As a growing number of agencies experiment with the use of artificial intelligence (AI) in operations and frontline services, public sector leaders from Thoughtworks and Amazon Web Services (AWS) talk to PublicTechnology about how to overcome key challenges and ensure successful outcomes
“Historically, most of the things that we’ve seen in the public sector have been based on machine learning. But then, overnight, we were all shaken awake by the arrival of large language models (LLMs),” says Scott Davies, Technical Director, UK Public Sector at Thoughtworks. “And the way you could engage with it was very different – you didn’t have to be a mathematician, you didn’t have to do lots of data science. You could literally type and get answers, and that was a revelation.”
Nearly two years on from the dawn of this new world, the growing interest in AI – including in government and the wider public sector – shows no sign of slowing down. Since LLMs went mainstream in late 2022, the government has released a guidance framework to support (and where necessary restrict) departments’ use of generative AI.
The state’s online shopfront, GOV.UK, is ramping up its trials of an LLM-powered chatbot to engage with citizens. And a range of individual public agencies have also started or expanded their use of the technology, including the Department for Working Pensions, Home Office and the Houses of Parliament.
Thoughtworks, meanwhile, has supported the United Nations in creating a Responsible Tech Playbook, which is intended to help the global diplomatic organisation ensure that emerging technologies are used in a way that promotes inclusion and transparency, and minimises potential harm. Specifically, the playbook emphasizes the ethical use of data and artificial intelligence by addressing issues such as algorithmic bias, data privacy, and transparency, providing strategies to mitigate risks and ensure AI promotes inclusion and minimises harm, you can find more information here.
Scott (pictured below right) says that Thoughtworks public sector customers are, “being really ambitious with their AI projects”. “Not only is it pushing forward the technology, it’s also pushing their engagement with citizens and other departments,” he adds. “They’re also openly having discussions about privacy and about the guardrails that ensure it gives safe answers. The government knows it needs to get to a place where the outcomes are repeatable. There are techniques now to improve that, and departments are rapidly adopting them and seeing positive results.”
Through its work with government departments, Amazon Web Services is now seeing a growing use of AI enabled by cloud, according to Jon Cook, Head of Specialist Sales Organisation – UK Public Sector at AWS.
“As we enter a new renaissance in technology and public service, it’s clear that we are seeing innovation across sectors. From enhancing national security to accelerating medical research and improving government services, the cloud and generative AI are empowering organisations to tackle some of the world’s most pressing challenges.”
Putting problems first
The government AI projects that stand out for their success and impact often take as their starting point “a really complicated process or set of rules”, where AI can be deployed in “navigating huge amounts of content or rules and getting to the result that you actually want quicker”, according to Scott.
“When you engage around a problem first rather than a solution, you’re more likely to get to a better place,” he adds. “And, if you can think about the world for the user – and what difference it will make for them when you’ve actually delivered, and however you’ve delivered it – that’s typically a better starting point: when you understand the constraints as well.”
For all the progress and positivity, public service entities – in common with organisations across all sectors – continue to face various familiar challenges in launching AI projects and, in particular, in building them out into wider-scale and more impactful deployments.
Scott picks out access to AI skills as an issue that has long been an obstacle for government AI programmes – albeit one that has alleviated somewhat in recent years, especially as technology platforms have become ever-more user-friendly.
“There’s also still the fear around where data goes, organisational boundaries, and how AI models keep up to date with all the latest guidance,” he says. “For example, if you train a model on tax legislation, how do you make sure it’s updated with all the latest information? And I think that leads on to what is often the real elephant in the room, which is maintainability – how you look after an AI system, and check that it behaves as it’s expected to.”
A challenge that is endemic to government is posed by funding models that are often built from money awarded as part of a three-year settlement – and then used to support projects in a world where the major platforms are undergoing significant change “every three to four months”.
To overcome such barriers, Scott advises that government agencies planning to embark on AI programmes of work should, from the outset, “think about what their challenges are… and certainly start with engaging citizens or dependent users really early”.
“Be really transparent about what you’re using it for, why you’re using it, and what happens to the data that goes into it,” he adds. “Think about how you are going to get humans in the loop to verify answers given by AI, and think about privacy as well.”
“Be really transparent about what you’re using AI for, why you’re using it, and what happens to the data that goes into it. Think about how you are going to get humans in the loop to verify answers given by AI, and think about privacy as well.”
