The recent shift forcing the Alan Turing Institute toward defence-focused AI research has sparked major questions about the UK’s innovation strategy.
Many argue that the government should have created or funded a dedicated defence AI institution with a clear mission, avoiding dilution of the Institute’s vital civil AI research and social innovation. The sudden pivot caused staff unrest, leadership upheaval, and risked ongoing societal research programmes.
Public trust and accountability are also crucial. National security projects need specialist oversight, ethical governance, and transparency—elements compromised when defence priorities are fused into a broadly purposed public research institute.
The UK already has specialist institutions developing defence AI. The Alan Turing Institute runs the Defence Artificial Intelligence Research (DARe) Centre in collaboration with the Ministry of Defence and intelligence agencies. The Defence Artificial Intelligence Centre (DAIC) integrates AI across military operations, while the AI Security Institute addresses AI safety and security risks. The Defence Innovation Organisation supports industrial partnerships with a significant ring-fenced budget. These dedicated bodies are designed to drive rapid advancements in national security AI.
Defence AI demands specialist infrastructure, security clearance, and operational protocols that a repurposed civil institute is ill-equipped to provide.
Despite hundreds of millions invested, the UK has yet to deliver a sovereign large language model (LLM) for business and societal use. Promising projects like BritLLM at UCL remain nascent, highlighting ongoing challenges in creating competitive homegrown AI systems.
The Alan Turing Institute was founded with a societal mission to advance data science and AI for public benefit, tackling challenges in health, environment, and public policy while promoting ethical innovation. Diverting its focus risks losing inclusivity, research diversity, and its role as a national innovation hub.
This strategic pivot reflects government priorities amid geopolitical uncertainty but many warn it may undermine the broader AI ecosystem and long-term societal gains. Balancing national security with a thriving civil research base remains a core challenge.
Ultimately, a purpose-built organisation for defence AI could advance national security while preserving the Turing Institute’s original societal mission. It’s time to for the UK Government to rethink its strategic AI R&D for the future.
* Christian de Vartavan is an eminent scholar and now CEO of a London blockchain consulting company and Associate, APPG AI, House of Lords.
Follow @libdemvoice.org on Bluesky
Like us on Facebook
Subscribe to our feed
Sign-up for our daily email digest
Professor Joad sees his play at the Theatre Royal, Northampton
The DVLA's problem with number plates
Is Kier Starmer the new John Major?
A bargain at 23 guineas?
20C night
Roland
David Raw
Chloe
Tom Bailey
4 Comments
Yes. One wonders why Defence AI isn’t within the remit of Dstl already. More to the point, does it make sense to invest in British LLMs when the whole LLM industry in the USA is experiencing such problems. Just throwing more and more data into the training set has shown diminishing returns and we have consumed almost all the text on the internet already. Perhaps we need to look at more limited applications where there are immediate benefits, like the analysis of X-ray, CT and MRI images where we know that AI diagnosis can outperform specialist consultants and the vast supply of data within the NHS would provide the sort of input that could create a really useful AI product for the UK to export.
I’m glad that you have highlighted the re-pivot of the Turing Institute to defence work from its initial broad-based initial mission to advance data science in general. Hopefully the new technology secretary, Liz Kendall, will take another look at the institute and its governance issues to see how it can/could be reconstructed to serve its initial mission – or indeed whether it should be closed altogether since it seems that much of its public-facing work is being replicated (or could be carried out) in other fora.
Defence is clearly a different issue, since defence-related data is clearly not “freely” available – you certainly cannot scrape the Internet easily. However, this should also involve collaborations with other countries, since restricting data to that obtained frin the limited conditions of the UK will not provide the insights needed to develop systems for deployment across the world.
Laurence makes a good point about whether it makes sense to develop a British LLM and whether more investment should be put into more limited applications. Indeed, there is an increasing amount of investment in the development of specialised language models and so I don’t think that there is under-investment. The one concern that I might have is the use of the data being provided to US companies from, for example, the NHS might be subject to a disclosure order under the US CLOUD Act.
As to a specific British LLM, I think that the jury is out on the use of such a model. There are issues with the US-based training of existing LLMs. However, it does seem that increasingly it is possible to instruct LLMs to produce British-focussed output. I have instructed my ChatGPT account to always provide UK-English spellings and focus on UK/EU/Australian issues and it does seem to have understood my requirements.
@Laurence Cox
> One wonders why Defence AI isn’t within the remit of Dstl already.
Looking at their website they have many scientific staff who are expert in things that are NOT so-called AI. Since they have day-jobs in their domain, it may make more sense to ask them to partner with existing ML experts in frameworks run by people with a track record in research management. I hear Dr John Bell may be available just now.
It takes time to spin up a new institute, or change direction of an organisation. Adding a new group to an existing organisation is quicker and lower risk. It is not a big PR splash or an ego boost or provide a set of new sinecures, but it can be quietly removed if it fails. Fail fast, yes, but fail SMALL.