2025 marks a decisive moment in our relationship with artificial intelligence. Across governments, industries, and communities, we are asking not only what AI can do — but who it serves and who it might exclude.

For years, the promise of AI innovation has coexisted with a growing awareness of its blind spots: bias, lack of representation, and the replication of structural inequalities through data. The challenge now is to ensure AI evolves as a tool for social repair, not reinforcement of injustice.

As an academic working on the intersections of AI, community safety, and digital inclusion, I have seen both the harm and the hope that accompany technological progress. The future depends on whether we can make AI more human-centred, transparent, and inclusive.

Seeing the streets differently: Turning hate into insight

Through my project StreetSnap, we are implementing image recognition to identify and analyse hateful graffiti in public spaces – turning what is often dismissed as vandalism and nuisance into real-time data about belonging and exclusion. 

By combing image recognition, practitioner reporting and creative arts interventions through the sister project of Flip the Streets, StreetSnap enables local authorities and residents to respond together, replacing hate with creativity.

This work reimagines what AI can be. Instead of a distant, data-hungry system, it becomes a lens of empathy – mapping the stories that shape our shared spaces. 

If you can’t see the data, you can’t see the problem – but if you can’t see the people behind the data, you can’t solve it.

Listening as data: The power of lived experience

Bias in AI is rarely just a technical flaw; it reflects the social inequalities embedded in our datasets. Communities most affected by discrimination are often the least represented in the data used to design digital systems. 

That insight inspired the work that is being developed on the Lived Experience Repository of Racism in Wales – a soon to be open digital archive that gathers existing studies and testimonies from people who have experienced racism and exclusion. This platform ensures that policy, research and innovation are guided by real stories, not abstract statistics. 

The lesson is simple: inclusion starts with listening. StreetSnap listens to the language of the streets; the repository listens to the language of lived reality. Both show that narratives are data too – essential for designing technology that reflects, rather than erases, human experience. 

Beyond bias: Building accountability into AI

Much of the public conversation about AI ethics remains focused on mitigating bias, but we need to go further. Accountability means asking who participates, not just how the algorithm performs. 

In these projects, I work with artists, young people, community safety teams and policymakers to co-design how data is gathered and interpreted. When diverse groups are part of the design process, the outcomes are not just more ethical – they are more trusted, relevant and resilient. 

True AI accountability cannot be achieved through audits alone. It depends on shared ownership, where communities shape the tools that shape their lives and experiences.

Digital inclusion as democratic infrastructure

Digital inclusion is often described as an access issue – broadband, devices, or skills – but it is also a democratic issue. If participation in digital systems determines access to services, safety and opportunity, then inclusion is foundational to social justice. 

Projects like StreetSnap and the repository demonstrate how inclusive AI begins with inclusive story telling. By valuing local knowledge, creativity and lived experience, we are not only collecting better data – we are redefining whose experiences matter in the digital public sphere.

This approach echoes the broader shift in responsible AI practice: from designing systems for people to designing them with people. 

The opportunity ahead: AI for human connection

The next phase of AI development must move beyond efficiency toward empathy. To build AI for good, we must reimagine what ‘good’ looks like in practice.

That means:

• Embedding ethical reflexivity into every AI project — continually asking who benefits, who is visible, and who is missing.
• Merging data and creativity, making social issues visible and actionable through digital storytelling.
• Empowering communities to co-create and govern the technologies that impact them.
• Bridging sectors — academia, policy, tech, and the arts — to ensure innovation serves the public good.

AI can, and should, help us see ourselves more clearly — not just predict outcomes, but reflect our values.

Key takeaways for leaders

• Prioritise representation: Build datasets that capture the diversity of lived experience, not just convenience or scale.
• Design with empathy: Invite communities into the design process to ensure AI reflects the realities it seeks to address.
• Invest in digital inclusion as social infrastructure: Equity in data access and participation is essential for ethical innovation.

Looking forward

Recognition in the Digital Leaders AI 100 list reflects not only technological progress but a growing movement toward ethical, community-driven innovation. 

As we look ahead, the most transformative AI systems will not be those that think like humans, but those that help humans think more compassionately about one another.

Read More AI for good

The post Reclaiming digital spaces: Building inclusive AI through lived experience and street data appeared first on Digital Leaders.

