Invisible Combat – The journey into the emergence of machine intelligent processes in contemporary conflict

7 Min Read
Drawing on two decades in oil and gas exploration and defence, Elizabeth Vallance-Bull traces how machine intelligent processes have moved from finding hydrocarbons deep underground to finding targets on the modern battlefield. She argues that data, code and AI now act as surrogates in warfare, that defence lags adjacent industries by decades, and that precise language must shape military requirements.
LVB
Elizabeth Vallance-Bull
- Co-Founder
| black and white consulting

This story starts in 2002, not in defence and national security, but in oil and gas exploration. A seismic dataset belonging to BP, terabytes in size from offshore West Africa, had just been delivered to the office of a company where I was employed as a geophysicist.

The aim of our work was to find a target – a reservoir, oil bearing sands – and establish their extent, then to drive ‘oil in place’ calculations, and ensure the economic value of the hydrocarbons for which we would be spending hundreds of millions on drilling. Drawing on the same techniques used to detect cancerous cell anomalies in human MRI scans, we set about our pioneering work. We were creating and refining high-definition frequency decomposition algorithms, fused with red, green and blue (RGB) blending, and iso- proportional slicing to create meaningful structural and stratigraphic multi-attributes. I use the technical language neither to confuse nor impress, and will return to this point later; it is important as expert practitioners that we use the right language from the get-go.

I remember very clearly the moment we revealed the results or our work (the hours of model refinement and iterations on our calculations) We visualised the target on a screen in high fidelity and granularity; the hydrocarbon bearing reservoir, its extent, as well as its geospatial and temporal context and relevance to the rest of the world. We had used machine intelligent processes, on terabytes of data, to find a target deep in the subsurface, hidden to the human eye, using high performance computing power.

Fast forward 19 years to 2021, Year 2 of COVID. James, a colleague, is in the back of a London Black Cab, at that time working on the UK COVID vaccination scheme (UKCVS). I am in a secure facility in London, and we are talking online with a company who were demonstrating software which identified mis-, dis- and mal- information from very many social media outlets across the globe.

The aim of our work on this occasion was to find anomalies in millions of open-source intelligence (OSINT) feeds, social media or otherwise, which were the root cause of incorrect information propagation and amplification about COVID 19 vaccinations. Furthermore, we were attempting to understand the human networks, and their associations with specific groups responsible, noting the rise of extremist far right behaviours at this time, alongside the associated increases in hate speech. We observed, in my case at the same level of wonderment as the seismic data discovery, knowledge or network graphs visually representing communities across the globe actively sharing and amplifying messages which were simply untrue and conversations that were augmented with deepfake imagery. Machine learning algorithms developed within the software functionality were able to isolate and monitor the anomalous narratives created by, and propagated across, nefarious communities and in some case organisations.

The images, which we suspected were fake, or at least out of context, were deconstructed through machine learning, deconvolving them to a set of pixels, the pattern of which identified them as being generated by artificial intelligence. The pattern of the pixels is striking in AI generated images; they form a bright white spot in the centre and a scatter of RGB colour sequences radiating from this. They also leave a digital fingerprint of the technology which generated them, opening a new threat surface, whose vulnerabilities can be exploited by adversarial actors.

Just over one year later in 2022, a colleague from the UK Ministry of Defence (MoD) and I were working together, in another secure location, as Russia continued its aggressive attack on Ukraine. We had been tracking the increasing escalation since 2014, and the recent uptick in cyber and virtual activity directed towards Ukraine that was attributed to Russia.

Many in the defence and national security communities had been observing carefully since 2017 how Russian aggression had altered, shifting from activity in the physical, kinetic, and lethal domain, to that of the virtual and cognitive. Widely attributed to Russia, the introduction of the novel cyber pathogen, NotPetya, which leveraged the exploit EternalBlue, initially developed by the U.S. National Security Agency (NSA), caused local disruption to power stations and financial systems, and almost resulted in bankruptcy for the global shipping giant Maersk. NotPetya has been estimated to have caused over $10 Billion of global damage across eighty companies, as well as its intended target country, Ukraine. That is the central point here; it was an act of war against Ukraine, the consequences of which had a disproportionate effect globally.

