How the Front Line can realise the Defence AI Strategy
The Ministry of Defence’s new Defence AI Strategy is an exciting step towards the UK’s ambition to be a world leader in the responsible development and use of AI. It has four objectives; transforming Defence into an ‘AI Ready’ organisation, adopting and exploiting AI at pace and scale, strengthening the UK AI ecosystem, and shaping global AI developments to reflect UK values.
Front Line Commands (FLCs) will be invited to share their responses to the strategy, in order to integrate their AI requirements into the next round of Command Plans in FY 23/24. With AI being such a wide-ranging field, what should they focus on?
There are many problems that AI or Machine Learning is good at solving and there are problems that AI is not best suited to. As AI gets closer to the Front Line, a deeper, cultural, understanding of where this line lies will be vital. The AI Strategy rightly talks about training senior officers to help them understand AI’s potential.
But true value will be unlocked when training is cascaded elsewhere. While at too low a level for the Strategy, we can assume that soon a Front Line engineer, or combat specialist, will be required to have deep knowledge of both the task at hand and the strengths and weaknesses of the AI tools available to solve it. These ‘AI Officer’, or ‘AI NCO’ individuals should be identified now.
Although it will take the Defence AI and Autonomy Unit (DAU) time to navigate the policy, including the legal and ethical barriers to Front Line AI use, the individuals that will deploy these tools should be identified and trained now, ready for when these elements are overcome. This will have the twin advantages of, firstly, helping FLCs to understand what they want to use Front Line AI for, while also introducing a culture of ‘AI Now’ among the Services, including recruitment of data science and AI experts, if required.
It is highly likely that the best uses of AI are yet to be identified and are currently inside the mind of a Royal Navy Weapon Engineer Leading Hand, or REME Corporal. These are the very people who know the use case to be solved, and the location of the dataset that will train subsequent ML models. These individuals need to be found, trained and used.
Preparing for AI Now
The strategy makes the distinction between AI Now (those tools which have already proven themselves in other settings and have demonstrated value) and AI Next (those less-proven tools with larger potential). FLCs should focus on AI Now use-cases and quickly identify combinations where meaningful datasets exist and where highest potential value can be quickly realised.
In many cases, as the Defence AI Strategy itself highlights, these will reside in preventative maintenance, logistics, training and knowledge discovery. These areas have already demonstrated significant value generation to private industry and wider Government, and come with a large ecosystem of experienced SMEs that can support development. FLCs should work now to understand where such data resides, whether it needs cleaning and labelling, ensure classification will allow contractor access and, of course, work out who ultimately owns it. Understanding these elements will take time, and the work to untangle these unknowns must begin now.
AI Now, by definition, is the quickest route to unlocking value within the myriad of datasets that Defence owns; but it is the FLCs who will acutely understand this data in detail, as well as the use cases that will generate most value.
As an aside, an understanding of where the budget for AI Next should come from should also be understood. Keeping future AI procurement tied to Platform spending is unhelpful, as this funding will be too slow to emerge if it remains linked to the purchase of large, complex systems.
Beyond the Top Level Budget (TLB)
In essence there are very few AI Now styles of project; predictive maintenance, knowledge discovery, improved image or signal detection (whether on-edge or elsewhere) or decision support (whether for humans or hardware). These can be grouped into needing similar algorithms, or shared problem-solving approaches – even if the details will differ.
For example, Knowledge Discovery or Technology Horizon Scanning requires cutting-edge NLP tooling to be trained, iterated, and tested on post-deployment reports (PDR). Although the PDRs of a Royal Navy Frigate or Armoured Infantry Battlegroup will differ greatly in style and substance, the underlying data science principles and required development environments will be roughly equivalent. So, why not share resources?
The same can be said for calculating the predictive maintenance needs of different platforms or understanding the intricacies of supply chain management. Cross-TLB organisations such as Dstl’s Defence AI Centre (Experimentation), or Digital Foundry’s Defence AI Centre (Operations) should be the first port of call, used as a hub of advice, the conduit to other organisations, and as a central register of FLC AI programmes. This will prevent re-inventing the wheel, speed up AI adoption, and allow the FLCs to better learn from others’ experiences.
The Defence AI Strategy is a great first step in the right direction and identifies the right overarching strategic goals. FLCs don’t need to wait for the next Command Plan cycle, or for the DAU to issue further guidance to get ahead of the curve. Identify and train key personnel, even to only cover the basics. Identify, access, clean, and understand the owners of key datasets. Finally, find those quick win, AI Now use cases. The best ones will be found and solved if FLCs leverage the DAIC, other TLBs, the commercial AI SME ecosystem and, of course, newly-trained, Front Line personnel.