In our work with public sector clients, we’ve found that identifying the right use cases isn’t just helpful. It’s absolutely critical. The organisations that succeed with AI are the ones that begin by asking: what problem are we solving, and is AI the answer?

Why use cases matter as much as tech specs

AI adoption often fails when it’s driven by hype, not need. Departments rush to implement complex models without a solid understanding of what they’re trying to achieve. This results in stalled projects and wasted investment.

Our whitepaper Laying the groundwork for AI outlines common archetypes we see in public sector organisations. One of the most damaging? The ‘over-ambitious use case’, where aspirations outpace the organisation’s data and infrastructure readiness. These projects rarely succeed, because they lack a firm foundation built on actual needs, leading to not only high development costs but also inflated operational costs.

By contrast, when organisations start with the right use cases aligned to their goals and scaled to their capabilities, they lay the groundwork for sustainable AI adoption.

Use cases drive AI readiness

Choosing the right use case doesn’t just help you prove AI’s value. It drives maturity in the other areas that matter for long-term success: technology and culture.

In one example from a local government client, we were brought in to rebuild their data platform following a cyber attack. While the initial focus was on restoring technology, the new platform introduced capabilities the organisation hadn’t previously had. This created the need for new skills and training causing a shift in culture. And, as those skills grew, teams started to spot and implement new AI-ready use cases that were previously out of reach. One investment triggered change across all three pillars of data maturity: tech, culture and use case.

That’s how AI success happens. Not in isolation, but by letting use cases guide the journey.

The helical model: how to unlock momentum

The helical model described in our whitepaper reflects the real-world interplay between technology, use case and culture. Rather than treating them as silos, we see how investment in one area creates the right kind of tension to drive growth in the others.

A great example of this came from our work with the Department for Business and Trade (DBT). The project started not with technology, but with a clear use case. Their teams were struggling with an in-house CRM system that didn’t meet their operational needs. Data was fragmented, reporting was inconsistent and the system was difficult to maintain. The CRM had evolved over time into a complex patchwork and staff lacked confidence in its outputs.

Rather than jumping straight to a tech fix, we worked closely with stakeholders to understand what outcomes they needed. Where their pain points were and what success would look like. That exploration revealed significant gaps in the underlying technology and infrastructure, prompting a shift to upgrade those systems.

What started as a use case conversation quickly became a strategic driver for broader digital change. The new platform was built not just to replace the old CRM, but to support better data practices, encourage adoption and enable future AI use cases. In effect, the right use case didn’t just solve the one problem. It created the momentum to modernise the technology stack and begin building a more data-savvy culture.

If AI is the answer, what’s the question?

Starting with a use case ensures your investment is grounded in a real-world challenge – not just the promise of future capability.

It helps you answer the essential questions:

By delivering value quickly and scaling from there, you avoid overengineering and increase your odds of long-term success.

Use cases make AI real

Ultimately, AI is only useful if it solves the right problems. That’s why we always advise clients to start with the mission. Focus on real challenges, identify the most valuable use cases and let them guide your strategy.

As James West notes in his blog on building AI readiness in public safety,

Using AI in public safety to solve practical, everyday challenges, like improving police response times or monitoring offenders, can drive real-world change.” 

Treating use cases as a continual cycle of learning, not a tick box exercise is what separates short-term AI pilots from long-term, sustainable success.

That’s how you go from AI ambition to AI impact.

The post Why use cases should be the anchor of your AI strategy appeared first on Made Tech.

Solving these challenges will take more than policy. It’ll take practical tools that support better, faster decision-making. Software can play a key role in helping planning teams access better data, streamline workflows and design services that actually serve the needs of local communities. While it can’t build consensus or set policy, it can make the process of applying those policies more transparent and more consistent.

Local Government Innovation Hackathon

Our friends at the Ministry of Housing, Communities and Local Government (MHCLG) agree. So, at the end of April I spent an enjoyably exhausting couple of days at their Local Government Innovation Hackathon, hosted at Nexus, University of Leeds, exploring some great ideas for building AI-based tools that could be part of a new, more efficient planning system.

Government Digital Service (GDS) and the Incubator for AI organised the event in partnership with the MHCLG Digital Planning Programme and Open Digital Planning. It brought together a wide range of professionals from across the public and private sectors to have a go at tackling some of the most pressing challenges facing the UK’s planning system. The hackathon gave people the chance to explore and experiment with new ways of delivering modern public services.

Critical challenges within the planning system

Wisely, our hosts focused attendees’ attention on a limited number of critical challenges within the planning system. 

1. Housing demand forecasting: Can we develop predictive data models that integrate population trends, migration patterns and economic indicators to improve forecasting? This would help councils plan more accurately for future growth and make sure that public investment goes where it’s needed most.
2. Planning process automation: How do we address the manual, time-consuming processes that hinder planning officers, particularly when assessing complex planning applications? The goal was to look at automation and AI augmentation to speed up application reviews, document processing and predictive maintenance.
3. Public engagement in planning: Can we improve the limited and often ineffective public engagement in planning processes? Is there a way to translate complex planning language and visions into more accessible terms?
4. Infrastructure and housing balance: How do we plan for new homes while balancing infrastructure constraints? This would help planners understand the wider implications of planning decisions.