An obvious and critically important question then arises: at what point do attributed, sub-threshold attacks spill over into an event which could plausibly escalate across the globe, triggering NATO’s Article 5 mutual defence agreement?

The larger point is that from 2017 the nature of conflict was changing. Enabling technologies were making code ubiquitously available to state and non-state actors, and technology vulnerabilities, in this case software, were being exploited. The digital degradation of Ukraine was gaining pace, creating its own challenges at the beginning of the conflict in 2022. The weaponisation of data had begun.

My colleague and I had been tracking how the conflict in Ukraine had brought about greater awareness of the organisation Bellingcat. Bellingcat had been reporting the atrocities of the Russo-Ukrainian conflict using its powers of crowd-sourced investigative journalism. The organisation uses vast quantities of OSINT and machine intelligent processes, to correlate and fuse data sources to uncover the facts around war crimes and crimes against humanity, providing legal evidence that supports the deliberations of the UN Security Council and other international bodies.

Our work focused on the methodologies Bellingcat had deployed to geolocate a photograph of the Russian Missile Programming Team purported to be behind programming many of the cruise missiles that hit Ukraine at the beginning of the conflict. Bellingcat had used multiple methods to identify the location of the photograph, and the details of those pictured in it, using various OSINT sources; satellite imagery, pattern of life, pattern of movement, facial recognition, image detection, all defined using machine intelligent processes. This was evidence which would later prove critical in defining war crimes.

How could we create, using similar techniques to Bellingcat, a capability which would support intelligence analysis for defence? We knew we had to develop a concept system which was able to harvest multiple, highly sophisticated, data sources, then apply data ontologies which would allow these sources to be integrated and, through machine intelligent processes, reveal multi-attributes from the physical and virtual domains, thus providing us with insight which would have otherwise remained invisible to the intelligence analyst. Our methodologies had to be auditable, leaving a digital trail of the code we had used to develop algorithms that interrogate the complexities and volumes of data we had to process, and analyse. Why? Because this was a multi-domain targeting tool, which would support human decision-making when recognising and engaging with assets of interest.

We posed our hypothesis in an approach paper entitled The British Army’s Approach to AI which was published by the Ministry of Defence in December 2023. The critical enablers, outlined in the approach paper, have been brought to life in what is now a live project, building upon the concept system described previously.

The system builds on the principles I first deployed in 2002 and takes into consideration how the nature of conflict has altered and evolved. It fuses, blends, and cross-correlates relevant data sources, revealing hidden attributes through refined algorithms which interpret, interrogate, and query data to define successful targeting outcomes. These principles are underpinned by our approach to the application of data analytics, machine learning, and AIs against the MoD’s ambitious, safe and responsible applications of AI doctrine, and a new directive known as Dependable Artificial Intelligence in Defence (Joint Service Publication 936).

I mentioned earlier the need for the correct use of technical vocabulary from the get-go. As a defence and national security collective, we must increase our digital literacy, or employ expert practitioners who understand the deep science that lies behind the words used to describe and prescribe the desired military outcomes intended for the technology development. Too often I hear the wrong words used; the explicit use of the right language is critical in ensuring we create solutions which will truly compete with those of our adversaries, rather than falling short because incorrect terminology was used at the requirement and scope of work setting phase. Carl von Clausewitz, the nineteenth century philosopher-practitioner of war, once observed that the analysis of war can prompt an ‘ostentatious exhibition of ideas.’ A ‘serious menace’, he suggested, is the ‘retinue of jargon, technicalities, and metaphors’ that ‘swarm everywhere – a lawless rabble of camp followers.’

Since leaving oil and gas exploration and moving into defence and national security (NS), seven years ago, with the explicit direction to “bring your learning from the oil and gas sector to Defence and NS”, I have observed the changing nature of contemporary war. I made this point earlier and would like to take the time to explain some of my observations.

The emergence of Cloud compute and the exploitation of the internet (dark web, and chat chambers, both of which have enabled the ubiquitous availability of OSINT) are the key differences between developing defence and NS capabilities now, and my work in 2002. Cloud has created ubiquitous availability of technology, democratised it, and created interconnectivity (and thereby the weaponisation of interdependence), which has further diluted the meaning of the state within a competitive international system.

