Skip to content Skip to footer

Pakistan Air Force Delves into Artificial Intelligence Development

On 27 August 2020, the Pakistan Air Force (PAF) announced that it formed an in-house ‘Centre of Artificial Intelligence and Computing’ (CENTAIC) to spearhead development in the field for both military and civilian purposes. In a statement, the PAF Chief of Air Staff (CAS), Air Chief Marshal (ACM) Mujahid Anwar Khan, said that CENTAIC will help the PAF incorporate artificial intelligence (AI) into its operational domain.

The scope of CENTAIC’s activities are not yet known to the public, but it may be researching into a number of key fields in AI, such as big data, machine learning, deep learning, predictive analytics and, potentially, natural language processing (NLP). Not only do each of these sub-fields apply to current and emerging air warfare applications – notably drone development – but the PAF will need to draw on these areas to drive its next-generation fighter aircraft (NGFA) aspirations. The creation of CENTAIC was inevitable.

The PAF had noted that CENTAIC could support the development of civilian applications. However, as the entity’s work matures in the way of aiding the PAF’s programs, CENTAIC may ultimately specialize in the domains directly relevant to the PAF. But the underlying infrastructure and knowledge/capacity creation could scale-out and result in sister-organizations – ideally both in the public and private sectors – to drive AI development for healthcare, manufacturing, resource management, and other areas.

However, in terms of specifically the PAF, the creation of CENTAIC could help the PAF indigenize many key – but non-tangible and less appreciable – inputs for its NGFA under Project Azm.

These inputs can range from algorithms for  guidance systems for air-to-air and air-to-surface missiles, image processing for TV/IR seekers, sensor fusion, human-machine-interfaces (HMI), and other applications. While not as physically tangible as gas turbines or transmit-and-receive modules (TRM) for an active electronically scanned array (AESA) radar, AI research could eliminate barriers to outcomes that are simply unavailable to Pakistan.

Sensor fusion one such example. There is no off-the-shelf solution (independent of physical subsystems), so the PAF would either have to acquire it as-is from a willing supplier (which may carry limitations, such as requiring it to buy specific hardware), or develop it internally.

Likewise, Pakistan could also develop a myriad of intellectual property (IP) in AI, which may help it enter research and development (R&D) partnerships as an active contributor instead of a passive bystander. For a country that lacks industrial inputs, AI could emerge as an ‘equalizer’ for R&D growth.

AI is Essential for Future Air Combat Applications

In this article, we will refer to AI as an overarching field that includes the following sub-areas:

Big Data: This is the process of identifying trends, correlations, and other insights from disparate sources of information. The intended outcome of big data is to improve decision-making.

The PAF can draw data from a variety of sources, such as – among others – instrumentation equipment installed at the Sonmiani Weapon Testing Range (WTR), the Damage Tolerance Analysis/Structural Health Management (DTA/SHM) tools used at Pakistan Aeronautical Complex (PAC), or Air Combat Maneuvering Instrumentation (ACMI) pods, specific sensors within its aircraft, flight sorties, exercises, and other areas.  Additionally, a little-known feature of the JF-17 is its heavily instrumented nature. As a result, PAF has thousands of hours of flight data from the JF-17. The establishment of CENTAIC may be driven by the desire to utilize this goldmine of flight data.

In turn, the PAF can use the data to help it plan deployments, lower maintenance costs, cut maintenance downtime, and many other discrete outcomes for the service. Big data may also help with development by helping with design through the analysis of data from wind-tunnels and instrumented ranges.

Machine Learning (ML): ML is a form of AI that tries ‘learning’ from past data and results so as to make better decisions in the future. ML involves the use of algorithms to understand the data (which may come from the sources we listed above) and, in turn, help the planner make an informed decision. While it may sound like common sense, seldom do humans have enough time to manually read and understand all of the facts of a situation. So, inevitably, humans could end up compromising or “going with their gut” in the absence of information. In contrast, ML can do the heavy-lifting analysis across gigabytes or terabytes of data quicker, so it leaves the human decision-maker with fewer blind spots or uncertainties. Examples of application of ML include electronic warfare, online emergency flight plan generation, flight-model learning from flight data, adaptive flight control systems, and online optimal guidance for munitions.

