资源环境科技发展态势分析平台

Greater algorithmic transparency in the public sector was the focus of a policy hackathon organised by The Forum and the think tank Reform.

Imperial AI and data academics joined a group of around 50 international experts including senior policymakers, regulators, industry leaders and think tank representatives, for solution-focused discussions on key policy questions posed by algorithmic transparency. 

The virtual policy hackathon, a format designed to enable creativity and collaboration for experts to think through concrete solutions to challenges, was hosted by The Forum, Imperial’s policy engagement programme, and Reform think tank. 

Applications of AI that seek to promote fairness and human flourishing

Tabitha Goldstaub, Chair of the UK Government’s AI Council, and Professor Aldo Faisal, Professor of AI and Neuroscience at Imperial, gave opening remarks at the event. 

Professor Faisal encouraged participants to use the opportunity of the hackathon to network and exchange ideas, and emphasised that in a domain that is so quickly evolving, this kind of thought leadership will have traction.

This was a perfect event to push cross-fertilisation of ideas on future and transparent uses of AI. This kind of cross-sector dialogue leads to the kind of changes that we all need. Dr Rossella Arcucci Imperial College London

Outlining some of the challenges when systems become increasingly complex and are able to make many decisions at scale, Professor Faisal said it may become unclear how overall patterns arise. He noted how engineers can set 'cost functions', which determine what an AI tries to optimise. 

These cost functions actually offer "tremendous potential for democracy", said Professor Faisal, and will allow citizens and voters to share the types of values they would like to see in society, that an AI can then effectively translate. When applying AI in society, politicians and policymakers need to help shape how we set these cost functions.

Professor Faisal, who is Director of the UKRI Centre for Doctoral Training in AI for Healthcare (AI4Health), spoke about the importance of AI skills and education, pointing to Imperial's strengths, such as the popular MSc in Artificial Intelligence, a conversion course for students from a range of disciplines, and the centres for doctoral training, such as AI4Health and Safe and Trusted AI, in collaboration with King's College London.

Tabitha Goldstaub, Chair of the AI Council, gave a keynote opening speech, where she mentioned the Council's AI Roadmap, published at the beginning of 2021, which set out a collective vision for AI development in the UK.

I'm really excited to see the results of your hard work and what innovative solutions you will be coming up with. I myself will be poring over those afterwards. Tabitha Goldstaub Chair of the AI Council

The Council, which includes Professor Nick Jennings, Imperial's Vice-Provost (Research and Enterprise), as one of its members, called on the government to build a UK National AI Strategy. In March 2021, the UK Government confirmed that it will publish a new strategy to unleash the power of AI.

Tabitha Goldstaub shared her thoughts on algorithmic transparency in the public sector to set up the discussions.

"Algorithmic transparency is a vital mechanism through which we can ensure that the law is being complied with. We need applications of AI that seek to promote fairness and human flourishing. This isn't easy, and in recent years we have unfortunately witnessed several cases around the world that have hit headlines where algorithms were developed in opaque ways and have led to harm.

"Greater algorithmic transparency is often touted as the solution to many of these problems, however it's not always the easiest thing to be transparent due to the technical limitations, particularly with machine learning algorithms and even more tricky with deep learning.

"I'm really excited to see the results of your hard work and what innovative solutions you will be coming up with. I myself will be poring over those afterwards."

Exploring practical solutions to the challenges posed by algorithmic transparency 

In the policy hackathon, a set of questions and problems identified by Reform were explored. Attendees were split into groups with a moderator and the aim was to come out with practical steps and solutions on an aspect of algorithmic transparency: 

1. Transparency in the early-stage design 
2. Transparency in the development stage 
3. Transparency during the implementation stage 
4. Redress and remediation 

Four Imperial representatives joined the discussions: Dr Rossella Arcucci, Research Fellow (Data Analytics); Professor Rafael Calvo, Chair in Design Engineering; Professor Aldo Faisal, Professor of AI and Neuroscience; and Dr Declan Mulkeen, Head of Development and Partnerships, AI4Health. 

This type of multisector discussion about the future of AI and its social impact is essential to guide work that leads to the type of improvements that society actually wants. Professor Rafael Calvo Imperial College London

Among the topics covered were what transparency requirements should exist on explaining why and how algorithms are being considered and used; how public sector organisations can ensure transparency over decisions made to those affected and effectively explain their basis to a non-technical audience; and how we can ensure that there is a wider public conversation on algorithmic use and that the public is aware of mechanisms to appeal algorithmic decisions. 

Dr Rossella Arcucci, Research Fellow (Data Analytics), jointly at Imperial's Data Science Institute and Imperial College Business School, said: "This was a perfect event to push cross-fertilisation of ideas on future and transparent uses of AI. This kind of cross-sector dialogue leads to the kind of changes that we all need."

Professor Rafael Calvo, Chair in Design Engineering at Imperial's Dyson School of Design Engineering, said: “This type of multisector discussion about the future of AI and its social impact is essential to guide work that leads to the type of improvements that society actually wants."

A post-event write-up will be produced as a result of the findings of the hackathon, which will be put forward to provide potential policy solutions. 

The Forum: Connecting our researchers with policymakers 

The Forum is Imperial College London’s policy engagement programme, connecting Imperial researchers with policy makers to discover new thinking on global challenges. Please get in touch if you would like more information: the.forum@imperial.ac.uk. 

Imperial College London is grateful for the support of the UKRI Strategic Priorities Fund for carrying out this work.