Warfare is increasingly waged through surrogates. These are technological tools, and human capital, which absorb the patron’s political, operational, or financial burden of conflict. The military are adapting emergent technologies, developing surrogates, and weaponizing them; data and artificial intelligence are examples of this.

Surrogates have the ability to disrupt the battlespace kinetically, the information space subversively, and the willpower of the adversary psychologically, all without a major combat operation. Surrogates scale rapidly, once code is released. They proliferate through ubiquitous and democratised access, horizontally (across states) and vertically (through state and non-state actors). Furthermore, emergent technologies can reinforce one another making them even more powerful.

The principles of data-driven decision making, and the use of AIs for targeting, are not new. The examples from my early career and the clear, golden, data-driven decision- making thread interwoven through them demonstrates this. It also identifies that defence and NS is at best some 30 years behind adjacent industries, like oil and gas.

Autonomous weapon systems have tended to dominate the discussion about AI in military applications, no doubt highlighted by events in Ukraine. Less attention has been paid to the use of AI in systems which support human decisions in armed conflicts.

Tools such as the concept system being developed by the Army Futures AI Team, are known as AI Decision Support Systems (AI-DSS), which are a novel concept for defence. These are computerised tools which use AI software to display, synthesise, and/or analyse data, and in some cases make recommendations, even predictions, to aid human decision-making in war. These are not unlike the systems and software I was using in 2002 for seismic data mining.

The use of AI-DSS for targeting purposes, in all domains; physical, virtual, and cognitive, demonstrates how the traditional domains of air, land, sea, space, and now cyberspace, have evolved and converged. The consequence is that an ever-growing proportion of the contemporary battlefield bears little resemblance to the armoury of previous global conflicts, for which international legal frameworks were originally constructed. This presents us with moral, ethical and political challenges, as conflict moves from a purely human endeavour to a world of digital proliferation and data-driven decisions.


This Defence Policy Brief, briefing paper is a lightly edited version of a presentation made by Libby Vallance-Bull to the House of Lords Committee on 1 May 2024 and was originally published on the CPUDS, University of Exeter.

Share this Brief

LinkedIn
Facebook
Twitter
Telegram
LVB

Meet Our Contributor

Elizabeth Vallance-Bull

Libby Vallance-Bull is Co-Founder of black and white consulting, where she advises on data exploitation and the scaling of AI technologies. She has more than 30 years’ experience across defence, national security, and oil and gas. She developed the strategy for the Rapid AI Delivery Taskforce (RAID TF), sponsored by the Chief of the Defence Staff, and co-authored the British Army AI Strategy. She serves as a reservist officer in a specialist technical unit and as a Non-Executive Director of the Police Science Council. She is a Fellow of the Geological Society and holds degrees from St Andrews, Aberdeen and Cambridge.

