Data Bites 61: trust, digital preservation, general elections and frozen pizzas

On Wednesday 21 October, Data Bites returned for its 61^st edition after a short hiatus.

For the uninitiated, Data Bites is an event series which aims to highlight interesting work on data (mainly from the public sector) and bring the different ‘data’ communities in and around government together. It’s been running since 2019, first at the Institute for Government (for 49 events) and then Public Digital (for 11 events).

As I pointed out in my introduction, much has changed since the first Data Bites in April 2019 – we’re on our sixth Prime Minister in the seven and half years we’ve been running – but the format remains unchanged. Four presenters each get eight minutes to present – there are eight bits in a byte, hence eight minutes in a Data Bite – followed by eight minutes of questions.

Watch the video

Anna Kwiatkowska: Trustworthy data and AI in HMRC

Anna, the chief data scientist at HM Revenue & Customs, spoke about the importance of understanding data, processes and systems before applying AI, and focusing on the problem you are trying to solve.

HMRC has a framework for thinking about trust in data and AI, which puts public value and purpose at the centre – one of the criteria is human oversight and control, which is not about approval of AI outputs but informed human judgment over AI-enabled decisions. This, and scrutiny, may look different depending on the use case, from drafting with Copilot, to complaints triage, to VAT risk modelling, the latter requiring more independent challenge and ongoing oversight. Understanding and accountability enables innovation – the question should not be ‘how much governance do we have?’ but ‘do we understand enough to act responsibly?’

Steve Daly: Re-imagining PRONOM

Steve, service owner (digital preservation) at The National Archives, talked about ‘Re-imagining PRONOM: A Sustainable Operating Model for Preservation Registries’. You can find PRONOM here; it recently won a Digital Preservation Award. Steve tells us more in his answers to the standard set of questions we send to all Data Bites speakers, below.

• Can you summarise your project in a few sentences?

This project is a radical re-imagining of the PRONOM registry which is the world’s database of information on digital file formats and identification signatures. It underpins most tools used by Digital Archivists to manage the long-term readability and accessibility of digital records.

• What problem are you trying to solve?

PRONOM is a critical resource to allow most archives to maintain going access to their digital records, but the system was nearly 25 years old and in need of refresh.  Rather than running a traditional IT application replacement project, we took a new approach to host the underlying data in an open public way, and make use of existing software development tooling for ongoing management of the data. 

• What difference will it make to citizens?

The system is publicly available but mainly aimed at archiving and digital preservation specialists. Mainly, citizens will benefit from the continued access to historic digital information, beyond the obsolescence of the file formats involved.

• How will you know when it’s succeeded?

We can quantify cost, environmental sustainability and security but most importantly we continually measure user satisfaction through feedback surveys on every page of the site, and through regular user engagement sessions.

• What are the main hurdles?

We had no dedicated funding or resource for this work so have delivered it alongside our primary goals. To be honest, working with zero budget actually focused the mind on a simple sustainable outcome. 

Louie Pollock: Describing (General) Elections

Louie, data science lead at the House of Commons Library, noted the impact a general election has on parliament’s digital estate (new people being added to systems, thousands of webpages having to change). But there is no official, centralised source of data (results are spread in different formats across local authority websites); it’s messy and not straightforward (boundaries change; by one measure, the Cooperative Party is one of the five largest in parliament but the parliament website doesn’t see it as such); and there isn’t a good historical archive of election data (it often requires going back to old BBC reporting with now-broken links).

The Commons Library team set out to develop an official archive. They came up with a better description of elections through a data model, trying to incorporate all the quirks of general elections. The resulting site currently goes back to 2010, but the team hopes to take it back further.

Frank Born: Spotting Red Lights

Frank, business consulting manager at Sopra Steria, explored ‘How AI can help reduce risks, improve compliance and make better decisions’. He used his uncle, a lorry driver, as a guide to our data and AI journey today.

Frank’s uncle spent so much time driving frozen pizzas from Germany to the UK that he thought it must be the only thing people ate in the UK – illustrating the perils of incomplete data, which can lead to perfectly reasonable conclusions that are completely wrong. Frank’s lessons focused on the quality of our data; actionable insights (the goal shouldn’t be more data, but better decisions); compliance (ensuring basic checks happen); maintenance (good maintenance is invisible, preventing problems before they occur); and experience. AI can  help spot the ‘red lights’ in our data.

Next time

Data Bites 62 takes place at 5.30pm, after Think Innovation for Government, on Wednesday 11 November. You can sign up for that here.

The final Data Bites of the year, number 63, follows Think Data for Government on Wednesday 2 December.