Imagine an AI that doesn’t sell products but helps rebuild after a disaster, or algorithms that don’t predict what you’ll binge-watch next, but pinpoint families in desperate need of help.

It’s not sci-fi, it’s the emerging frontier of AI for the public good – a movement driven by data scientists and policymakers who believe that AI is capable of solving some of society’s greatest challenges.

Among them is Dr Laura Gilbert, chief analyst and director of data science for 10 Downing Street, who gave this year’s opening keynote address to a packed audience at DIGIT Expo, beginning with her experience of bringing data-driven insights and evidence-based methods into traditionally rigid government departments, where fresh ideas face skepticism, and sometimes outright hostility.  

“People don’t necessarily want you to pitch up with a load of recommendations when they’re ‘experts’,” said Gilbert. 

“They already know what they’re doing, they’re really quite confident, and in some ways that’s unhelpful. It seems like they don’t want people messing around. So I started to really think, ‘what are we doing here?’”

Bringing data to decision-making

That question led Gilbert to consider not just the way that data is collected, or even how it’s used, but how data can be framed in ways that leads others to change their minds, and ultimately make better, evidence based decisions. 

That can be particularly tricky when dealing with people working in government, who come in with their own ideas about what the term ‘public good’ means in practice, and often with their own unique spin on the data. 

Changing preconceptions among policymakers has proven to be a challenge, and Gilbert is emphatic that it isn’t her job to tell those in power what to do or how to think. Instead, her role is to try and get them to look at the evidence and make informed decisions.  

“We have to be really careful about the way we frame data, evidence and information when we’re talking to people, if what we want is for them to make an objectively good decision,” said Gilbert.

For Downing Street’s data team, that means gathering immense amounts of information to build accurate predictive models, then finding ways to present those models that are accessible to people from opposing sides of the aisle, with different experiences and biases. 

“We talk to people about information in a way that we hope makes them more receptive to listening to it, more receptive to thinking a little bit differently, and perhaps making slightly different decisions.”

According to Gilbert, over the years that approach proved to be a successful way of weaving data and politics together, ensuring that, even if the results vary, the most important political decisions are founded in consideration of their potential impact.        

Driving public good through aspirational AI

Today, the role of data scientists in government is shifting. It’s no longer about persuading policymakers with graphs and figures, or simply presenting evidence to drive change.

With the rise of LLMs like GPT-4, the data models that were once at the cutting edge of decision making have been replaced by something more complex, immediate and impactful, but finding ways to use them hasn’t been a smooth process in the stilted, bureaucratic world of government.

“The first thing that happened [after ChatGPT was released] was that four people, or six people immediately declared themselves the government head of AI, and a lot of people were getting very excited,” remembers Gilbert.

“I sat down and thought to myself, we’re not really gripping the opportunity, because you don’t want to assume all of the risk and not pick up the opportunity of AI to try and improve public services, to try and save money.”

As others across government were jostling for position, Gilbert and a team of like-minded individuals set up the Incubator for AI, a programme designed to make the most of the emerging tech by focusing on the aspirational possibilities of its use.

Described as an ‘agile technical delivery team’, the project brings together experts from industry, academics, and the public sector to build AI tools that it’s hoped will improve lives, drive growth, and deliver better public services. 

Working within the Department for Science, Innovation and Technology, the Incubator is a means for the government to cut across multiple domains, using the technical skills of those involved to rapidly design, test and deliver AI products.

Speed is clearly of the essence, with the year-old project claiming that 11 million lines of code have been written for the products under development, and that its most mature tool already supports 9,000 interactions a month. 

Among the kind of projects the incubator is focusing on now are improvements to public facing services using genAI, fraud detection, triaging and optimisation, casework management, and the creation of data infrastructure.

As Gilbert herself pointed out, reading through that list of goals is a little dry, boring even, but they have real world applications that can bring huge benefits to some of the most marginalised people.

“Last year, we were asked to have a look at the asylum system, and the problem we were trying to solve is that a lot of people spend an awful lot of time in asylum hotels.

“If you’re a family raising children, spending two years in a hotel in a new country, that’s not a good experience. We want to put people into homes and communities much faster, and also save a lot of money.

“So, we built an algorithm that matched the people to the available accommodation, and it takes the amount of unused rooms from about 73% to well over 99% utility. So, you’re saving money and you’re also getting people to the right places.”

According to Gilbert, the central idea behind such AI tools is their versatility. At first, an algorithm might be used to help asylum seekers find permanent homes, but it can just as easily be attuned to finding and prioritising connections within the National Grid, a project that could save the public purse around £74 billion over the next ten years.

Making government more human with AI

Those are lofty ambitions, but Gilbert foresees AI as being more than a tool of the government to meet its promised policies around immigration or energy security. 

Among the applications she is most proud of, said Gilbert, is Caddy, an AI-driven assistant developed with Citizens Advice as a casework support tool. 

Up until now, those using Citizen Advice services, which often include some of the most vulnerable people, have relied on volunteers scouring confusing government webpages to provide guidance, with mixed results.

According to Gilbert, Caddy is proving to be a solution to this problem. By feeding in all the relevant information, the AI sets off to find specific answers to difficult questions, allowing Citizens Advice volunteers to give out balanced, informative advice. 

“We released our first trial in Manchester and actually the support staff really loved it, the advisors really loved it.”

“The results were really good. We had a 60% improvement in the number of calls that had positive outcomes for the public, and the advisors felt they were two and a half times more confident. It’s quite a bit faster as well, to get those answers back.”

Likewise, MedGuard, another AI application, is being designed to help pharmacists and patients manage prescriptions, as well as identify any risks with medication that might otherwise be missed, potentially saving both time and lives. 

The point of all these tools isn’t to replace people with machines, but rather to make room for more human to human interactions.

“We want to use AI to make government more human,” said Gilbert.

“It sounds really counterintuitive, because people tend to worry that AI is going to automate everything, and that they’re going to lose their jobs, but actually you really can and should use AI to get a lot more human interaction.”

Recommended reading

• Bank of England Survey Finds 75% of FS Firms Now Use AI
• FCA Warns Finance Firms About AI Fraud Risks
• What Key GenAI Advancements Will Shape the Future of Procurement?

Earning public trust with transparent AI

All of these impressive tools begin with the same foundational principles. They are safe and secure; transparent and explainable; accountable and with clear governance; fair, and open to both challenge and improvement.

Adopting these principles is perhaps more important for government run AI projects than those in the private sector, due to both the direct impact these projects can have on millions of lives, as well as the power that comes with having access to so much raw data to build these models on.

Without robust frameworks, AI can run rampant with disastrous results.

In 2020, hundreds of students staged a protest outside the Department of Education over the use of an algorithm to determine exam grades. This decision came after exams were cancelled due to the pandemic, and the algorithm, intended to predict grades, sparked widespread controversy with almost 40% of students receiving marks lower than anticipated.

In that instance, media attention forced the government to backtrack, but what about cases in the future where it’s not so clear that AI has introduced bias? How can the public possibly trust systems that most of us don’t even understand? It’s a problem that Gilbert is especially sensitive to, and one she is keen to show is at the forefront of mind.

“People trust government less than they trust marketers and advertisers, and the only way you get people to trust you and believe what you say is by being very transparent and very open.”

“We’re very open about what we do, so if anyone thinks we’re not doing it right, they can tell us”

That’s not a challenge, but rather an invitation, and one that Gilbert hopes people within the technology community take up. It’s a call to action for transparency, collaboration, and accountability, because for AI to truly serve the public good, it must first earn the public’s trust.