Value-driven application of AI: how do we get next-generation digital services right for people and businesses

Monitoring the horizon

The first step is to carefully monitor the market and hype around the technology. Government employees and stakeholders must remain curious as to what’s out there and keep an eye on the fast-paced world of AI development. However, this must involve taking the inevitable ‘next big thing’ claims with a pinch of salt. Getting too caught up in every new AI development will naturally lead to organisational weariness and change fatigue.

The race to value in AI has a trail behind it littered in casualties (as reported here) that we can and should learn from. So, rather than trying to implement too much too quickly, the focus should be on maintaining an innovation function that can trial new developments adjacent to and on behalf of product teams and regularly share findings. GDS and i.AI are two great examples of this in action – maintaining excitement and curiosity in the technology while developing talent and helping teams learn from each other.

Remember that there’s plenty of great content available to help with your own AI experiments, including this extremely helpful resource from David Knott’s team within CDDO on Generative AI.

To drive additional benefit, look for opportunities for cross-departmental innovation teams or events (e.g. hackathons) where users can combine talent and solve challenges together. As we know, being a key element of the Civil Service Diversity and Inclusion Strategy, diversity of thought is a catalyst for success.

Focusing on value

With the exciting range of capabilities on offer, one could be forgiven for thinking that AI is a solution looking for a problem; the antithesis of the goal and value-driven development approaches many of us have come to recognise and passionately adopt. Perhaps instead we need to again assume our curious mindset and – recognising some common groupings of capability such as Image Recognition, Speech Recognition, Search, Predictions, Understanding, Content Generation etc. – look at Live Digital Products for opportunities.

We should be incorporating the User-Centered Design techniques and feedback loops that are commonplace in the development of Government Services to identify where the service and experience could be improved.

There are some natural candidates which could dramatically improve users’ experience (be it citizens or Civil Servants) with Digital Services. For example, content search is often a torrid experience despite accomplished search architectures and tooling existing for decades. What they often lack that AI neatly steps in to address is Natural Language Understanding. Through the implementation of a modern Generative AI solution architecture (like Retrieval-Augmented Generation or Agentic-AI), it’s possible for users to interact effortlessly with search.

For example, our Cyber Threat Intelligence Portal (CTIP) provides customers with up-to-date threat intelligence and has AI-enabled search to help users make faster and more informed decisions. This innovative feature, powered by Azure Cognitive Services and OpenAI's API, came as a result of our strategic relationship with Microsoft and was one of the first production implementations of this approach.

But this is starting to happen in Government too. The FCDO recently launched its AI-enabled search function in private beta, designed to ensure that citizens abroad can gain access to critical information faster than ever before.

Establishing justified trust

One of the most remarkable releases of 2024 comes from DSIT, an “Introduction to AI Assurance”, which provides a toolkit for assurance (Measure, Evaluate, and Communicate) and coins the phrase “Justified Trust”.

Justified Trust can be expressed as the intersection between trust and trustworthiness, where, according to DSIT, “a person or group trusts the use of an AI system based on reliable evidence”. Achieving this of course is no mean feat. It requires proof that solutions are safe, secure, explainable, transparent, fair and ethical – and that we have the relevant governance in place to monitor the development and delivery of AI technologies. This will depend upon the following aspects being in place:

• Excellent data hygiene (AI is only as good as the data pipeline feeding it)
• Robust security controls
• New (and exciting) testing techniques
• The generation of synthetic data (e.g. to address bias in source data)

Like many of our Government customers, we have been on a journey of experimentation and incremental adoption of new AI capabilities into production, with a focus on justified trust. To help our internal adoption, we developed our AI Assessment Service, and are now able to offer this to our partners across Government. The service provides a posture assessment against the AI assurance landscape, considering all phases of the AI development lifecycle to help customers build justified trust.

Ultimately, there are exciting times ahead. Embracing an approach that monitors the hype, identifies the most valuable use cases and is focused on establishing a scaffold of justified trust will be essential to ensuring that we as a collective build innovative, next-generation digital services that are right for businesses and for the people using them.