Scott Davies, Thoughtworks
This should involve project leaders “getting on board early with security governance, risk compliance to get their input”.
Jon from AWS suggests that to successfully roll out AI, public bodies should identify and prioritise high-value use cases that align with specific business objectives and deliver measurable outcomes.
“Without a clear understanding of both the technology and how to measure business value from the proposed solution, public sector organisations may see their innovation investment return limited or unidentifiable.
Successful AI adoption follows many different paths, each of which is unique to the organisation’s mission and requirements. Across thousands of customers, we’ve identified a common sequence of events we call the AWS Cloud Adoption Framework for Artificial Intelligence, Machine Learning, and Generative AI (AWS CAF-AI).”
More than anything, organisations should not assume an AI project is necessarily going to be an AI project at all.
“First, think about what’s actually going to solve the problem – and there are certainly cases where AI is appropriate, but also those where it isn’t,” Scott adds.
One small step
As programmes proceed, ongoing best practice and support should replicate the kind of incremental approaches seen in any other digital project founded on Agile principles.
Teams should be “thinking about what the little micro steps might be, because then people can see progress and people buy into it more as they see it – rather than if you try to shoot for the end result from the start, which can very often disappoint”, Scott says.
And, as with any other technology investment, any deployments of AI in government should give careful consideration to “how are you going to update it and how do you avoid lock in”.
Scott adds: “You need to think really carefully about how you make those decisions so you can be independent of models and how you can switch between them when one has new or better features or is simply cheaper because the more tightly coupled to them you are, the harder that will be to change and you will either end up paying more for it, or falling behind.”
This kind of flexibility can be made easier by the use of public cloud, according to Jon (pictured below) from AWS.
“The cloud provides essential capabilities that enable AI and machine learning to be applied at scale. Cloud platforms allow organisations to store the large amounts of data required for foundation models cheaply and securely. Additionally, the cloud provides access to virtually unlimited specialised computing power needed for training models on that data. For organisations implementing AI solutions, cloud platforms like AWS offer comprehensive sets of AI services to build innovative solutions. This includes tools like Amazon SageMaker to build, train, and deploy ML and foundation models at scale, Amazon Q for accelerating software development and leveraging an organisation’s internal data, and Amazon Bedrock, a fully managed service that makes foundation models from leading AI companies available through a single application programming interface.”
For the foreseeable future, it seems clear that the public sector’s use of AI will only continue to increase.
In judging the success of these projects, Scott advises considering whether implementations achieve the intended outcomes, and how much time or money is saved. But there is another metric that is perhaps even more important.
He says: “If people start to trust more, see that it is helping them, it builds increased confidence and keeps them coming back – that is probably one of the best measures.”
Scott Davies – Technology Director – Public Sector, Thoughtworks
Scott is the Public Sector Technology Director at Thoughtworks, with over 25 years of experience driving technological innovation to address the unique challenges of public sector organisations. He has collaborated with a diverse range of government and public service clients to deliver impactful, citizen-focused solutions that create meaningful change.
Specialising in software delivery enablement, Agile transformation leadership, and fostering effective ways of working, Scott is skilled in stakeholder management, solution architecture, and hands-on software engineering with a focus on continuous delivery. A passionate advocate for open source and a dedicated mentor, he is committed to building high-quality software and empowering teams to succeed.
Jon Cook: Head of Specialist Services – UK Public Sector, AWS
Jon has over 25 years’ experience working for IT services, software and cloud vendors in sales leadership, go-to-market, and product management roles across markets in EMEA and North America. He joined AWS to build a team, helping to accelerate the cloud-native journey for large enterprise and public sector organisations. Initially focussing on the Cloud Contact Centre market, he then moved to lead a team helping customers across EMEA with modern workplace redesign through the use of cloud delivered desktops and applications.
In his current role, he has built an organisation with deep commercial and technology expertise across AWS’s specialist service domains such as GenAIML, Analytics, Data Strategy, Cybersecurity, Legacy Migrations and Modern Workplace. This team is solely focused on helping the UK Public Sector transform the delivery of citizen services utilising the latest innovations.
Jon brings an interest in operational execution and scaling businesses through partnerships with like-minded organisations from global systems integrators to local boutique ISVs.
