Why Northern Ireland Can Lead the World in Practical AI
An opinion piece by David Crozier CBE and Dr Donnacha Kirk
An opinion piece by David Crozier CBE and Dr Donnacha Kirk
There is a quiet assumption embedded in much of the current discourse on artificial intelligence, namely that the future belongs exclusively to a handful of American frontier laboratories, with a few Chinese competitors following close behind. Bigger models, bigger data centres, bigger cheques. For nations and regions without hyperscale infrastructure, the implied role is that of consumer, renting intelligence by the token and accepting whatever terms, prices and priorities the frontier providers set.
We believe that assumption deserves to be challenged, and we believe Northern Ireland is precisely the kind of place from which to challenge it.
The history of general purpose technologies teaches a consistent lesson. The economic value of electricity, of computing, of the internet, was not captured principally by those who invented them. It was captured by the economies that diffused them most successfully across ordinary firms. The scholar Jeffrey Ding has made this argument compellingly in his work on technology and the rise of great powers: the long run winners of the AI transition will be those who get the technology into the hands of the everyday economy, not those who build the largest frontier model.
This reframing matters enormously for a region like ours. It means the race that determines prosperity is not the one being run in California, nor the one being developed around King's Cross. It is the race to close the gap between what AI can do and what an under resourced twelve person firm in Ballymena, Strabane or Newry can actually deploy. And in that race, remarkably, the field is far less crowded. The frontier laboratories are locked in competition with one another at the top of the market. Almost nobody is competing seriously for the diffusion of practical AI into small and medium sized enterprises. That space is open, and Northern Ireland is unusually well placed to occupy it. To be precise about the claim: diffusion is a complement to serious national investment in compute and frontier capability, not a substitute for it. This is a parallel race, one a region can win while the harder one is still being contested, and it is the race currently going unrun.
The evidence already bears this out. Independent research published this year by Trinity College Dublin and Microsoft Ireland, surveying 300 senior leaders across the island of Ireland, found that 62 per cent of Northern Ireland organisations already have AI tools implemented and in active use, compared with 39 per cent in the Republic. This is not a claim we make about ourselves. It is a measured, external finding, and it points to precisely the advantage this thesis depends upon, that Northern Ireland’s businesses are further along the road to practical adoption than is generally understood, and further than many larger, better resourced economies nearby.
For the overwhelming majority of businesses, the questions that determine productivity are not answered by a trillion parameter general intelligence. They are answered by systems that understand a specific firm's operating environment: products, processes, terminology, regulation, customers. A precision engineering business in Mid Ulster, a food processor in Portadown, a professional services firm in Belfast, none of these needs a model that can solve Erdős problems. Intelligence is not the limiting factor. The value of an AI application comes from access to well-ordered proprietary data, remote sensing and integration with existing systems and people. These companies need a model that knows their world intimately, runs affordably, keeps their data under their control and improves the margin on every job they quote.
This is the domain of Small and Specialised Language Models and appropriately scaled AI more broadly, systems trained or fine-tuned on domain specific data, deployable on modest hardware, often on premises, and tuned relentlessly for a defined set of tasks. Nor is this confined to language. Vision systems running on inexpensive edge devices can grade produce, detect defects and count stock on the factory floor. Compact time series models can sit physically beside large machinery and predict maintenance needs before failure. These are the tasks that define the physical, regional economy, in agri-food, in advanced manufacturing, in the trades, and they are precisely the ground that cloud based frontier services are least able to contest.
There is also a tier of firms for whom the entire cloud AI proposition is dead on arrival. A MedTech company handling patient data, a law or accountancy practice bound by duties of confidentiality, a manufacturer whose process data is its competitive moat, none of these will ever send their crown jewels to a third party endpoint. For them, AI that runs on their own premises, provably under their own control, is not a preference. It is the precondition of adoption.
All of these needs can be met by existing technologies but many SMEs are intimidated by the pace of AI change and do not know where to start. The big AI labs are not putting much effort into closing that gap.
The temptation for an organisation trying to diffuse applied AI is to deliver bespoke projects, one firm at a time. Bespoke work matters, it is where the most inspiring demonstrations come from, but no team of any size can transform an economy of tens of thousands of SMEs through direct engagement alone. The multiplier comes from something different: repeatable patterns.
A pattern is a proven, documented combination of hardware and software. Patterns delivered using affordable off-the-shelf kit and open source models are ones that a firm can pick up, deploy and, critically, maintain itself. Retrieval over a firm's own documents stood up in a day. A sovereign, on premises inference appliance built from high street hardware. A sector specific fine-tune whose data curation recipe and evaluation set, developed once for one agri-food or MedTech firm, become a reusable asset for the whole sector. The pattern, once proven, costs almost nothing to replicate.
For some more technically confident firms, documenting a workable pattern will be enough for them to pick it up and run with it but, from our on the ground experience, lots of companies need more support. Whether through online sessions, in person workshops or on site tutorials, ironically face to face teaching remains the best way to promote new tools and technologies.
By 1914, the Department of Agriculture and Technical Instruction had 138 instructors travelling across Ireland teaching farmers cutting edge agricultural techniques. Part of the Agricultural Extension movement, these instructors aimed to "extend" the latest learning on food production from the universities and model farms to every corner of the island in ways that made sense for small farmers in their own situation. Horace Plunkett, the Department's founding force, charged each instructor to be "the guide, philosopher and friend of the existing farmers".