For Unilink, a long-standing provider of secure digital solutions across justice and public safety, AI represents both an extraordinary opportunity and a complex responsibility. Over the last 30 years, we’ve built and deployed digital solutions for prisons, probation services, immigration detention, and secure hospitals. Now, the inclusion of AI across our product suite is opening new possibilities; not simply the automation of routine tasks, but the augmentation of decision-making, personalised offender support, risk modelling, and operational insight.

From digital transformation to digital justice

The Ministry of Justice’s (MoJ) digital strategy sets out an ambitious vision: a justice system that works better for everyone, through the smarter use of data, automation, and design. In 2025, this vision is taking shape. AI and digital technologies now support everything, from digital court processes and secure communication to case management and prisoner engagement. 

AI’s real potential, however, lies in turning complex, fragmented data into meaningful insight. Machine learning models can identify trends in case workloads, forecast resource needs, or highlight where individuals may be at risk of disengagement. Predictive analytics can improve safety and planning, while automation reduces the administrative burden and frees staff to focus on more humane services that are grounded in evidence and insight.

Ethical and transparent by design

However, as AI becomes more deeply embedded in justice operations, ethical and responsible deployment is essential. The sector deals with some of society’s most sensitive personal data, and the consequences of algorithmic decisions can be profound. We believe that for digital justice to succeed, ethical governance and human oversight must remain central.

That means explainable AI – systems that can show how and why a decision was reached. It means rigorous data governance that ensures quality, security, and fairness. And it means maintaining human oversight at every stage. Technology can support better judgment, but it should never replace it. Embedding these principles will help to ensure public trust, protect individual rights, and deliver genuinely fair outcomes. 

Human-centred services

Digital justice is about people as much as technology. When designed around user needs, it can transform the experience of both service users and staff, and make justice more accessible and empowering. Adaptive interfaces can support users with low digital literacy or those for whom English is a second language. Chatbots or virtual assistants can provide 24/7 access to essential information. Data visualisation tools can give frontline workers a clearer view of progress, helping them to intervene earlier and more effectively. 

Crucially, digital services can help individuals to build the digital confidence and skills that extend beyond the justice system, supporting rehabilitation, employability, and reintegration into society, and enabling long-term positive change. In this context, technology isn’t just modernising justice, it’s humanising it. 

Innovations that are shaping 2025 and beyond

Looking ahead, Unilink is focused on three key innovation strands for 2025 and beyond:

1. Adaptive digital services – in-cell devices or wing kiosks that don’t simply offer menus or book visits but learn from user interactions to improve accessibility and engagement
2. Predictive analytics for smarter operations – using AI to anticipate demand and improve resource allocation
3. Integrated data ecosystems – securely connecting justice, health, and rehabilitation data to provide a holistic view of an individual’s journey.

These innovations demonstrate just how AI can enhance transparency, improve outcomes, and make the justice system work better for everyone that it serves.

Towards a digital and ethical future

Across the UK, justice agencies and technology providers are collaborating to make this vision a reality. Companies such as Unilink, which develops secure digital systems for justice, are helping to turn AI’s promise into practical tools that support both efficiency and rehabilitation, from intelligent self-service and communications platforms to data-driven operational insight. The challenge now is to scale innovation responsibly, ensuring that, as technology becomes more capable, it remains accountable, explainable, and human-focused.

Digital innovation has always been about solving real-world problems. In the justice sector, this means using technology to promote fairness, reduce harm, and improve lives. AI is simply its latest manifestation. It’s not being introduced as a replacement for human interaction but as a way to make those interactions more meaningful, informed, and effective, and it must be guided by empathy, ethics, and human insight. 

In our view, the justice system of the future will be digital first, but human always, to enable a smarter, fairer justice system where data informs decisions, digital empowers people, and technology serves the public good. 

Read More AI for good

The post AI and digital innovation – Building a smarter, fairer justice system appeared first on Digital Leaders.

This AI will need to be accessible, transparent and yet intuitive to a wide application-domain experts, who have much common knowledge and constraints, as well as individual blind spots and baggage. This field is rather separate from common data-driven decision-making within data-rich applications, which is grunt, high-frequency, decision-making that has been applied in many areas. It is time for AI to move up to strategic and radical novelty-seeking challenges. 