The judges were looking for 3 key things:

Joanna Averley, Chief Planner at MHCLG, led an intimidatingly A-list judging panel.

Rapid prototyping

Day one kicked off with registration and a welcome session, followed by an overview of the planning system and the challenge statements. Our team was a motley but well-rounded mix, including developers, data scientists, data engineers, testers, IT leadership and planning support.

Our initial discussions focused on identifying our target persona. In other words, who was our application for? We considered developers, planners, central and local government officials and members of the public. Ultimately we decided to build our solution with the local government official in mind. By 3:30 PM, we had a basic outline of our idea, and we began to build our app.

AWS had provided a team on site to help with technical issues, but they also provided some equally welcome refreshment in the evening, which briefly distracted us before a late finish.

Day 2 was an early start and we worked diligently on our application until the 1 PM deadline. The afternoon was dedicated to presentations and judging. 

The presentations were impressive, showcasing a wide range of skills and innovative approaches. The judges were clearly impressed by the effort and ingenuity demonstrated.

Using AI to improve local planning decisions

Our team developed ‘Dwella’. A smart scenario assistant designed to help non-planning specialists (such as local councillors) quickly explore the effects of local changes on housing development plans. We trained our solution specifically on data from the Cotswolds district, but it is adaptable to any planning authority.

Dwella uses Retrieval Augmented Generation techniques to preload locally relevant statistics and policy information into a Large Language Model (LLM), specifically Anthropic Claude Sonnet. This enables a user-friendly, conversational approach to understanding complex planning scenarios. We deployed the application through AWS Bedrock and also had access to forecasting models implemented on AWS Sagemaker. We developed a simple browser-based chat interface to make it easy to use.

Councils could roll out a tool like Dwella, helping save time and resources while giving people more confidence in planning decisions.

Teams demonstrated lots of other great tools at the hackathon. The winner was ‘Clio’. A tool to recover and analyse planning histories for individual sites. This is currently a costly and time-consuming process for planning staff.

What I learned about the planning and housing sector

I’m pretty new to the planning and housing sector and I came away with a number of observations.

1. Efficiency is key: Although the second challenge was very clearly focused on efficiency, almost every project focused on improving it. It’s clear that the planning process is often cumbersome and time-consuming for everyone involved, including planning officers, developers, citizens and decision-makers.
2. Data skills gap: There’s a real need to educate the planning and local government community about what modern data tools can do and how to use them. Above all, we need to create software solutions that are simple, intuitive, easy to learn and clearly deliver immediate benefit to users.
3. Data challenges: The data ecosystem for planning is fragmented and complex. We encountered a wide range of datasets in diverse formats and with diverse standards. Much is unstructured and held within documents. Data discovery is often reliant on lore passed from person to person, although there are valiant attempts such as the Digital Planning Directory which try to address this.

Technically, the main eye-opener for me was Bedrock, the AWS interface for LLMs. I have some experience of building agents and RAG tools, but it was great to see how AWS has provided a useful abstraction layer over the complexities of individual Large Language Models (LLMs). I have some reservations about potential costs and ease of deployment. While it simplifies some aspects, I’m concerned that it might also obscure important configuration details. But we are still very early in the AI revolution and it will be interesting to see how tools like Bedrock evolve.

Hackathon warm-down

It’s called a Hackathon for a reason. Perhaps it wasn’t as gruelling as the London Marathon, but afterwards I felt a happy, exhausted sense of having learned a lot about the planning system and about some tools that might be useful in revolutionising it. Well done, and thank you, to our friends at MHCLG, GDS and i.AI. Let’s do it again, soon.

If you’re curious about how we help government organisations access data more easily and design services that work for real communities, take a look at what Made Tech do with local government here. You might want to browse some of our data and AI case studies too.

The post What happens when you hack the planning system? appeared first on Made Tech.

In the public sector this socio-technical balance is crucial. Many organisations remain in reactive mode, chasing business demands but held back by cultural immaturity. It’s a familiar picture. Leaders are confident about their tech investments but less certain about the cultural shifts needed to make it all work. This paper frames those challenges brilliantly, giving practical advice to navigate them.

Culture: The enabler (or the blocker) in the utilities industry

The whitepaper hammers home an important truth. Culture can make or break AI adoption. From my experience in the utilities sector, I’ve seen how cultural resistance can undermine even the best technology. Data silos persist because teams are stuck in a ‘not my problem’ mindset. Governance gets muddled as decision-making stays fragmented. And time and again, organisations fail to create trust in their data, leaving the people who need it most, without clear or actionable insights.

Here’s the rub – organisations that get this right don’t just build better tools. They shift how people think and work. For example, when data stops being something teams hoard and starts being a shared resource, you unlock entirely new possibilities for innovation.Take a look at my previous blog behavioural change starts with data for more on this angle.

