With the substantial increases in defence spending which we are seeing across Europe and beyond, there’s never been a more important time to shape an emerging defence technology. Globally, AI investment is measured in the tens of billions, with the US, China, and several other nations vying for leadership. Global AI investment in 2024 was $110 bn, up 60% in 2023. 74% of that investment was in the US, followed by China, the UK and France. It’s encouraging to see the UK in at number three, although Europe as a whole achieved only 12% of global investment. These numbers represent fundamental drivers for future prosperity and power.
The title of this blog may raise some eyebrows. “AI in Defence.” It’s a broad category! Invariably, such a broad topic leads to too much fine detail to be of any real importance, or be so high-level as to consist entirely of platitudes.
It’s a bit like giving a talk in 1948 on “the potential of transistors in the future of defence” – another general-purpose technology which held the potential to make almost everything we did easier, faster, cheaper, more accessible. This makes it a good mirror for thinking about the transformative potential of AI.
So, let’s start by talking about the transistor. I asked ChatGPT to help me with this – which, satisfyingly, involved enlisting the help of AI running on 7 trillion transistors. That may seem like a nice narrative coincidence, but it’s actually indicative of the relationship between the two.
So, from 1948 onwards, what did the transistor do for defence?
- Communications: Portable radios, enabling real-time battlefield communication.
- Radar: Compact radar systems, enhancing situational awareness.
- Weapons: Lightweight guidance systems in precision weapons.
- Analysis: Real-time battlefield analysis and improved C2 systems.
- Simulation: Advanced wargaming and training.
- Intelligence: Enhanced signals intelligence and cryptography.
- Electronic Warfare: Compact radar jammers and electronic countermeasures.
- Stealth: Active cancellation systems and advanced surveillance.
- Navigation: Precise navigation and positioning systems.
- Autonomous Systems: Drones, UAVs, and robotic ground vehicles.
Considering that list, we can ask for each – “does AI further enhance this?”. The answer is, for almost all of them, yes it does. Why is that the case? The answer is that AI and transistors are both tools that enable the same thing: faster – and better – processing of information. To put succinctly, “AI is an amplification of things that we already do”.
The Core Problem
The transistor did two things: firstly, it allowed for miniaturisation. I don’t think that’s a core aspect of what AI will achieve, although it may help. The other thing is that it enabled calculations to be made faster. AI certainly helps with this. Let’s try to imagine fighting an adversary with a broad AI capability, if we had none, using our analogy.
What would it look like if we had to face an adversary who had the transistor, when we did not? The adversary would have overwhelmingly superior communication, intelligence, weaponry, and logistics, leading to overwhelming strategic and tactical dominance over us.
We would rely on bulky communications systems, prone to failure. Our adversary would enjoy secure real-time comms, decentralised C2, and a well-coordinated joint force. We would have bulky, static radar systems, and slow and manual intelligence analysis, making us vulnerable. Our adversary would have miniaturised, mobile radar systems, better SIGINT, and the ability to rapidly understand and respond to our actions. They’d also have EW capabilities that we would have no answer to.
We’d be stuck with “dumb” munitions, requiring far more ordnance and units to achieve the same effects. Our adversary would win decisive battles with fewer resources, stretching both our forces and our economy.
Our aircraft would rely on visual targeting. Our adversary would have beyond visual range air-to-air weaponry, integrated air defence systems, and all-weather operating capabilities, resulting in complete air superiority.
Given the above, we would lose on:
- C2 and joint force coordination
- Intelligence analysis
- SIGINT
- EW capabilities
- Smart weaponry, and volumes of ordnance required
- Air superiority
It’s not hard to imagine a similar differential in capability caused by AI superiority.
The Geopolitical Stakes and The Data Imperative
The geopolitical landscape is accelerating this AI arms race. We see major powers, investing heavily in AI research and development. The number of AI-related publications from these adversaries is outpacing other nations, indicating a focused effort to achieve dominance. This isn’t just about academic prestige; it’s about strategic advantage.
And here’s a crucial point: in the realm of AI, data is king. It’s never been about having the “best” algorithms in isolation. The algorithms are, increasingly, becoming commodities. At Hadean, we took the strategic decision two years ago to not develop our own AI foundation models, because we believed they’d be commodified and their cost competed down to the price of the silicon. This has been borne out. The critical factor instead is about who has the most comprehensive, relevant, and actionable data. This data fuels the training of AI models, making them more accurate, more adaptable, and more effective, as well as giving trained models something to chew on to produce valuable outputs.
Therefore, data capture and ownership are paramount. We must prioritise getting our data into a unified, accessible environment where it can be analysed and understood. This is the foundation upon which all effective AI applications are built. To see the importance of this, we expect the speed of progress in large language models to slow, because they have already consumed the entire internet as training data: they’ve run out of fuel. We cannot permit this to happen in defence AI. For models in defence, we need the equivalent of what Tesla has done: get the systems that you hope to one day be autonomous on the road, get human users to drive them, and gather the data on how they do it well, and what their failure modes look like. Tesla has been doing this for nine years, and it’s only just beginning to yield results in terms of initial autonomous driving capabilities.