Many distinct sectors are deeply interested, as there is almost always no prior experience nor relevant case data on the table when they have to make high-stakes calls.  Yet, many existing AI players and users are either conflating both straightforward and hard issues; believing (wishing) that the present AI data-driven paradigm is the only game in town; or else being simply focussed on doing what everybody else is doing (but hopefully better) – transferring supervised and unsupervised decision-making, anomaly detection, object recognition, and all types of data-driven inferences into different fields of applications. 

They are all merely swimming while the next generation AI, working on no-data crises, will be “flying”. 

 And, as Nietzsche said, “He who would learn to fly one day must first learn to stand and walk and run and climb and dance; one cannot fly into flying.”

What is under the hood?

Such AI must be a hybrid, combining a generative layer with a logical-foraging-evolutionary layer. Ideally, this accomplishes two things in response to any given challenge.

1.         Iteratively finding hypotheses (response options), while prizing possibly “novelty”. It is termed “imaginative” or “creative” since such hypothesis generation is not merely regurgitating and combining incremental hypotheses that are presently available.
2.         Refuting or validating any hypothesis, depending on whether it does or does not contradict any sector knowledge or constraints and accepted logic. The rejected hypotheses are termed “fallacious”; while the accepted hypotheses are termed “irrefutable”.

This process can maintain a growing archive of all (so far) irrefutable hypotheses; which is useful in successively defining “novelty” measures that drive types of Novelty search. There are many possible alternative approaches to various elements within both steps. But the principal aim is clear at a high level.

Hybrid AIs such as these have attracted some recent attention during 2025: at Vancouver ICML 2025, where the Exploration in AI Today workshop, where AI moves from exploitation (emulating intelligent functions based on its ability to recombine, paraphrase, and simulate, yet rarely discovering novel things) towards the exploration of new ideas and knowledge discovery.  

The role of government and public funding

Whilst there are obvious radical and entrepreneurial elements in developing next generation AI, the process by which the UK Government supports such innovation is all hobbled by adherence to consensus-seeking peer review; or else by pre-defining fields of interest and applications (thinking inside the box). At a high level, governments desire R&D of radical, distinctive, AI concepts and applications, within “Sovereign AI”, yet the mechanics have little appetite for risk. They should instead champion controversial R&D: as a taxpayer, I want every AI programme to have a policy of investing, say 25% of the resource, into ideas that would not have an expert consensus (and avoid groupthink); and that are so controversial that they would start a fist fight in a pub full of experts. 

It is a problem of framing. The taxpayers want risk and growth from potential high-impact and fail-fast investments, yet the programme managers want to avoid failures and to remove project and business risks. Consequently, the Government confounds its own strategic mission by top-slicing the ambition and risks, and very often deploys expert consensus to justify investment (as “excellence”) to HMT. Furthermore, by pre-defining strategic and focus areas of interest, it maintains groupthink and eschews radical ideas. The government only address the known unknowns. Yet the only thing we know about the unknown unknowns is that they are out there. No wonder that most disruptive paradigm changes and next-generation AI, usually emerge from the venture space. 

Read More AI for good

The post Explorative and creative AI appeared first on Digital Leaders.

The logic of scale is seductive. Organisations like OpenAI, Google DeepMind, and Anthropic argue that bigger models bring predictable gains in reasoning and creativity. But evidence is emerging that this assumption is starting to falter and that generative AI may not sustain the exponential demand for GPUs that so much global investment is built upon.

If that proves true, history offers a useful analogy. In the 1950s and 60s, computing was dominated by vast, centralised mainframes that only the wealthiest institutions could afford. The assumption then, too, was that bigger central terminals would always be better. The real revolution, of course, came later when compute power migrated from research labs to the desks of workers around the world.

Hyperscale models are the new mainframes

Today’s hyperscale language models are our new mainframes: powerful, but centralised, fragile, and controlled by a handful of global firms.

According to RAND (2024), the economics of foundation models are already “precarious,” with soaring compute and energy costs delivering ever-smaller performance gains. Beyond the economic and environmental implications of this, they also raise deeper questions about sovereignty.

AI sovereignty: Beyond data location

True sovereignty means controlling how data is processed, protected, and governed. That’s impossible when relying on models hosted or operated abroad. Even “UK instances” of foreign models cannot change the fact that model training and governance policies are decided elsewhere.

As Harrison Kirby, Author of “Exploring GenAIOps” and co-founder at Great Wave AI, has argued:

“AI sovereignty isn’t just about data location—it’s about control, governance, and accountability. Just think back to COVID-19, when even the closest allies competed for scarce vaccine resources. What happens to our ‘sovereign’ instances of overseas tech when priorities diverge?”