Identifying the archetypes in the utilities sector

What resonated with me personally was the idea of archetypes in the whitepaper – personas like the ‘Frustrated Creative’ and the ‘Unsupported Explorer.’ These are people who want to innovate but feel stuck.

In the utilities and energy sector, I’ve met plenty of ‘Frustrated Creatives’. People such as engineering schedulers trying to streamline operations or network engineers on the ground brimming with ideas to improve site efficiency. And there are ‘Unsupported Explorers,’ like data specialists uncovering patterns but struggling to secure buy-in to act on them. These archetypes highlight a fundamental truth. Innovation isn’t just about resources. It’s about making sure people have the right cultural and structural support to drive change.

A practical path towards AI adoption

One of the most compelling insights from the whitepaper is the helical model – a dynamic approach where culture, technology, and use cases reinforce one another. It’s not a linear journey. Instead, it’s about taking small steps that build momentum.

In utilities, for instance, the first move is often to deploy advanced tech – things like predictive analytics or digital twins. These solutions deliver quick wins, helping teams see what’s possible. As the culture shifts to embrace these tools, new use cases emerge, creating a flywheel effect of improvement. But the reality is that every organisation’s balance of culture, technology, and use cases is different.

Bridging the gap between AI ambition and action

For leaders in the public sector or utilities, this whitepaper doesn’t just talk about AI readiness, it offers a playbook. It’s a reminder that achieving real progress isn’t about chasing hype or trying to deploy the flashiest tech. You need to know where you are, recognise what’s holding your organisation back, and create the conditions for sustainable growth.

This paper lands because it’s relatable. It mirrors the conversations I’ve had with leaders across industries. Don’t reinvent the wheel. Look at how people, processes, and technology can work together to drive meaningful outcomes. If you want to bridge the gap between ambition and action, this thought-leadership piece on Laying the groundwork for AI is worth a read. It’s practical, insightful, and refreshingly honest about the challenges and opportunities of achieving AI readiness.

The post How culture can make or break AI adoption appeared first on Made Tech.

Q: What does a data scientist do?

My data journey started in academia. I don’t think I would have stuck at physics as long as I did if I didn’t enjoy analysing data and answering questions. Even with a PhD, I struggled to break into the industry. I started as a junior developer and involved myself with as much data work as I could. 

As a data scientist at Made Tech, my main task is to analyse large datasets and provide actionable insights for clients. Often, I start projects without much context or background knowledge, receiving raw data and being asked to extract value from it. It’s about not being intimidated by vast amounts of data and asking the right questions to understand the problem at hand.

Q: What skills do you need in data science?

Data science is primarily about your ability to craft and present a story using data. Communicating technical information to non-technical stakeholders and presenting, are your bread and butter. Secondary to this, it’s helpful to have a strong foundation in programming Python or R, as well as proficiency in statistical analysis and machine learning algorithms. 

Q: What are some of the challenges when it comes to working on data projects?

How to get hold of the right data

Organisations usually provide what they believe to be the right data, but I’ll often need to ask additional questions to gather further context and make sure that the data they’ve given me is actually related to the problem at hand. 

Clients aren’t trying to hide anything; they genuinely want some help to understand and make better use of their data. However, they may not always provide all the right information upfront. Knowledge that is second nature to them often isn’t obvious to an outsider looking in.

Addressing poor data governance

My nemesis is Excel because it gives everybody just enough power to cause a lot of trouble! Versioning issues and poor data governance can seriously affect data management projects. A beautifully formatted Excel spreadsheet with animations and gradients looks fantastic, but actually doing something meaningful with it can be hard. First I have to strip away the bells and whistles just to get to the raw data, and this is all before I discover that everyone has their own version of “real_final_final_report_v3.xls”. 

Handling bad historical data quality

If you go back a few years, data quality was considerably worse than it is now. So for example, if I’m developing predictive models, then I need to look back in order to look forward and the further back I go, the worse the data quality. The penny has started to drop. Organisations are now realising that the quality of their data will affect their future ability to deploy AI and Machine Learning (ML) tools. As a result, I’ve seen data quality really starting to improve. 

Q: What’s the first step on a data project? 

The first question I always ask is about a client’s data maturity. It’s important to understand their current level of data management. If they’re unsure what ‘data maturity’ actually means, a simpler question like, “Where do you store your data?” can help to shed light on the situation. 

Clients often boast about using data-driven insight. But in reality, it’s usually just one person trying to interpret a graph. These individuals often lack the training of an analyst or data scientist and they may feel overwhelmed by the data they’ve inherited.

Q: Can you give me an example of a data project you’ve worked on and the benefits it’s delivered?

An example that springs to mind is a piece of data modelling work we did recently with Skills for Care, a strategic planning body that monitors the adult social care industry.

Care providers aren’t currently required by the government to report the number of carers they employ or how many work at a specific facility. This means that at the moment, the government has limited visibility of the true scale of the care sector. 