Co-Founder
black and white consulting
itisblackandwhite@pm.me

Related Briefs

A Trident 2 D5 Missile breaking the surface.
22 Min Read
Dr Charlotte Levy
Timothy Levy
SDR 2025 and the Defence Investment Plan recognise the priorities, but investment is not readiness. Timothy Levy and Dr Charlotte Levy test whether Britain’s programmes add up to an integrated deterrent across five layers: cyberspace, air and missile defence, command and control, conventional forces and nuclear posture. A shock in one layer can spread inward, narrowing the “decision corridor” in which leaders can make controlled choices short of nuclear use. They find a vulnerability gap: legacy capabilities are declining faster…
Civil and nuclear pic
6 Min Read
Andy Stirling
| Science Policy Research Unit (SPRU), University of Sussex
Phil Johnstone 
| Global Studies, University of Sussex
The government's nuclear deregulation drive rests on assumption, not evidence, never comparing nuclear with cheaper, faster alternatives, downplaying safety risks, and denying trade-offs outright. A less visible driver, they suggest, is military pressure: sustaining Britain's submarine-building base costs over £5 billion a year.
Royal Navy F-35B fighters of HMS Prince of Wales intercept a Russian military aircraft
13 Min Read
Matthew Heaslip
| University of Portsmouth
Britain's conventional deterrence has failed. Hostile actors — Russian aircraft and submarines and Iranian drones, street-level subversion — are testing UK resolve with growing confidence. Decades of deferred spending have left the Royal Navy and RAF critically overstretched, with maintenance backlogs meaning only a handful of warships and combat aircraft are operationally ready. The planned uplift to 2.7% of GDP is insufficient. The Treasury seems to have failed to realise that deterrence is cheaper than war; it must recognise that…
UK ARMED FORCES STEPPING UP FOR NATO ACROSS LAND SEA AND AIR
15 Min Read
Paul Cornish
| University of Exeter
The British Army is threadbare and strategically directionless, its credibility eroded by talk of a “first duty” that goes unfunded. Paul Cornish asks what an army is for: taking and holding territory, and deterring aggression by being demonstrably able to do so. He finds a coherent answer in SDR 2025: a force of at least 100,000, two divisions at the heart of a NATO strategic reserve corps, and a “high-low” mix in which technology multiplies rather than replaces mass. Endorsed…
London-,September,,18,,2023:,The,Ministry,Of,Defence,Main,Building
20 Min Read
Steve Prest
The MOD is reorganising its procurement system again, but Steve Prest argues the organisation chart was never the problem. The failure is systemic. The department commits to more than it can afford, masks shortfalls with promised “efficiencies”, lowballs costs to win approval and imposes in-year savings that slow delivery. Optimism bias and risk models blind to tail risks make it worse. So do functions that can veto programme leaders, producing compliant failure rather than imperfect success. The remedy is known:…
Sir Gwyn Jenkins
7 Min Read
Peter Spayne
Gaps in capability and low numbers of vessels have become front page news. The concept of a ‘hybrid navy’ — championed by the First Sea Lord — is designed to address these gaps and shortfalls. A hybrid navy combines traditional crewed warships with remotely crewed vessels and autonomous systems, as well as contracting in live streams of commercially sourced data. Uncrewed systems access environments too dangerous for manned ships, while frigates, destroyers and carriers retain essential roles in complex combat.…
HMS Dedender-Sea Viper
6 Min Read
Peter Spayne
Iran has learned from Ukraine how to trap Western navies in an impossible choice. Deploy a billion-pound warship and risk losing it — broadcast live to the world — to a swarm of cheap drones. Or stay home and be accused of weakness. Either way, the West loses. Peter Spayne explains why this dilemma is by design.
The Next Generation Fighter Aircraft - A European Dilemma?
5 Min Read
Keith Hartley
| University of York
There are two parallel sixth-generation combat aircraft programmes being developed in Europe — the UK-Italy-Japan Tempest (GCAP) and the Franco-German-Spanish FCAS. The USA has only one, the F-47. This article examines whether merging both projects into a single six-nation programme would deliver cost savings through shared development costs and economies of scale, while acknowledging the governance and cost-escalation risks. An Anglo-French industrial joint venture, modelled on the Jaguar's SEPECAT arrangement, might be part of the answer.
this-shock-to-the-western-alliance-should-send-a-shock-wave-through-whitehall:greenlandandnato
5 Min Read
Roderick Crawford
| Policy Brief
The lesson from the near crisis over Greenland is that the UK will need to step up its role in securing the Transatlantic alliance, not least in taking the lead role in securing the High North. Providing security for the US may now be more important than contributing to the security of the continent. It is certainly a mission the UK is uniquely equipped and positioned to fulfil. A credible UK defence policy will require substantial additional investment that can…
a-growth-strategy-that-imperils-defence
10 Min Read
Ron Smith
| Birkbeck University of London
The Defence Industrial Strategy should have been subtitled “Making the Weapons We Need”, not “Making Defence an Engine for Growth”. Growth has become an objective, distorting decision-making about the primary function of policy: efficient acquisition of the armaments the UK needs. The SDR compounds the problem, seeking to achieve deterrence now whilst seeking to deliver this within a ten-year financial framework. And the problems don’t end there.

Email A Letter To Our Editor

Scroll to Top