Beyond sovereignty, there are structural security risks. Large models must process data in unencrypted form — meaning every sensitive input, from defence intelligence to patient records, is briefly visible inside a foreign-controlled system. How long can that remain acceptable?

Finally, sovereignty must mean accountability.  In recent weeks, both AWS and Azure have experienced global scale outages created from errors incurred in the US.  Other countries have had to accept the problems but have had no ability to identify or mitigate the issues. This issue of being a passenger in someone else’s technology “vehicle” will only get worse in an age of hyper-scale GenAI.

The rise of distributed, sovereign AI

Fortunately, a new alternative is emerging. Advances in model compression now allow smaller, domain-specific models to run securely on local devices. These systems can operate offline, respect privacy boundaries, and collaborate through encrypted orchestration layers – like the UK’s orchestration challenger; Great Wave AI – which are the digital equivalent of an AI operating system.

This is an important historical parallel: control of the operating system has been the foundation of US technology dominance for four decades. Owning the orchestration layer, rather than the models themselves, could be Britain’s strategic advantage.

Practical examples of sovereign AI in action

We are already seeing promising pilots in policing where compact language models could be deployed directly into patrol cars, giving officers secure, instant AI assistance even without an internet connection.

These local models can handle sensitive data within police networks, providing decision support without exposing information to external servers. That’s what sovereign AI looks like in practice: distributed, resilient, and locally governed.

Britain’s opportunity: Lead in standards and orchestration

Britain’s opportunity lies not in building the next trillion-parameter model, but in building the system that connects them – the orchestration layer that determines how intelligence flows securely across our digital infrastructure.

This approach does more than avoid the hyperscale arms race; it positions the UK at the forefront of the next computing revolution, where value comes from interoperability, governance, and control.

By leading in standards, transparency, and secure federation, the UK can set the global benchmark for trustworthy, distributed AI ensuring that our future systems are not only powerful but sovereign and secure by design.

Read More AI for good

The post Britain’s future lies in sovereign, distributed AI — Not in building bigger models appeared first on Digital Leaders.

We’re treating AI procurement as a binary choice between build it ourselves or buy it from a vendor. But there’s a third option hiding in plain sight, one that could determine whether Britain becomes an AI powerhouse or just another customer. 

The problem we’re not talking about 

Recent research shows that UK organisations plan to increase AI investment by 32% by 2026. Yet nearly 90% report they’re not delivering customer value from their AI efforts. Meanwhile, 62% cite an urgent AI skills gap, with agentic AI capabilities most in demand. 

Think about what that means. We’re investing heavily in technology we can’t fully control, with skills we don’t have, to solve problems we’re still defining. 

The symptom that worries me most is that 83% of UK organisations report “shadow AI”. That’s British employees bypassing official channels to use tools that actually work. When your official path is too slow or too locked down, people find another way. They always do. 

But here’s the question nobody’s asking: Why isn’t our official AI delivering what the shadow version offers? 

Why “Buy” isn’t working 

I understand the appeal of proprietary AI platforms. One vendor, one throat to choke, predictable support. It’s an IT tale as old as time. Simple. 

Until it isn’t. 

The pattern is familiar, where initial promises sound great, then walls appear. Features move to higher tiers. Integrations get deprecated. Prices change, sometimes by 1000% after an acquisition. Your data lives somewhere you can’t inspect it, processed by algorithms you can’t examine, behind terms that shift whenever convenient for someone else. 

You thought you bought a solution. You actually bought a dependency beyond your control.

For government, this isn’t just expensive, it’s strategically dangerous. When operational control and autonomy are top cloud sovereignty priorities (cited by 72% of UK respondents in the same research) it’s important to recognise that you can’t build digital sovereignty on rented foundations. 

Why “Build” doesn’t scale 

Building everything in-house sounds empowering, but it’s a trap of a different kind. 

Every department reinventing authentication. Every agency building its own document processing. Every council creating bespoke chatbots from scratch. 

This isn’t innovation. It’s sprawl. It’s waste. And most of all, it’s bad for taxpayers. 

British tech talent, which we are told we don’t have enough of, gets consumed rebuilding commodity capabilities that already exist. Meanwhile, the truly differentiated work that could set UK government apart never happens because our best people are stuck reinventing wheels. 