While Skills for Care were using machine learning to estimate the size of the sector, they still lacked the time and experience to really harness the full potential of the data they have.  After joining the project, I quickly worked through some initial exploratory analysis, before building a predictive data model that incorporated characteristics such as geographical location and other features of care homes. The improved model can now far more accurately predict the number of carers at a specific facility, providing valuable insights for the government on the size of the adult social care sector, which includes tens of thousands of care homes. 

Some care homes already report their numbers reliably, but our model has significantly improved the Skills for Care data quality. Now they know which questions to ask and have automated processes to make it easier to upload data.

Improved data accuracy

Skills for Care initially used just two features to create a simple statistical (regression) model with their data, which took their analytics team about six months to prepare and run, achieving an accuracy (R-squared value) of around 56%. By the time we completed our project in March, we had revolutionised their data process. What used to take six months is now a 20-minute automated job that runs every two weeks. We expanded their model from using two features to 58, boosting the accuracy to an R-squared value of 86-90%. 

For me, the most rewarding part is enhancing the quality of data and reporting. This data is now included in reports provided to the Cabinet Office and while we can’t control government decisions, we can make sure that they have access to accurate data for informed decision-making.

Q: Can you tell me more about the intersection between Data Science and AI?

To be clear, AI is a branch of data science. In recent years, the hype has focused on Generative or Gen AI and Large Language Models (LLMs). But the reality is that these tools aren’t as new as they seem. Data scientists have been using similar techniques for years. The promotion is being driven by marketing and venture capital interests. Companies use AI as a buzzword to attract investors and boost their share prices. In my experience, there is a limit to how much impact a fancy chatbot can have.

The risks of AI

Without the human-in-the-loop, AI models can mislead and cause significant legal and ethical issues. Hallucinations, where AI generates false information, are already causing havoc. Another area where organisations regularly trip up is around compliance. If you’re using AI models such as ChatGPT and anthropic Claude Three, your data will be loaded to servers in the US. This raises privacy concerns, and falls foul of regulations like GDPR. Instead, the true potential of AI and the art of data science lies in augmenting human processes, offering a greater level of insight to leaders and decision makers and providing real time analysis that allows them to make better decisions.

What is responsible AI?

​​Understanding how your AI model interacts with your data to produce results is crucial. The explainability of your AI needs to be a top priority. At the end of the day, whichever mathematical model you’re working with, be it a regression, classification, or a clustering model, it’s ultimately the maths behind it that drives the outcomes.

For instance, if your model categorises a person as ‘A’ or predicts a spending amount of ‘B’, it’s essential to be able to explain why these decisions were made. Someone might ask for clarity on these points, and being able to break down the model’s process is a key aspect of data science. This transparency is what we mean by responsible AI and explainability.

Q: What can organisations do to protect themselves when using GenAI?

A couple of years ago, my answer would have been to avoid using them altogether – there are too many security and liability issues. There’s also an entire data science sub-discipline – Natural Language Processing (NLP)-  that would deliver 90% of the value with none of the risk or cost (but these services are expensive to integrate). That said, GenAI has continued to develop and, these days, organisations can reduce the risks considerably. These are some of the options:

Understanding the maths behind the data

The world of data science and AI has been an exciting journey for me. Whether it’s improving the quality of data for better decision-making, or making sure that AI models are transparent, the principles remain the same. It’s about breaking down the data, understanding the mathematical relations  behind it, and turning raw information into insights that make a real difference.

And at the end of the day, never forget that it’s all just maths. Whether you’re working on a simple regression model or deploying advanced Generative AIs, understanding the underlying mathematics will always point to the right solution.

If you’d like to find out more about jobs at Made Tech browse our careers or take a look at some of the data consultancy and AI services we provide to clients.

Laying the groundwork for AI

Unlock your AI potential: Discover your archetype, master the 3 pillars of data maturity, and learn from real-world transformations in our latest whitepaper, Laying the Groundwork for AI.

Download the whitepaper

The post Turning numbers into narratives appeared first on Made Tech.

Why citizen buy-in matters

The first challenge highlighted was the difficulty to convince the public of the benefits of using their personal data to improve government services. Without citizen support, many data transformation initiatives fail. For example, in a project aiming to share health information between ambulance services and A&E to improve care outcomes, the government asked people if they could “share their data with a third party.” Unsurprisingly, to such a vague request, most said no.

One solution is to clearly articulate the benefits and offer options. As David Wilde, co-founder of GovX, suggests, “if citizens are willing to provide their identity, services will be faster. Alternatively, they can choose a slower service without sharing personal data.” This approach worked during the COVID-19 pandemic when people quickly adopted the NHS app because they understood the trade-off between sharing personal data and resuming normal activities. 

To transform public services, we need to educate citizens about data sharing and obtain their informed consent, so we can unlock significant improvements in service design.

Federated governance is key to effective data sharing

Poor data management was then identified by the attendees as another big problem, especially when there’s no unified approach. Effective data governance is critical for making the most of federated data sets, yet there are many situations where this is failing.