This is where concepts like C4ISR (Command, Control, Communications, Computers, Intelligence, Surveillance, and Reconnaissance) become critically important. We must ensure complete integration of data from disparate systems, providing a holistic, real-time picture of the battlespace. This is valuable independently of AI, but it also sets AI up for success, so it can get on with the job of very, very rapidly sifting, identifying, and surfacing signals in the noise. It’s also true that the only real way to have AI that nobody else can have is to own a training dataset that nobody else has, and this is a possible route to that.
The point is, to start the process of finding needles in a haystack, we must first put ourselves in the position of owning a haystack in which we believe there are needles.
The Challenges and Complexities of Solutions
The challenges are significant. Integrating AI into existing defence systems is complex. Aligning legacy data systems, ensuring interoperability, overcoming bureaucratic hurdles, and navigating security classifications are major undertakings. The “hard part,” isn’t necessarily developing the AI algorithms themselves, but the data pipelines, the organisational change required to make AI truly operational.
This is where a rich home-grown AI ecosystem will also provide value. As noted, with UK AI startups receiving the third highest VC funding inflows globally, we have a strength here to leverage. We must be wary of the fact that – despite the availability of excellent and continuously-improving open-source models – some secret models may remain, their existence only becoming apparent on the battlefield. We must ensure that we have the sovereign talent to analyse and unpick these capabilities in short timeframes.
A Vision of the Future
Via AI – software can become truly responsive and reactive, adapting in real-time to the demands of the battlespace, tuning up the OODA loop, and making the synthetic environment into the new operating environment.
The one major key to achieving this is in enhancing and deepening systems connectivity and data capture. AI doesn’t come from nowhere, and once you have it, it can’t operate on nothing. You have to capture the data – there’s a mantra that in AI, whomever has the most data wins (not whomever has the best algorithms), and there is a clear hypothesis that whomever has the best AI will win in conflicts.
This is something that I was pleased to see Dstl doing about eighteen months ago, when they orchestrated a series of beach landings on the Solent, solely to capture what they looked like from a variety of sensors, for the use in the creation of AI.
This is important, but it should not have been a specially-organised event. It should and must be routine when we operate to capture and save all motion, events, and modalities, so this must be baked into our software systems, networks, and platforms.
Creating this information space where AI can deploy, train, and operate – alongside humans who are also deploying, training and operating – will be transformational for both. It will enable decision support systems that are much more powerful than their predecessors, with the ability to rapidly spot connections, correlations, and threats – flagging information to humans and making predictions about what might happen next, along with the associated risks. It will allow us to create more reactive, responsive, realistic training environments, powered by rich pattern-of-life and human terrain, enabling us to train as we fight. The precursor to all this is connecting systems, enabling interoperability, and collecting data, and is underway today.
The British Army’s ambition to double its lethality in three years and triple it by the end of the decade will only be achieved if we are successful with our use of AI. General Sir Roly Walker’ vision of a fifth-generation land force, leveraging AI and autonomous systems, is a bold and necessary step. It’s about creating an “any/any network,” connecting any sensor to any effector, building an “internet of military things”.
This requires a shift in mindset. In Sir Roly’s vision, commanders must become systems integrators, understanding not just tactics, but also the technology that underpins their lethality. This is what some are calling “techcraft” – the fusion of fieldcraft and technology.
Direction of Travel
The innovation ecosystem is key to achieving this vision. We need to move beyond the fragmented landscape of government labs, start-ups, and traditional defence industry players. We need to foster collaboration, streamline the transition of technologies from research to deployment, and overcome the barriers to scaling. Initiatives like the US Joint AI Centre (JAIC) and the UK’s Defence AI Centre (DAIC) are crucial steps in this direction. Personally, I’m a huge fan of co-creation, where the problems of the operators are exposed directly to the people who might think of solutions, rather than being filtered through layers of bureaucracy before an RFI is produced.
Some points to finish:
- Data Sovereignty is Strategic: Control over data is paramount. We must prioritise data capture, storage, and analysis within our borders, within our existing defence systems, and aligned with those of our trusted allies.
- Infrastructure is Critical: Invest in the infrastructure to support AI deployment, including high-performance computing and robust data pipelines.
- Think Systems, Not Just Platforms: The aim is to create integrated systems that give advantage and reinforce the whole AI defence ecosystem, not just faster/better platforms.
Conclusion
AI in defence isn’t just about keeping pace with our adversaries. It’s about fundamentally reshaping the battlefield of tomorrow, ensuring security, and safeguarding sovereignty in an uncertain world. It’s about achieving a decisive advantage, not through mass alone, but through superior intelligence, faster decision-making, and seamless coordination. It’s about deterring conflict by demonstrating an overwhelming capability to respond effectively. The future of defence is being written now, and it is being written in the language of AI.
Remember, the UK sees the third-largest investment in AI of any nation, and that is a fantastic position to start from. We have an excellent talent pool, thanks to our world-class universities and the presence of world-class R&D labs. Let’s ensure we take advantage of that.