The third way: Commoditise the common, differentiate where it matters 

Government should be asking “What are the common, underlying foundations that everyone needs, and how do we make those a shared utility?” 

Authentication. Document processing. Natural language understanding. Computer vision. These are becoming commodity capabilities. They need to work reliably, but they’re not where government creates unique value. 

The differentiated value, the work that truly serves citizens, comes from applying these foundations to uniquely governmental challenges. Better service delivery. Faster benefit processing. More responsive emergency services. 

Enterprise open source provides the bridge. 

This isn’t the “open source” that people interpret as “download it and figure it out for yourself.” Enterprise open source means professionally supported, fully integrated platforms built on open foundations. You get the stability and support of a commercial relationship, and you retain flexibility and choice. 

The code is inspectable. The community improves it globally. Security vulnerabilities get caught because thousands of eyes are watching, not just one vendor’s security team. And if you ever need to, you can take it in-house, fork it, or switch providers.

That optionality rebalances the equation. Vendors must earn your business annually, not just lock you in and extract value. 

What this looks like in practice 

When researchers optimised the Linux kernel and cut data centre energy use by 30%, every Linux user benefited. Not just premium customers. Everyone. That’s how innovation compounds. 

Enterprise open source foundations already underpin critical national infrastructure around the world, including the UK’s Ministry of Defence and National Energy System Operator. Linux. Kubernetes. The tools powering the internet. They’re proven, working, and reliable, but they are not getting the strategic attention they deserve in AI discussions. 

France and Germany built La Suite, an open source alternative to Microsoft 365, not due to vendor hostility, but from understanding that digital sovereignty requires controllable foundations. 

China is investing heavily in open source AI, and many would say that it’s not purely due to community spirit, but because whoever controls the foundations controls the future. 

What Government should do now 

Make enterprise open source foundations the default for government AI. Not because it’s cheaper (though it often is), but because it’s smarter. When something breaks, you can inspect and fix it. When you need to pivot, you can. When you need someone to call for help at 3am, experts are waiting at the end of the line. 

Invest in talent differently. Train civil servants on technologies they’ll own, not just operate. Skills built on open foundations transfer everywhere. Skills built on proprietary platforms trap people and waste the talent pool we have. 

Contribute to global communities. Join and shape the open source projects that matter. Help set tomorrow’s standards before they’re set for us. 

Think in layers. Commoditise the common infrastructure. Let hundreds of departments collaborate in a consistent way and stop rebuilding authentication or cloud platforms. Liberate them to focus on the differentiated services that only government can provide. 

The real question

Of the UK respondents surveyed, 83% believe the UK is or could become a global AI powerhouse within three years. That’s a smaller proportion than in Spain (99%), Sweden, Germany, and the Netherlands (all 98%). 

Our lower confidence isn’t just British cynicism. It reflects a real constraint. We’re building on someone else’s foundations, with someone else’s rules, at someone else’s prices. 

The AI revolution isn’t just about models and compute. It’s about who gets to use it, how it works, and who decides what happens next. 

The question isn’t whether to build or buy. It’s whether to depend on proprietary foundations that can shift beneath us, or invest in open ones we can control, improve, and build upon. Especially when this means enterprise support is readily available. 

The UK doesn’t need to reinvent AI. We need to commoditise the common parts and unleash our talent on the problems that matter. 

That’s not a technology decision. It’s a sovereignty decision. 

And, the window to make it is still open – for now.

Read More AI for good

The post Beyond build vs buy: Why UK Government needs open foundations for AI appeared first on Digital Leaders.

This isn’t simply about replacing physical cards with digital ones; it’s about creating the foundational layer upon which a more intelligent, responsive system of governance and commerce can be built. This infrastructure is crucial for realizing the immense potential of AI to transform healthcare, education, and business operations. Crucially, the success of this strategic pivot depends entirely on getting the infrastructure right from the beginning.

The potential for AI to transform how we deliver healthcare, education, social services, and business operations is immense.

However, the success of such initiatives depends heavily on getting the infrastructure right from the beginning. To ensure our national digital ambition succeeds, we must leverage international evidence, examining both successes and critical shortcomings.

What can we learn from these efforts, and how should they impact the UK’s AI strategy?

Download the Digital Leaders White Paper

Read More AI for good

The post Digital Identity – Lessons for the UK’s AI strategy appeared first on Digital Leaders.