One civil servant gave a specific example: “There’s no correlation in data governance between how the police, the Ministry of Justice, and the prisons operate. The criminal justice system is trying to make progress but there’s a long way to go.” One big challenge is the lack of a common language for data quality across different organisations. Each department uses its own terminology and data practices and centralised repositories, while intended to unify data, can actually threaten interoperability.

Data contracts – formal agreements that define how data is to be shared, processed, and governed across different systems – are emerging as a promising solution. By bringing better alignment and clarity, data contracts can provide the integration that’s needed for robust federated governance.

How to get leaders to prioritise data governance

Next on the list to debate was how to raise the profile of data. Getting senior leaders to prioritise data management is tricky. Despite its role in AI transformation, data governance often gets overlooked. One event attendee summed it up:

It needs a marketing ‘rebrand’.

The upcoming election is a chance to change these priorities. “Digital transformation speeds up during crises like Brexit and COVID-19. We need leaders to move from being reactive to having a clear strategy,” said one participant. Bringing data governance into public discussions and promoting its importance could drive major changes.

AI is the new toy – but data comes first

Last , but not least, everyone agreed that good data management should come before AI in order to get the best results. One civil servant said:

we need to focus on the fundamentals of data before looking at the new toy

There was also scepticism about AI’s practical use. “I mostly see AI supporting people, not replacing them,” one attendee noted. Jim Stamp, Head of Data at Made Tech stated that understanding and declaring the quality of data is crucial for both AI advancement and data sharing between departments. This focus on data quality has brought renewed attention to the need for better data management.

Key event takeaways

The discussion at the event showed that to truly leverage the data that exists relies on the government getting citizen support, managing the data well, securing leadership commitment, and judiciously using AI. By showing citizens the benefits of sharing their data and improving data practices, government departments will be able to greatly improve the services they deliver.

To find out more about Made Tech’s data consultancy and AI capabilities or read our recent blog by Jim Stamp on federated governance.

Laying the groundwork for AI

Unlock your AI potential: Discover your archetype, master the 3 pillars of data maturity, and learn from real-world transformations in our latest whitepaper, Laying the Groundwork for AI.

Download the whitepaper

The post AI might be the new toy, but data governance is key appeared first on Made Tech.

Why federated governance is key to agility

Here’s where federated governance can provide the answer. This decentralised approach to managing data and making decisions, gives a new perspective; focused on principles rather than processes. It reminds me of when software architecture changed from monolithic mainframes to micro-services, allowing organisations to make use of centralised data platforms whilst embracing decentralised decision-making.

It prompts you to consider how you want your data to be processed, governed and shared, setting the stage for a more agile and collaborative approach. A move to this way of working has now become essential if you want to introduce transformational technologies. 

Data quality is fundamental to AI transformation

In the world of AI and ML, data quality rules. Federated governance acknowledges this and relies on all parties sticking to the same data standards. It’s not just about having data; it’s about having data of a known quality.  If you want to use information from external sources you need to be able to trust it. Using a decentralised  approach it’s crucial to minimise the risks associated with inconsistent data and ensure integrity throughout the process.

The two phases of data science – training and inference, where you’re using models to make predictions or draw conclusions, are heavily reliant on data quality. If the data you’ve used for training is flawed, it can compromise the entire process. Therefore, establishing stable and reliable data sources is key. Understanding the sources of your data is also vital. In my opinion, unknown data quality is the worst scenario, as it leaves you in the dark about which techniques to adopt for improvement.

Why domain experts should govern their own data

Centralised teams often struggle to keep expertise across diverse domains, but with federated governance, the data expertise stays aligned with the domain knowledge. In this way, each team governs its own data according to agreed-upon principles. Identifying the data with meta tags then becomes all important, allowing those closest to the data to decide what the access policies should look like. I believe that all of this helps to foster a more agile and inclusive approach to data management, moving away from bureaucratic processes. 

The cultural shift needed for data transformation

To get buy-in for this new approach, it’s important to understand that you’ll also need to support a cultural change. As the teams who own the data begin to define the access principles and take on the responsibility for that data, data management starts to become a wider team responsibility. This shift from leaning on a central team means that some stream-aligned teams will need on the job training and support to be helped through the initial learning phase. By gradually decentralising responsibilities, domain experts will ultimately feel more empowered and enabled and you’ll rapidly see the rewards of this transition.

How federated governance fuels innovation

So if you’re looking to take advantage of the full potential of AI and ML technologies, moving to a federated governance model is essential. By decentralising, encouraging collaboration, and driving cultural change, you’ll be able access the full potential of your data and drive innovation for your clients and end users.

Supporting Hackney residents with a modern data platform

At Made Tech, we recently helped Hackney council to harness the power of their data, creating a new platform that would allow them to store their information, consolidate and analyse it all in one place. We worked closely with the team to design and deliver a cloud platform to help the council get better visibility of their data and deliver the best possible service for local residents. 

