A No10 data science unit could create more problems than it solves

By posting a job advert for someone to head a new data science team in No10, the government issued another statement of intent about its determination to make better use of data. The prime minister’s chief adviser, Dominic Cummings, has blogged extensively about his belief in the power of data and data science (as set out in his infamous call for ‘misfits and weirdos’ to work in government), and Michael Gove’s Ditchley Annual Lecture stressed the value of data and the need for improved numerical proficiency in government.

As we have said many times, better use of data would help across government, from procurement to parliament, and from people management to public service delivery. The focus of ‘10ds’ (which stands for ’10 data science’) on using data science to improve decision making is also welcome: our forthcoming report on how digital technologies could be used to improve government policy making covers similar ground. But this particular job advert leaves many existing issues unaddressed – and risks creating even more.

Government still needs to ‘fix the plumbing’ of data

Good data science requires good data – which many analysts in government are unable to access. Much of the data collected, stored and processed by departments is still poor quality and subject to significant gaps, difficult to find and share, and locked away in legacy IT systems. 10ds might be able to improve the situation on a very small scale, with a dedicated data engineering team to improve the quality and timeliness of the data available to the prime minister and the Cabinet, and capacity for collecting new data when necessary. But this is a drop in the information ocean. Building a well-rounded picture of government and society, and empowering the rest of Whitehall to use data science, will require an overhaul of data use way beyond No10.

The new Data Quality Hub (led by the Office for National Statistics, providing tools and guidance to improve data quality) and Data Standards Authority (led by the Government Digital Service, to create data standards and foster collaboration across government) should help the government, but as John Manzoni acknowledged in a letter to the Public Accounts Committee about the Challenges of Using Data Across Government in April, more work is needed. As we have argued previously, government needs a long-term plan backed up with practical steps, a much greater willingness to invest in skills and systems, and clear high-level leadership. The National Data Strategy, first announced in June 2018, needs to provide clarity on these fronts when it is finally published later this year.

The government should also state whether and when it plans to fill the government chief data officer (GCDO) and chief digital and information officer (GCDIO) roles that it first advertised for in 2017 and 2019 respectively, and explain what relation these will have to the work of 10ds. Without the leadership and coordination these roles could provide, 10ds – and the other new bodies – risk becoming just more initials in the alphabet soup of an already-crowded data landscape of multiple government bodies with overlapping roles and responsibilities.

No10 needs to keep data ethics front and centre

There is no discussion of ethics in the information pack accompanying the 10ds job advert. This is an unnerving omission. Awareness of the ethical challenges of collecting and working with data should be a crucial part of the role, particularly given growing concerns about biases in data, where datasets reflect the perspectives of the individuals, organisations and societies responsible for collecting them. Imagine a job application system based on data from a firm biased against female applicants, or skin cancer diagnosis systems trained only on white skin, as the Committee for Standards in Public Life has highlighted. If data is unrepresentative, incomplete or otherwise flawed, it will lead to damaging decisions. 10ds will need to work with subject matter experts in departments and beyond to understand the limitations of the data it is using.

The government still needs to earn the public’s trust when it comes to the use of personal data. The uproar around GOV.UK pushing to collect user data, a completely normal step for essentially any other website, is indicative of how delicately government will need to tread to avoid a backlash which could set back government’s use of data for years. Greater transparency about what data government is collecting and what it is using it for is needed, along with a greater willingness to engage in public debate about how data and AI should be used by government.

The unit raises questions about the relationship between Downing Street and departments

The job advert sets out how 10ds will offer ‘challenge and feedback’ to departments. Given the limitations of any data analysis work, testing assumptions and find weaknesses sounds sensible, but as we observed in our Behavioural Government report, ‘red teaming’ works best when it comes from someone within the same group, or from someone who identifies with it; otherwise it leads to defensiveness. The relationship between Johnson’s No10 and government departments has often been fractious, with SpAds and permanent secretaries forced out of their roles. Further ‘challenge’ from 10ds might not be welcome.

There is potential for 10ds to play a useful role in bringing the use of data to the heart of government decision making and in fixing existing problems. But a 10ds which antagonises the rest of government, fails to win public trust, and doesn’t improve the use of data across the whole of Whitehall, has the potential to cause more. No10 says it is looking for someone who could ‘look through a different lens’; for now, however, the job of the No10 data scientist requires rather more focus.