The post Using AI as a critical friend: Redefining strategic leadership appeared first on Digital Leaders.

As data scientists, our role is to develop models that are accurate, efficient, and fair. However, when the AI systems are deployed into production environments, the importance of explainability increases, especially when a system’s decision-making can impact people’s lives. Helping our clients trust and understand the AI’s decision-making process is critical, particularly in highly regulated industries where ethical considerations are mission-critical. Equally, business leaders, product managers, and policymakers all need to be able to explain and justify the outcomes of AI models to customers, regulators, and the public, ensuring that AI systems are developed and used ethically and responsibly.

A lack of transparency can lead to reduced trust, legal complications, and unintended biases, negatively impacting both individuals and organisations.

In this piece, we explore how we use Explainable AI (XAI) techniques at Informed Solutions and how they can be leveraged to encourage a culture of transparency.

Why is explainable AI (XAI) important?

High Value, Trust Based Innovation

Users need to be able to trust the recommendations and decisions produced by AI systems to use them effectively, and if a user is unable to understand how an AI arrived at its conclusion, they are less likely to rely on it.

This lack of trust can have serious consequences. In healthcare, for instance, doctors may be hesitant to act on an AI’s diagnosis if they don’t understand the reasoning behind it, potentially leading to missed or incorrect treatments. Moreover, when users don’t trust AI systems, they are less likely to adopt them, which limits AI’s broader impact across industries. This can hinder innovation, prevent businesses from realising operational efficiencies, and slow down the transformation of industries that AI has the potential to revolutionise. Without XAI, therefore, the potential value of AI tools could be compromised.

Regulatory Compliance

Regulations such as GDPR mandate that AI-driven decisions be interpretable, especially in high-risk domains. Companies that fail to ensure AI explainability may face legal consequences, financial penalties and reputational damage, so it is essential that data scientists consistently uphold XAI as a standard when working on new systems.

Identifying Bias

The data used to train an AI model may contain biases, and if the decision-making process isn’t transparent, it becomes challenging to detect and correct discriminatory patterns. As such, Explainable AI helps organisations audit their models, identify and adjust biases, ensure fair outcomes and reduce the risk of harm from decision-making.

Improving Performance

AI models can sometimes generate unexpected outputs, and when they lack transparency, diagnosing and fixing these errors becomes difficult. Explainability, therefore, allows developers and data scientists to trace errors, refine model performance, and continuously improve system accuracy.

Five strategies to improve explainability

1. Choose Interpretable Models

Some AI models are inherently more interpretable than others due to their simple rules and relationships. Decision trees, for instance, are models that make decisions by splitting data into different branches based on certain rules. It’s like a flowchart where each node (or “question”) divides the data into smaller, more specific groups based on the values of the features (like age, income, etc). These decisions continue until the tree reaches a “leaf”, where a final decision or prediction is made.

At Informed, we follow a philosophy akin to Occam’s Razor—choosing the simplest, most interpretable model that still meets the required level of accuracy. For instance, when tackling a straightforward binary classification problem, we would favour a classical machine learning technique, such as logistic regression, over a more complex deep neural network, provided it delivers sufficient predictive performance.

2. Implement Explainability Techniques

For complex models, XAI techniques such as SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations) can provide insights into AI predictions by showing the ‘why’ behind the prediction or recommendation.

SHAP values, for instance, help explain how individual features influence a model’s prediction. In simple terms, SHAP assigns a “weight” to each feature, showing you exactly how much each feature (like “age” or “income”) influenced the final decision. This method is based on a concept from game theory, where each player’s contribution is calculated in a fair way.

LIME, on the other hand, works by creating a simple, interpretable model that mimics the behaviour of a more complex model for a specific instance or prediction. For example, if you’re trying to figure out why an AI recommended a specific product to you, LIME can create a simplified version of the AI for that one recommendation and show you which parts of your preferences (such as past purchases) were most influential in the decision.

At Informed, some of the problems we tackle involve complex relationships that necessitate the use of “black box” deep learning models. In these cases, interpretability techniques like SHAP and LIME enable us to look beneath the surface and better understand the factors driving specific predictions, helping us maintain transparency even with more advanced models.