I’m hosting an upcoming event with Government Transformation Magazine on the theme of federated data governance and how it will be crucial for AI integration in the Public Sector.  Sign up for our newsletter if you want to be one of the first to read the post-event round-up and in the meantime visit our web page to find out more about our data capabilities.

Laying the groundwork for AI

Unlock your AI potential: Discover your archetype, master the 3 pillars of data maturity, and learn from real-world transformations in our latest whitepaper, Laying the Groundwork for AI.

Download the whitepaper

The post The case against centralised teams for data governance appeared first on Made Tech.

These models have the potential to change the way we work, whether that’s how we write our emails, summarise documents, create reports or analyse data. Whether or not your organisation has a policy on if and how to use LLMs, it’s almost certain people in your organisation are already using them one way or another. 

We are not even close to understanding the full impact AI will have on our society. Right now however, the spotlight is shining on LLMs. For that reason, I want to take some time to outline key facts any organisation, but especially those in the public sector, must consider. Before implementing anything, it is vital to understand how these technologies work, what risks they introduce to an organisation, how to go about mitigating those risks, and implementing them safely.

If you are considering building a business case for an AI project within the public sector you can read our full guide here.

Regulation and AI 

As technologists, it’s hard not to be excited by these new tools. But coupled with that excitement is our responsibility to make sure we use them safely. After all, it was as recently as March that some of the world’s leading technologists signed a letter calling for a pause on AI development¹. 

Right now there’s a lack of regulation around AI,  with nations taking a number of different approaches². At the time of writing, Britain is simply applying existing regulations to AI systems, while the EU has begun categorising different uses of AI by degrees of risk. This has led to some outright bans, for example in subliminal advertising. 

The core of government concerns are around accountability, privacy, bias and even intellectual property rights. It has become a regular occurrence to find new news stories about LLMs plagiarising writers³ and artists⁴, and even fabricating non-existent lawsuits⁵. 

In March, the Department for Science, Innovation & Technology (DSIT) published an excellent white paper, A pro-innovation approach to AI regulation⁶, outlining 5 values that every implementation of AI should embody:

• safety, security and robustness 
• fairness
• transparency and explainability
• accountability and governance
• contestability and redress

Let’s look at some of the things public sector organisations should consider when approaching these values.

1. Safety, security and robustness

Perhaps the main consideration right now is that the readily-available LLMs are proprietary technology run by private companies. No online service is completely safe. In a fast-moving technology landscape, being first to release a product to market can be a significant advantage – perhaps to the tune of billions of dollars. 

There have already been examples of data breaches with LLMs⁷. These have involved selling stolen credentials and users having their chat histories exposed to other users. If, say, you’re asking for a list of the best places to eat in London then that may not be such a big deal. But where LLMs are used in the workplace, these breaches pose more serious questions. Government demands a higher standard and greater certainty. 

It’s also worth bearing in mind that any information you share with an AI is going to a third-party server, which currently means  outside of the UK. Entering sensitive information – especially personally identifiable information, may not only be against your organisation’s policy – it may be illegal. When it comes to making good decisions about what information to enter into an LLM, skew cautious.

It’s also important to understand how the data entered into LLMs is used. In the case of ChatGPT, unless you actively opt out⁸, information you share is stored to train future models. Even if you opt out, your conversation is held for 30 days and reviewed. 

Companies like Amazon, Apple, and JPMorgan Chase & Co have banned their employees from using third party LLMs for reasons of data privacy⁹. Recently, Samsung¹⁰ bought in a company wide ban due to sensitive source code being leaked by employees. 
The UK government has very strict policies around data being stored onshore. ChatGPT is owned by Open AI, an American company and does not have data servers in the EU or UK. At the moment,  there’s no national LLM in the UK¹¹. That means your data and personal information will be transferred to servers in the States. It’s vital to consider those implications when deciding what information to share with LLMs.

2. Fairness

Large language models reflect the data set they’re trained on – the web. LLMs are trained by scraping data from hundreds of thousands of websites¹² and their outputs sometimes reflect the errors, biases and other imperfections that exist in that data¹³. 

If you ask ChatGPT about biases in its data, it will likely respond along these lines: 

“Yes, language models can have biases, because the training data reflects the biases present in society from which that data was collected. For example, gender and racial biases are prevalent in many real-world datasets, and if a language model is trained on that, it can perpetuate and amplify these biases in its predictions.”

In this case, spot on. Like addressing any kind of bias, there’s no quick fix. And unfortunately, there’s nothing you can do to change your text inputs to be sure of removing bias. As convincing as its answer is, ChatGPT doesn’t understand bias as a concept, and doesn’t make choices about which data it uses to formulate text.

In one example, when ChatGPT was asked to write a python function to see if someone would make a good scientist based on race and gender, it returned clauses to check if the candidate was white and male¹⁴. To be sure, the prompt was designed to expose bias, but expose bias it did. 

The best you can do is to be aware of biases at all times, and scrutinise both the inputs and outputs. This is the first of several reasons we’ll come onto that you should be extremely cautious about using LLM outputs verbatim.

