani AI can be built to look up this material before answering, like a careful clerk checking the manual, making it an expert in Pakistani procedure and languages. That is its library, and for that purpose, the Pakistan Digital Authority (PDA), under guidance from the Ministry of IT and Telecommunication (MoITT), has already issued the first Machine-Readable Legislation (MRL) and Policy as Code.
The second is what is true today about a particular person, plot of land or business. That information stays where it belongs, with NADRA, FBR, provincial revenue authorities and other custodians, and never enters an AI model. Instead, the AI uses WASL like a telephone: it asks the right office a specific question, with the required permission, and the office answers from its own system. WASL keeps no copy and does not retain the underlying record. It carries sealed envelopes it cannot open and records a tamper-proof receipt that the exchange took place.
This makes a bigger idea practical. Pakistan could host its own sovereign AI model, tuned for Urdu and regional languages, and offer government employees a marketplace of specialised agents built by departments, universities and Pakistani start-ups.
WASL could make this into a national market rather than a pile of pilots. Because every provider in a field serves data in the same national format, an agent built for Punjab’s land records works with any province following the standard. And every agent is accountable: WASL treats an AI like any other user, requiring it to prove its identity, obtain consent where personal data is involved, and leave a record of which model acted, for whom and why. There are no special AI back doors. The agents draft and recommend; officers decide.
The biggest gains, though, will be felt by ordinary citizens, through services offered by government departments and regulated institutions that plug into WASL, with the citizen giving consent through PAK-ID. Today, asking why your pension, domicile or benefit application is stuck usually ends with ‘come to the office’. An AI assistant on a phone line or WhatsApp, speaking Urdu, Punjabi, Sindhi, Pashto or Saraiki, could, with your permission, check the live record at the relevant office and tell you exactly where your case stands and what to do next.
A student applying for a scholarship or a family applying for support approves once, and the AI fills in verified details directly from the offices that hold them. No photocopies, no attestations, no second trip. With a patient’s consent, a doctor’s AI assistant could gather records from different hospitals, even across provinces, and summarise them before the consultation, so a patient in Muzaffargarh doesn’t have to carry a plastic bag of old reports to every appointment.
A small shopkeeper with no paperwork asks a bank’s AI what loan he could get. With his consent, it confirms his land holding and tax filing status and nothing more. Often a verified yes or no is enough: a subsidy check needs to know that a household’s income is below a threshold, not to read the tax return. That is how AI-based lending can reach millions of people the banks cannot see today, without turning their private lives into a commodity.
And sometimes the service comes to you. If a retired government servant consented to his pension office being notified of changes in his civil record, the registration of his death could prompt an AI agent to prepare his widow’s family pension case and call her in her own language to confirm her bank details.
Some questions are not about individuals at all, but about the bigger picture. What patterns can we spot across whole districts and entire populations? Where is dengue spreading fastest? Which districts lag on maternal health? WASL supports analysis in which each provincial department computes its own figures and shares only aggregated statistical results. The same approach points towards national AI models that learn from government data while the data never leaves its owners.
Consent rules, audit trails and data standards are not red tape around the exciting part. For AI, they are the exciting part. No hospital, bank or benefits office will trust an AI whose sources are unknown and whose actions cannot be traced. Under WASL, an AI cannot grant consent on a citizen’s behalf, cannot see anything a human officer in the same role could not, and can trace every action it takes. The groundwork for personal AI agents acting on a citizen’s behalf has also been laid, but it wisely stays switched off until policy settles who is liable when an agent errs and what stops it from overreaching.
Pakistan can remain a buyer of AI built elsewhere, fluent in many languages but ignorant of Pakistani life. Or it can put AI to work on its own terms, connected to its own authoritative data, speaking its own languages and answerable to its own laws. Models will keep getting cheaper. The lasting advantage is trusted, consent-based access to the facts only the Pakistani state holds. The model is not the moat; the connection is – and WASL is where it begins.
This is the third in a series on Pakistan’s digital public infrastructure.
The writer is the chief digital public infrastructure officer at the Pakistan Digital Authority and former managing director of the KP Information Technology Board.