3. Establish Transparent AI Governance

Organisations should establish AI governance frameworks that define clear guidelines for transparency. For instance, they should include guidance on documenting model development processes, as well as maintaining audit trails and ensuring explainability standards are met across AI implementations.

At Informed, our AI Charter commits us to putting ethics, safety, responsibility, and security at the core of everything we do. This means designing AI solutions that are safe, transparent, robust, and fair from the outset. Explainability plays a central role in upholding these values across our organisation. To support this, our “Well-Assured Framework” provides a comprehensive checklist that guides our team through key considerations, ensuring every solution we deliver is trustworthy and aligned with our principles.

4. Provide User-Friendly Explanations

AI explanations should be tailored to their audience. For example, a data scientist may need in-depth mathematical insights, whereas an end-user would require simpler, more intuitive explanations. Creating role-specific transparency, therefore, ensures that all stakeholders can meaningfully interact with AI systems.

Because we work closely with our clients, it’s essential that we communicate our models in a way that’s accessible and easy to understand. By prioritising explainability from the outset of a project, we’re able to offer clear, concise overviews of how models are developed, ensuring our clients remain informed and confident in the solutions we deliver.

5. Conduct Regular Audits

By setting up proactive alerts and tracking key indicators, such as shifts in data distributions, unexpected model behaviour, or signs of bias, we can respond promptly and appropriately. This not only helps maintain the integrity and performance of our models over time but also reinforces our commitment to delivering responsible and trustworthy AI solutions that adapt as real-world conditions change.

Conclusion

As AI usage becomes ubiquitous and continues to shape critical decision-making processes, ensuring explainability and transparency is a necessity. Organisations that prioritise XAI principles will not only comply with regulatory requirements but also build trust, mitigate risks, and enhance AI performance. Therefore, by adopting these practices, we can continue to ensure that AI systems are transparent, reliable, and more widely trusted and adopted.

Read More Bias &Ethics

The post The importance of implementing explainable AI appeared first on Digital Leaders.

I’ve walked the halls of Whitehall both as a civil servant and now as a supplier. And I can tell you this: the game’s changed. The speed’s cranked up. The shape of the blocks is different. And new power-ups have emerged that fundamentally alter the rules. 

The 2025 game-changers: New pieces, new physics

The core challenge remains: you have a board already crowded with legacy blocks and a stream of new digital demands falling from above that you must expertly slot in — and mixed in with these are rogue shadow IT pieces. 

But three revolutionary changes have altered the physics of the game. 1. The AI block: Your biggest opportunity and threat 

The single biggest change in recent years is the explosion of Generative AI. This isn’t just another falling block; it’s a polymorphic, unpredictable piece that can change shape mid-air. 

• As an opportunity, it’s the ultimate ‘line-clearing’ tool. AI can analyse and refactor legacy code, translate ancient programming languages, and automate testing, dramatically accelerating your modernisation efforts. 
• As a threat, it’s the most dangerous form of shadow IT you’ve ever faced. Staff using unsanctioned AI tools to write code, analyse sensitive data, or communicate with citizens can introduce profound security holes, data privacy nightmares, and algorithmic bias that your department will be held accountable for. 

2. The Platform power-up: Reshaping the board Itself 

The most forward-thinking departments in 2025 are no longer just playing blocks; they are strategically re-engineering the game itself. This is the rise of mature platform engineering. It’s about building a stable, automated foundation—a set of “Paved Roads”—that makes it incredibly easy for your teams to do the right thing. This platform provides the common components (identity, payments, cloud hosting, security monitoring) so service teams can build and deploy new, compliant digital services at a speed unimaginable just a few years ago. 

3. The economic “gravity”: No time for mistakes 

The post-COVID economic climate and relentless pressure on public finances have increased the “gravity” in our game. Blocks are falling faster. The political and public tolerance for multi-year, multi-billion-pound IT transformation projects that fail to deliver

immediate value is zero. You’re expected to clear lines—to show tangible results—far more frequently. This fiscal reality fundamentally challenges the wisdom of slow, sequential replacement strategies. 

How to play in 2025: A strategic approach beyond overhaul

A complete overhaul isn’t always feasible. But that doesn’t mean you’re stuck. The new rules of the game are about strategic, surgical interventions. 

From ‘tolerate’ to ‘actively contain’ 

In the past, you could “tolerate” a stable, non-critical legacy system. In 2025, that’s dangerously complacent. An unpatched but previously isolated system can become your biggest vulnerability when a staff member inadvertently connects it to an unsanctioned AI tool. 

