Skip to the content

Promoting data science and AI in public policy


Mark Say Managing Editor

Get UKAuthority News


Dr Cosmina Dorobantu
Image source: The Alan Turing Institute

Dr Cosmina Dorobantu finds it exciting that technology can do some things that humans struggle with, especially integrating knowledge from various sources and making sense of it.

This underpins her work as one of the co-founders – along with Professor Helen Margetts – of the Public Policy Programme at The Alan Turing Institute, the national institute for data science and artificial intelligence.

A big element of this is in using the capabilities to bring together those strands to feed into policy making.

“It’s something the public sector struggles with,” she says. “It has various policy areas as we saw during the pandemic, with epidemical models and economic models that don’t talk to each other but work in siloes.

“This is where the potential is as we have the technologies and methodologies to bring them together and create more generalised models for policy in which you can see how a health intervention will affect the economy and the other way around.”

Ethical foundations

Dorobantu has been with the institute since 2018 when she and Margetts set up the programme to help organisations use data science and AI to solve long running policy problems and develop ethical foundations for the process.

“We wanted to create a version developed with public sector problems in mind, which are among some of humanity’s most consequential problems, and I think these technologies are well positioned to help,” she says.

The programme is based on four challenges: using data science and AI to inform policy making; improving the provision of public services; building ethical foundations for the use of data science and AI in policy making; and contributing to policy that governs how they are used.

An important element of all these is in the complex interrelationships between factors in the policy field.

“Policy as a whole is an interrelated field,” she says. “When the chancellor decides in the next where they invest money, that invested in health could also affect other areas like education and labour markets.

“All of these are interrelated; there is no policy you can implement in a void. Policy is an integrated whole but we are used to doing it in siloes.

“That’s a key example of what we need machines to do for us. Our brains can’t do that very well.”

Team resources

The public policy team is funded by the Turing, which in turn receives backing from sources such as UK Research and Innovation and funding calls for specific programmes, and has about 50 researchers. Dorobantu says this provides some slack in system to provide a reasonably quick response to a government body wanting to deal with an issue, and enables it to work on some projects for free.

It also collaborates with academics in the UK and overseas, providing a national resource for work in the field.

“No other country in the world has something like this at the moment,” she says. “They may have centres within universities but do not have national centres with groups of dedicated researchers for government work.”

She sees a growing appreciation of its potential for the public sector, with more organisations now having dedicated data science teams, and most being keen to learn more about what that and AI can do for them. But there are barriers to overcome.

“Just getting the data you need for a project is still a problematic area and the public sector could do much better. There is still some siloeing between policy areas and I would like to see that brought together.

“A lot of time you see departments working on a policy area internally. For example, if I want to build a machine learning system there is a tendency to look at what others have been built in their department and try to adopt it to their problem. But you can often learn more by looking across to another department. There is greater scope for collaboration; these technologies can be generalised for other questions.”

Star projects

Asked about stand-out projects since the programme was set up, Dorobantu cites ongoing work with the Department of Business, Energy and Industrial Strategy on aggregating data for a model of labour markets, and its work with Ofcom on the Online Harms Observatory, a dashboard of insights on the subject.

She also highlights its work on policy measures to bring more women into data science, and says she is most proud of its development of guidance on the use of AI in the public sector. This has been followed up with its contribution to the government’s development of principles for regulating the use of AI and the Turing’s lead role in the pilot of the AI Standards Hub.

“We’re moving from leading the national conversation on AI ethics to leading the international conversation,” she says. “We’ve been an integral part of UNESCO’s first international agreement on AI ethics, and been involved with UNICEF’s ethical guidelines on children. There are so many huge projects in that space.”

She is also aware of problems that still need to be solved, citing the example of local authorities trying to use machine learning to identify children at risk when they are still struggling to understand all the ethical implications, and when the data available does not provide a complete, unbiased picture.

Cautious stance

“I’m nowhere near convinced we have the right data or tools to make the right decisions on children’s lives. I wouldn’t dare, unlike some of the private companies, at this point to tell local authorities we could build a system for them.

“It’s not just the technology, it’s the data itself. A local authority will have data on pieces they have had in their areas, but we know that minorities are over-represented in the datasets, and a middle class white kid at risk of harm is not as readily picked up.

“The data also does not pick up the good bits about a child’s family environment; it’s a system to identify risk and harm but not the good things. Questions like this make you worry.”

But this is not restraining the ambition of the programme. Projects in the pipeline include the development of tools to help government bodies harness machine learning – for which there is not yet a timeline – the work on the AI Standards Hub with the British Standards Institute and the National Physics Laboratory, and modelling on topics such as the housing market, international trade and wealth accumulation.

Underpinning all this is the aim of encouraging public sector bodies to share what they can, and to build their own capabilities in the field.

“The nice thing about being a research institute is that we are not like a consultancy, building the same system for 20 different departments,” Dorobantu says. “My researchers want to build a system and move on. We want to tell them to learn from what other people have done. We want to feed into what will improve the work for them; train them up and hand it off to them.”

Register For Alerts

Keep informed - Get the latest news about the use of technology, digital & data for the public good in your inbox from UKAuthority.