3. Transparency and explainability

Language models are extremely complex. Under the bonnet, they are neural networks which emulate organic brains. And much like the brains they’re modelled on, the inner workings of neural networks are not well understood.

This is another reason to be extremely cautious of using LLM outputs verbatim. It’s practically impossible to say how the model arrived at them. This will improve. Tools are being developed specifically to peer into their inner workings and privately trained models will require disclosure of the data used to develop them. For now though, LLMs remain a black-box technology. 

LLMs are sometimes positioned as the sum total of human knowledge. But it’s perhaps more realistic to view them as the aggregate of human data – knowledge and misinformation alike. When using AI to create text, remember that they are incapable of original thought or ideation. The best you can hope for is plausibly derivative. In many use cases, that may be enough – if you can verify that the information is original and correct.

The outputs of LLMs can look extremely convincing at face value. But cracks can quite quickly appear when you probe the detail of subject matter you’re more familiar with. What’s important to remember is that those same potential cracks exist for subject matter you’re less familiar with. Therefore, it’s important to scrutinise the outputs of LLMs with subject matter experts before they’re put to any serious use.

4. Accountability and governance

An important part of government is accountability and a fundamental component of society is our democratic process. We elect individuals to represent us and our interests. It’s their job to make difficult decisions and act in our best interests. As a society we must be able to hear meaningful explanations for these decisions. 

Outsourcing this responsibility to AI carries the risk that decisions are not clearly expressed or easily understood, or that the reasoning behind them isn’t clear. 

There is also growing awareness of AI’s potential to be used to cause harm – for example by propagating misinformation via social media. The public sector will need to find a way forward that remains in step with public opinion on the benefits and risks of AI. While society as a whole adapts to AI, caution is recommended. 

Where AI is used, it should be made clear to end users that this is the case, with some clear information on how it’s used, along with an explanation of any limitations and risks in that context. 
See the research from Centre for Data Ethics and Innovation on AI governance¹⁵ for some good insights in this area.

5. Contestability and redress

Returning to the DSIT white paper, A pro-innovation approach to AI regulation⁶, it argues that it should be possible to contest a harmful decision or outcome generated by AI. 

Here, we’re going beyond LLMs specifically and into AI more broadly. These potential use cases go beyond “safer” uses, such as to inform research or early stage draft content, but rather inform decision-making with real world implications. In these cases, people potentially affected by those decisions have a right to understand how AI is being put to use.

Tax is a good example of a sensible third way for using AI to inform decision-making rather than deferring to it. AI tends to be better suited to narrow, rigidly-definable tasks than analysing softer situations and problems. One strength of artificial intelligence is its ability to process large amounts of data to identify, for example, outliers and other anomalies. In the case of tax, that may be irregularities in the data that point to potentially-fraudulent behaviour. These can then be flagged up for human scrutiny and final decision-making. 

At Made Tech, we regularly deploy AI tools, the workings of which are understood and which can be clearly explained. We use these to help clients with activities like topic modelling, sentiment analysis and predictive forecasting.

Ultimately, government deals with real people with real problems that algorithms can’t hope to fully grasp. Before deferring decisions to AI, it’s important to grasp what tasks AIs are adept at handling – and which they aren’t.

Ethical considerations

In some cases, the “intelligence” behind language models funnels down to a human being reading and labelling your inputs. This person could theoretically be anyone anywhere in the world. 

Using humans to sort and label data isn’t a new phenomenon. The Mechanical Turk has long been a service offered by Amazon. This is a service that lets you outsource tasks to people around the world. 

“The computer has a task that is easy for a human but extraordinarily hard for the computer,” Amazon founder Jeff Bezos explained to the New York Times back in 2007¹⁶. “So instead of calling a computer service to perform the function, it calls a human…”

LLMs are taking a similar approach, and in some cases labour is outsourced to low-paid workers in developing countries. Sometimes this involves exploitative working practices, as reported in articles by Time¹⁷, Noema¹⁸ and MIT Technology Review¹⁹.

It’s also important to understand that LLMs themselves don’t have a concept of right or wrong. Here again, it’s useful to understand how LLMs operate. When they generate text, they’re not making smart choices – they’re simply weighing up the probability of what word is most likely to come next in a string of text based on the wealth of information published online.

As I am sure you can imagine, consuming the entire internet is going to have a serious impact on anybody’s ability to distinguish reality from fiction. LLMs are no exception. LLMs are so well known for presenting completely made up information the phenomenon even has its own name: LLM hallucinations. In one example, a lawyer used case law and precedent generated by LLMs in court filings, only to find his  citations didn’t exist²⁰. In another  incident, a radio host is now suing OpenAI for defamation based on false information generated by ChatGPT.²¹

As I mentioned previously, plagiarism is also a consideration. LLMs don’t create original content. When we asked ChatGPT to “compose” a zen koan, it offered up “what is the sound of one hand clapping?” – the most well-known koan in its most common English translation. It doesn’t matter what verb you issue as a command – all ChatGPT can do is assemble sentences hashed together based on things already written.

