Weaponising Quants: Can math make you a better war fighter?

We live in interesting times. Not only is history rhyming with the past, as nations shift their militaries towards near-peer competition, new technologies such as AI are being added to the verse.

According to SIPRI, 2.7 trillion USD was spent on defence globally in 2024, 37% more than in 2015. A closer look at these numbers reveals that spending is refocusing on equipment not intended to fight non-state combatants, such as ISIS or Al-Qaeda, but rather other national militaries. As Europe awakens to the threat posed by Russia, and China and America square up over Taiwan, things like tanks, fighters, ships, and others that focus on increasing “mass” are increasingly back in vogue.

The issue with near-peer competition, however, is that advantages in numbers and technological superiority are less pronounced. As anyone who has played a strategy game before knows, be it chess or an RTS, when both sides are equal, what matters most is making the right decision at the right time. Hence, the growing popularity with some in the Pentagon of a sci-fi sounding concept: using quants and AI to predict an adversary’s next move.

Formally known as Quantitative Analysts, quants traditionally use mathematical models to predict market behaviours. This enables them to make informed bets using data as esoteric as Nebraska’s weather to predict minuscule changes in price in the oil being pumped through pipelines. But with AI, their ability to process even vaster amounts of information has expanded significantly. Perhaps to the point that they might be able to use math to “guesstimate” the future.

Or, to put it in more dramatic terms, the hope is that mixing quants and AI will one day do for war at a strategic level what Deep Blue did for chess.

What Exactly is a Quant?

In the simplest, perhaps reductive, terms, a Quant is a math or computer specialist who creates complex mathematical models (essentially less concise mathematical formulas with adjustable parameters) to predict relationships between variables. Less of: if “x” happens, then “y” happens and more of: if “x” happens, then the probability of “y” is…..

While not exactly a flawless crystal ball, anything that can predict the future, even with a whole lot of caveats, is of unimaginable value. If you can accurately predict the fluctuation of stock or commodity prices, even by a minute amount, you can make serious money from it.

(It is also important to note that quants operate in several other fields, including healthcare and engineering; any field that requires making predictions with math requires a quant, but for this article, we will focus primarily on those who work in predicting human behaviour.)

To put this into perspective, Renaissance Technologies, considered by some to be the quant firm that started it all, is estimated to have achieved a 66% annual return over the last decade. The demand for quants is such that salaries can exceed $500,000. If anything, this is a good indication that they must be doing something right.

As for the day-to-day of a quant, it mainly involves staring at a computer screen, creating, perfecting, and updating their model. This is done by analysing past data to uncover patterns.

Quantitative analysis can only take you so far, though, and so at some point in the decision-making process, a qualitative analysis is usually incorporated with a qualitative judgment. This helps address the model's blind spots.

While this may seem divorced from the day-to-day of the ordinary soldier, the truth is that if you are predicting prices, you are predicting human decision-making behaviour. So, in theory, if you can mathematically predict decision-making, then you should be able, given you have all the information needed, to predict an adversary's next move during a conflict (at least, to a degree).

Join that with AI’s ability to automate using vast data sets and perform quick calculations, and these predictions can be sent to the commanding officer fast enough for it to be of use. After all, there’s no point in having a model that will tell you what the enemy will do tomorrow in a week.

Despite its benefits, AI is not without its limitations. Companies and governmental agencies still have to take a “trust but verify” approach towards it, and also ask who trained it? And what data has it used? If the data is bad, then the predictions will suffer. If the person who taught it has ingrained biased views, then maybe the predictions are incomplete. Nevertheless, it's undoubtedly an incredibly useful tool with applications across multiple fields.

In the commercial world, we have already begun to see companies like Seerist, which specialise in country risk, leveraging quants and AI, utilising open-source data (i.e., non-classified data accessible to all) to predict instability and country actions, including military ones. So, the next stage is to bring that speciality fully into the military domain.

Datapoints, Sensors, and Decision-Making Superiority

The obvious question is: how would such a thing work? Well, mathematics and warfare are already intertwined. From calculating the angles and climatic conditions for a sniper shot to developing new ICBMs, anyone in the business of war knows that someone must pull out a calculator sooner or later.

Furthermore, while certain aspects of warfare today would be utterly baffling to someone like Carl von Clausewitz, much of it remains remarkably the same, with a few tweaks - for example, exchanging horse rations for gas. This is all to say that while some things change, others remain constant, aiding anyone predicting enemy movements.

And where military strategies have evolved, a tremendous amount of effort from both the private and public sectors has been expended in uncovering and analysing them from all sides. To illustrate this point, a plethora of accessible journals, research papers, and books are available on the strategy, operations, and tactics that could be employed by the Chinese to take over Taiwan.