The new rule: No system is an island. “Tolerable” systems must be actively contained. It means putting them in a secure enclosure, strictly controlling all data flowing in and out, and continuously monitoring them for anomalous activity. You must have a clear, costed, and regularly reviewed plan for their eventual decommissioning, even if it’s years away. 

• Example in practice: Following a near-miss where a legacy HR system was almost exposed via a shadow AI plug-in, a major department initiated an “Active Containment” programme. They used modern tools to create a security wrapper around the old system, logging every single access request and using AI to spot unusual patterns, effectively buying them time to plan a safe replacement while neutralising the immediate threat. 

From ‘optimise’ to ‘augment & automate’

“Optimise” used to mean a “lift and shift” to the cloud. It was a good first step but it’s no longer enough. The 2025 approach is to use new tools to fundamentally enhance your existing assets. 

The new rule: Use AI and automation to augment your systems and teams. This means using AI-powered tools to automatically refactor and modernise codebases, cutting down multi-year projects to months. It means wrapping legacy systems not just with basic APIs, but with intelligent APIs that can clean, validate, and enrich data on the fly. 

• Example in practice: The Department for Work and Pensions (DWP) has been a leader in tackling its vast legacy estate. Imagine them launching a pilot using AI-powered tools to translate millions of lines of critical COBOL code into a modern language like Java. This doesn’t just “optimise” the system; it transforms it from an untouchable black box into a modern, manageable asset, saving years of manual effort and significantly reducing risk. 

From ‘pace’ to ‘platform-led evolution’

The old idea of sequentially replacing systems at a certain “pace” is too slow for the demands of 2025. The modern game is played in parallel.

The new rule: Invest in your platform to enable continuous, simultaneous evolution. By providing a secure, automated platform, you empower multiple service teams to modernise their own applications at the same time. This is the vision outlined in the government’s DDaT Playbook and is being realised by departments like His Majesty’s Revenue and Customs (HMRC). In their tech blogs, HMRC engineers describe their move towards an “Internal Developer Platform” which streamlines the process of building and deploying new digital tax services, allowing them to focus on user value instead of wrestling with infrastructure. This platform approach is a direct descendant of pioneering GDS services like the GOV.UK Platform as a Service (PaaS). 

From ‘systematically replace’ to ‘strategically decompose’

The monolithic “big bang” replacement is dead. The financial and operational risk is too great in the current climate. 

The new rule: Don’t replace the whole thing; strategically decompose it. You treat the legacy monolith like a patient in surgery. You identify the most critical functions locked inside (e.g., a specific calculation, a case management decision) and carefully extract them, rebuilding them as small, independent services on your modern platform. You slowly and safely “strangle” the old system piece by piece, showing concrete value and reducing risk at every single step. 

• Example in practice: The Home Office, faced with its sprawling legacy immigration casework systems, has shifted to this model. Instead of a single “Future Border” programme to replace everything, they are delivering value slice by slice. They launched a new, standalone digital service for student visa applications. It works beautifully for the user and plugs into the old monolith on the back end to access necessary data. Over time, more and more of these slices will be carved off until the old system has nothing left to do. 

The grand strategist of the great legacy escape

The game has changed. The speed is faster, the pieces are more complex, and the stakes are higher. But you also have more powerful tools at your disposal than ever before. 

Your role as a leader is not to play every move. It is to understand the new rules of the game. It is to champion the shift from a passive mindset of the past to an active, dynamic fit for 2025. It means harnessing AI, investing in your platform, and empowering your teams to dismantle your legacy piece by valuable piece. 

The challenge is immense, but the opportunity to build truly responsive, resilient, and effective public services has never been greater. 

Don’t miss this Digital Leaders AI week webinar: Join Chad and senior experts from the Home Office, the Cabinet Office, and the Department for Business and Trade, as they share insights of what it takes to implement AI effectively and responsibly in government. Register now. 

Originally posted here

Read More Regulation

The post The great legacy escape is a game of Tetris… but not as you know it. appeared first on Digital Leaders.

We’re living through an “AI revolution”. And there’s no doubt that it is having an impact across every aspect of our lives. But as we get to grips with the implications of this disruption, I wonder if we’re looking at it through the right lens.

The post Why you might be focused on the wrong AI appeared first on Digital Leaders.