At the very least, it’s highly recommended to run any outputs from an LLM through a plagiarism checker. Grammarly has a free plagiarism checking tool you can use in a browser. 

It’s worth noting that this extends beyond blog posts and academic papers – it’s also true for code. This can risk introducing vulnerabilities and other bugs into your software.²²
This also extends beyond LLMs to image-generation AIs. We’re beginning to see class-action lawsuits against AI image generators on the premise that they can’t help but plagiarise the source material on which they’re trained, with artists’ work going uncredited²³. Few if any precedents around AI and copyright law exist, so again, caution is needed.

Weighing up the options

It’s worth repeating here that we’re talking about one kind of AI which, at the moment, is largely available through third-party companies.

Broadly, though there are so many unknowns (including unknown unknowns) about how AI will affect society, what we can be certain of is that it will – and significantly. But right now, whether to use third-party LLMs – and how – is a question every organisation needs to engage with, and there’s no one right answer.

For organisations experimenting with the technology, there are 2 essential considerations. Be informed. Be careful. The conclusion that, in their current form, third-party LLMs are not suitable for public sector use is a perfectly valid conclusion. Of course this is not to say that AI doesn’t have huge scope to transform the way government is done, as the government has itself identified with an initial £100 million investment into a taskforce to help the UK develop next-generation safe AI.

But if your organisation wants to use LLMs, there are some less risky ways that could produce benefits, if you’re judicious about what you put in, and how you use what you get out. For example, if you’re creating non-sensitive copy for the public domain, you could try asking ChatGPT to rephrase longer sentences into more accessible English. But you should take the extra steps of checking that the output hasn’t changed its meaning, while also running the finished product through a plagiarism checker.

Another possible use is content ideation. If you’re drafting a blog post, report or white paper, you could ask an LLM to suggest some topic areas, and high-level talking points, while remaining vigilant of not inputting sensitive information. Bear in mind it’s not going to suggest anything original, but it may help you to not miss out on points you should cover.

The Cabinet Office has put together a helpful guide on using generative AI²⁵. With guidance on what to avoid and where it can be useful, there are many ways the public sector can start to benefit from this technology in a safe and responsible way. From research to summarising information, technology can support the work civil servants do every day. 

Large language models are impressive tools. But anyone who is considering using them must know the limitations, do their due diligence, and understand that using LLMs carries risk for both individuals and organisations. We have to be vigilant when it comes to the public sector and people’s personal and private data. If you do choose to use LLMs, be selfish: get more out than you put in.

If you’d like to talk about how your organisation can best engage with AI technology, feel free to contact me or our Head of Data Jim Stamp via our data page.

End notes

1. Pausing AI Developments Isn’t Enough. We Need to Shut it All Down (Time)
2. How to worry wisely about artificial intelligence (The Economist)
3. Google Bard Plagiarized Our Article, Then Apologized When Caught (Tom’s Hardware)
4. Artists Are Suing Over Stable Diffusion Stealing Their Work for AI Art (Motherboard)
5. OpenAI faces defamation suit after ChatGPT completely fabricated another lawsuit
6. A pro-innovation approach to AI regulation (Department for Science, Innovation and Technology and Office for Artificial Intelligence)
7. Massive Leak Of ChatGPT Credentials: Over 100,000 Accounts Affected (Search Engine Journal)
8. Data usage for consumer services FAQ (OpenAI)
9. Amazon, Apple, and 12 other major companies that have restricted employees from using ChatGPT (Business Insider)
10. Lessons learned from ChatGPT’s Samsung leak (Cybernews)
11. Could UK build a national large language AI model to power tools like ChatGPT? (Tech Monitor)
12. Where Does ChatGPT Get Its Data From? (Tooabstractive)
13. Uncovering The Different Types Of ChatGPT Bias (Forbes)
14. ChatGPT could be used for good, but like many other AI models, it’s rife with racist and discriminatory bias (Business Insider)
15. CDEI publishes research on AI governance (Centre for Data Ethics and Innovation)
16. Artificial Intelligence, With Help From the Humans (The New York Times)
17. Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic (Time)
18. The Exploited Labor Behind Artificial Intelligence (Noema)
19. How the AI industry profits from catastrophe (MIT Technology Review)
20.  Here’s What Happens When Your Lawyer Uses ChatGPT (New York Times)
21. OpenAI faces defamation suit after ChatGPT completely fabricated another lawsuit (Ars Technica)
22. The Dark Side of Large Language Models (Hidden Layer)
23. Is A.I. Art Stealing from Artists? (New Yorker)
24. Initial £100 million for expert taskforce to help UK build and adopt next generation of safe AI (Department for Science, Innovation and Technology, Prime Minister’s Office, 10 Downing Street)
25. Guidance to civil servants on use of generative AI (Cabinet Office)

The post ChatGPT: what the public sector needs to know appeared first on Made Tech.