Finally, these days, the battlefield is littered with sensors from both military platforms, which are becoming more digitised, and commercial sensors as well. The F-35 can be said to be more than just a fighter with advanced computers and sensors; it is essentially a flying computer and sensor that just happens to be capable of engaging both air and ground targets. Furthermore, as militaries seek to enhance collaboration among their branches (e.g., Air Force, Army, Navy, Space Force), as well as a greater interest in commercial data gathering (example, social media), the sharing of data is expected to expand. Even at the soldier level, militaries are equipping them with more networked sensors than ever before, such as computers for the field. The speed of the digital age also means that, with enough processing power, this data can be gathered and acted upon in real-time.

Unfortunately, this also creates another issue on the modern battlefield: an overwhelming amount of data. According to an Australian publication from the Department of Defence Science and Technology on Intelligent Decision Superiority, 85% of potential high-value data collected from next-generation platforms and systems may never be looked at because they are manually processed. One can only imagine the amount of data created by a force the size of America’s or China's. With so much data, even deciphering what is useful and what is not is a Herculean task for quants, especially since one of their added values is finding non-obvious patterns across traditionally siloed information. This is where AI’s potential to automate much of this work becomes invaluable.

This is all to say that even a relatively well-equipped army that does not approach the levels of a superpower has access to a vast source of historical and current data points, which can, in theory, be used to develop models which would predict an enemy’s force behaviour. This, combined with the democratisation of AI, means that we may not have AI generals commanding humans into battle anytime soon…  But we’re likely to have field reports compiled with the help of AI, suggesting future manoeuvres.

Of course, predicting a competitor’s behaviour is part and parcel of the military commander’s job since war has been a thing. What militaries want to do with AI and quants is narrow the gap between certainty and uncertainty in the decision-making process. For those who believe in this, the data is available, and now the ability to process it is also broadly accessible. There should be no reason why commanders can’t achieve greater decision superiority. It's certainly not going to be easy, but for those who see their militaries facing other militaries in “their weight category” in the future, this may prove invaluable.

The holy grail would be a system that can predict the future with extremely high accuracy. This system would help a commander outmanoeuvre any force, perhaps to the point where success is so assured that it wouldn’t even make sense for the enemy to engage, a real-life checkmate. In this dream scenario, the future Wiki page for a war between China and America over Taiwan would cover not the day-by-day realities of a catastrophic fight… but the movements made in the run-up that ensured conflict never erupted.

What might this look like? We’re glad you asked.

What would a potential conflict using AI look like in the near future?

Picture the scene. The year is 2027, and you have been entrusted with the responsibility of maintaining Taiwan's independence. As commanding officer in the region, you receive daily reports on China’s activity.

These reports are compiled through numerous open-source analyses and real-time military intelligence. AI would be on the lookout for signals, which on their own don’t mean much, but together could be cause for alarm. These signals would be quantified and fed into an AI mathematical model designed to predict China’s next moves. This model would have been created by studying past data extensively.

One day, the AI reports that many social media sources have gone silent, and those few still functioning mention increased activity by the Chinese forces. The AI recommends raising the alert level and increasing vigilance as the probability of invasion has increased. Not wanting to escalate the situation, but still cautious, you request that additional resources be activated to monitor Chinese activity.

Five days later, the CCP announces further military exercises around the Taiwan Strait, leaving you to decide if this is a ruse for the actual invasion. Past trade data indicates the nation has enough stockpile of raw materials for an offensive. When combined with meteorological information, as well as intelligence indicating that enough fresh blood for the predicted casualties has been transferred to nearby hospitals, the AI raises the probability of invasion even further, while also providing a window of time on when it would take place. You are now compelled to act.

You request that your naval, air and space assets position themselves in strategic locations. During peacetime, such a move would have incredibly dire diplomatic consequences; however, the AI assures the probability of invasion is high, and multiple sources confirm this, making the manoeuvre worth the loss of political capital.

With your assets now in place, the task of successfully taking Taiwan suddenly becomes so much harder. With their own calculations having changed from “this is hard but do-able” to “this will be a catastrophe,” the CCP war planners back down. The invasion plan is called off. Essentially, you have protected Taiwan without firing a single shot. Sun Tzu would be proud.

To be clear, this is a gross oversimplification. A real AI system would employ much more complex data analysis and numerous additional variables. But the point wasn’t to sketch out a solid prediction for the future, but to demonstrate the variety of data one could examine to create such a system.

Unfortunately, as anyone who has used Chat GPT for their university reports and got a C minus has realised, we are still a long way from the above scenario.