Trusted independent trainers, telling not selling. The technology is different today but the approach is the same. This is how a small institution like the Artificial Intelligence Collaboration Centre (AICC) moves a whole economy, and it is how the work compounds rather than merely accumulates. And it is already happening: since establishing in 2024, the AICC has provided hands on support to more than 120 SMEs across the region, most of them building AI proofs of concept tailored to their own data, business problems and use cases.
One living vestige of the Agricultural Extension project is The Archers, an everyday story of country folk with a side helping of advice to help improve productivity in an era of postwar food rationing. Perhaps we will know AI diffusion is really succeeding when the residents of Ambridge start swapping vibe coding projects over pints in The Bull.
There is also a harder edged argument. Every business, and indeed every public body, that builds its critical processes on rented frontier intelligence is accepting a strategic dependency, exposure to pricing decisions, model deprecations, terms of service changes and geopolitical currents entirely outside its control. That last point is no longer hypothetical. Recent episodes in which export controls abruptly, if briefly, withdrew access to leading frontier models reminded every dependent business how quickly the ground can move beneath it, and serious scenario work in Europe now war-games futures in which access to frontier intelligence is rationed, country by country, as an instrument of statecraft. Token based pricing introduces cost volatility that few finance directors have yet fully priced in. Appropriately scaled, locally run AI converts that unbounded recurring liability into a modest, predictable capital cost, and it allows a firm, quietly, to opt out of a considerable amount of geopolitical risk.
We should be honest about the risk on the other side of this ledger. A model fine-tuned today is frozen at today's frontier. The field will move, and a firm left maintaining a depreciating asset it cannot update has been done no favours. The frontier labs might pivot to directly supporting SMEs and "eat our lunch".
The answer is a principle we hold to firmly: we do not give a firm a model or even a fixed application, we give it the durable capability to keep choosing the right approach. Every pattern must be simple enough to revisit and extensible as the technology improves, supported by objective, vendor neutral evaluation covering performance, cost, reliability and data security. Armed with this, a firm can measure quality for itself and swap the underlying model, local or hosted, open or frontier, as the field moves. That evidence based neutrality, the willingness to tell a firm when a hosted frontier model is now the better answer, is exactly what makes the AICC trustworthy where a sales engineer is not.
Two commitments run through everything above, and they are not decoration. The first is responsibility. Every pattern the AICC promotes travels with structured assessment of governance, ethics and assurance, because an SME's first AI deployment must be one it can defend to its customers, its regulator and its own workforce. Trust, once established, is the cheapest distribution channel there is. Lost, it is nearly unrecoverable, and a single high profile failure would set regional adoption back years. Responsible adoption is not a brake on the flywheel. It is the flywheel.
The second commitment is that adoption must be pro-worker. The evidence on technology diffusion is unambiguous, it succeeds where the workforce is brought along and stalls where it is imposed. Because patterns are taught rather than installed, it is the firm's own people who acquire the skill, and the productivity dividend arrives as augmentation, better quoting, faster compliance, fewer hours lost to unplanned downtime, rather than displacement. Where roles do change, the same capability building that stands up a model can reskill the person working beside it. A region that adopts AI with its workers, rather than at their expense, will adopt it faster, embed it deeper and keep the gains at home.
This approach is also, increasingly, where the national debate is moving. Conversations in Westminster and beyond have turned towards technological sovereignty, towards making AI work for British firms, workers and communities rather than importing a strategy shaped elsewhere, and towards greater regional control over how the technology actually lands, a philosophy of place and devolution that its most prominent advocate has lately badged Manchesterism. Whichever way the political weather turns, the direction of travel is unmistakable, and Northern Ireland already has a working model of it. A devolved, regionally rooted institution, teaching sovereign, appropriately scaled, responsibly governed AI to the everyday economy, is not a proposal awaiting a white paper. It exists, it is delivering, and its lessons are available to every region of these islands.
The final step in the thesis is the most ambitious, and the most important. The expertise developed by solving this problem for Northern Ireland's SMEs is not consumed in the solving. Every proven pattern, every evaluation harness, every sectoral data recipe adds to a body of exportable know-how, and the demand for it is global. Every advanced economy, and a great many emerging ones, faces the same SME productivity challenge and the same unease about frontier dependency. If a pattern can deliver impact in Ballymena, the same pattern can travel from Bilbao to Baltimore.
Northern Ireland has form here. We turned hard won expertise in cyber security into one of the world's leading clusters, exporting services and attracting investment far beyond our shores. A decade ago Estonia made itself the world's reference point for the digital state. There is no structural reason Northern Ireland cannot become the world's reference point for practical AI adoption in the everyday economy, the place other regions visit to learn how it is done.
None of this is inevitable. It requires sustained investment in skills, patience with long term institution building, firms willing to experiment and share what they learn, and an unwavering commitment to responsibility and governance, because that is where the durable advantage lies. But the prize is considerable, a more productive indigenous economy, a meaningful reduction in strategic dependency, and a new export industry built on expertise rather than infrastructure. Small models, proven patterns, a small region thinking bigger than anyone expects. That is a thesis worth backing.
David Crozier CBE is Director of the Artificial Intelligence Collaboration Centre (AICC), a partnership between Ulster University and Queen's University Belfast, supported by Invest NI and the Department for the Economy. Dr Donnacha Kirk is Deputy Director of AI Technology and Research Services at the AICC.
Innovation thrives through connection. Whether you're an SME, researcher, or professional exploring AI, we’re here to help.