Deep Learning: This is a sub-field of ML. It involves running different programs through different layers of data to pull conclusions or insights from several sources. Basically, where ML would use algorithms to interpret the data and help a human decide, deep learning tailors algorithms into ‘artificial neural networks’ that could make decisions autonomously. In other words, deep learning can enable for drones that can work on their own with limited human input or control. It may even help with the development of new generation terminal-stage seekers meant to attack moving and/or hidden targets. Deep learning will enable the operation of drone swarms and may lead to decision making AIs in loyal wingman drones.

Predictive Analytics: This is basically the use of data to determine what is most likely to occur next. One good example is aircraft maintenance. The PAF can use its existing inventory/requisition logs as well as its sensors on-board aircraft to set-up predictive maintenance and preventative maintenance schedules. In turn, it can reduce long-term repair costs, cut down-time, increase availability rates, especially for the JF-17.

Natural Language Processing (NLP): NLP enables computers to communicate with people. This might not seem important in an aviation context today, but it will be in the future. For example, a single NGFA could generate vast amounts of data from a dozen major sensors. If, for example, an enemy aircraft is trying to ‘paint’ the NGFA by radar, the NGFA could alert the pilot with a specific light or screen. This may be doable if that was the only information the pilot needed, but the pilot may also want to know about the current status of his/her ‘loyal wingman’ drones, or when to send a loitering munition into a terminal stage. The pilot has limited time and energy to read a giant screen of alerts. With NLP, the NGFA could converse with the pilot, thereby cutting the risk of information overload.

Potential Marquee Air Warfare Applications

Though the projects at CENTAIC may be at their early stages, one can see the PAF direct its AI/ML efforts the following areas. These projects can tie into its NGFA efforts both directly and in complementary ways.

Unmanned Combat Aerial Vehicles (UCAV)

Be it ‘loyal wingman’ drones or deep-strike aircraft, future UCAV employment will require varying degrees of autonomous operation. In fact, Pakistan may already have a basis to start developing a ‘loyal wingman’ UAV by leveraging its existing cruise missile and target drone technologies. These drones would still be a significant step-up from air-launched cruise missiles (ALCM) in terms of size, flight control technology, and engines, but it would be the next step in this line of development for Pakistan.

However, the lack of a concerted AI effort would have precluded the development of loyal wingman UAVs – CENTAIC may alleviate this issue. Something with the capabilities of the Kratos Valkyrie may not be on the horizon in the near-term, but simpler solutions, such as an electronic countermeasures (ECM) decoy, electronic intelligence/radar emission finder, or dual-loiter-and-decoy munitions could be a starting point.

In fact, these aircraft can serve as the basis for developing the first generation of AI inputs. It would be a prudent idea for the PAF to start these projects as soon as possible if it intends to deploy armed UCAVs in the loyal wingman and long-range strike roles in the 2030s.

Cruise Missiles

Pakistan can also start looking at adding AI to its cruise missiles. In iterative terms, the improvements can come into play in newer versions of existing designs, i.e., Babur, Ra’ad, and Harbah-series. AI could assist with managing modes, e.g., the point at which a missile should transition from its INS/GPS guidance suite to its terminal-stage seeker. This could allow for loitering over a target area, or even enable a system like the Ra’ad to intelligently dispense bomblets over an area (or operate as an ECM decoy) by utilizing image processing ML algorithms to recognize specific targets.

The PAF Must Set an Example for Leadership

Establishing CENTAIC as an entity focused on AI development is a positive step, especially if it leads to the creation of the requisite technology and training infrastructure. However, in the long-run, CENTAIC cannot be Pakistan’s sole driver of AI efforts. It will not be long before CENTAIC largely specializes in creating the solutions the PAF needs, but that would detract the entity from its dual-military-and-civilian focus.

In time, the PAF will eventually have to branch the latter work into distinct entities, ideally with the private sector taking the lead on most civilian organizations. However, the PAF could scale CENTAIC infrastructure – especially in terms of capacity development and physical hardware– to support private endeavours.

Universities are an avenue ripe for investment in AI and computing as they can often provide the required large datasets and large numbers of engineers needed for research into these areas. This will allow CENTAIC to offload a lot of their basic research work to universities and focus more on application specific work. This is on top of the significant benefits this model would provide to universities in terms of research funding.

Show CommentsClose Comments

Leave a comment