While AI is certainly playing an increasingly significant role in areas like optimising logistics, analysing the global stage is another league entirely. It took about a decade to progress from an AI that plays checkers to one that plays chess, and two decades for the same achievement to be made with the Go strategy game. The primary reason for the difficulty is that Go is highly complex at a strategic level due to the vast number of possible moves. And Go still has strict rules, whereas war and global politics do not. One doesn’t need to look further than Ukraine’s recent attack on Russia’s strategic bomber fleet with hidden drones - or Israel’s pager attack on Hezbollah - to be reminded that all is fair in love and war.

But there’s a vast gulf between the utopia of a war that never has to be fought and a world where quants and AI can work together to give your forces a battlefield advantage. It’s this latter outcome that militaries are trying to work towards today, a process that’s as difficult as it is potentially rewarding.

Stocks to Bombs, from Quant to Military Data Scientist

The layman view of AI tends to fall into two categories: either an unknowable superbrain that’s so powerful it effectively acts as magic, or else as an overhyped nothingburger that can’t do anything without human helpers.

As usual, the truth falls somewhere in the middle. Even with AI, quants collaborating with the military are going to have an unholy amount of work to do. They will need to gather and quantify massive amounts of historical data, as well as collaborate with engineers to establish methods for collecting data from a wide variety of sources in real-time.

Some of the difficulties will also include translating qualitative data into usable quantitative data for processing, and then back into recommendations to present to someone who may not understand all the mathematical concepts behind it, and therefore needs the results retranslated into plain old English. More likely than not, this would also involve combining numerous mathematical models with even more data sources. And this doesn’t even touch the parameters set by ethical and political considerations of creating such a system, nor the challenges posed by reviewing the answers such a system may give and working backwards from it.

The amount and type of work are such that a new kind of quant is needed. One that acknowledges the rather large gap between modelling price fluctuations and military decision-making. As a result, we shift from Quantitative Analysis to Data Science.

The overlap between quants and Data Scientists is significant, as both utilise mathematics to decipher complex relationships between values, providing actionable insights. Data scientists, however, conventionally work on a much broader spectrum with more complex variables than just solving financial fluctuations, and require a PhD, unlike a Quant. Add a military focus, and the result is a military data scientist (MDS).

Typically, the title "Data Scientist" is a catch-all term. Their responsibilities can range from collecting data to utilising AI and deep learning. Similar to more conventional roles, the smaller the organisation, the more hats individuals wear and the less specialised their tasks are, while larger organisations can afford to have more people do more things and specialise. In an organisation such as the Army, a data scientist would also work together with engineers, developers, and others. Although still very much someone who uses quantitative analysis, it is essentially on another level.

And it will need to be, because competitors will do everything in their power to negate or mitigate the other’s decision-making capabilities.

The Destructive Power of Imagination in an AI World

It would be naïve to think that most militaries aren’t trying to leverage AI to increase their chances of winning wars. Recognising this, it would be equally naïve to believe that they are not pulling significant resources to find ways to nullify or mitigate the advantages AI brings.

While trying not to turn this closing chapter into a thesis on warfare intelligence, there are broadly three ways to mess with an AI: deny or corrupt the data it relies on, attack its systems, or be unpredictable.

The first two ways are pretty straightforward. An AI will only be as good as the information fed into it. From complex methods such as using deepfakes and electronic warfare to something as simple as lying on national television, there are numerous tools one can use to degrade an AI’s effectiveness. And if you can’t deceive it, then you can try and break it, with either a physical attack on its servers, an electronic attack on its sensors or a cyberattack against its software.

But if you really want to outsmart an AI, then perhaps the most effective thing you can do is be unpredictable. To not think just outside the box, but so far outside it that the box is barely a speck on the horizon. Pearl Harbour, September 11, and the recent Ukrainian drone strike on Russia’s strategic bomber fleet were all planned by people who were thinking far beyond the limits of what their opponents believed was possible.

With AI, the same rules apply; to beat it, you will have to think outside its “box”. Until AI can develop an imagination not bound to past information and probability, it will be unable to foresee such black swan events. Ironically, given the number of writers and artists AI may soon put out of business, it could be old fashioned human creativity that future militaries come to value the most.

In summary, if you are a Data Scientist wanting to leverage the incredible power of predictive mathematics for warfighting, rest assured that militaries around the world will also be equally devoted to ensuring that you spend many sleepless nights updating your model because they created a variable you didn’t think about.

Welcome to the new arms race, soldier, I hope you like math and coding!

Sources

• blogs.icrc.org
• citeseerx.ist.psu.edu
• interestingengineering.com
• interestingengineering.com
• militaryembedded.com
• sdi.ai
• seerist.com
• thedefensepost.com
• thiele.ruc.dk
• ukinvestormagazine.co.uk
• armyupress.army.mil
• businessinsider.com
• fnc.co.uk
• forbes.com
• institutedata.com
• investopedia.com
• investopedia.com
• ipforum.net
• sipri.org
• youtube.com