Algorithmic transparency in the public sector

This event-write up brings together the ideas generated at Reform and Imperial College London’s The Forum, policy hackathon ‘Algorithmic transparency in the public sector’.

In recent years, public sector organisations in the UK and abroad have increasingly made use of algorithms to assist in decision making. From moderating examination results to detecting financial fraud, hiring staff to allocating police resources, a growing number of applications for algorithm use has emerged in the public sector. Advances in this area can bring large benefits – algorithm assisted decision making may be more consistent and accurate than human decision making.

However, concerns exist that the use of algorithms to make complex, high-impact decisions – such as whether a person is eligible for benefit payments or social housing – may lead to unfair outcomes and reinforce existing biases. Many stress that technological innovation must be matched by a high level of public accountability and scrutiny over safe algorithm use. Greater transparency has been posited as a solution to a number of the problems posed by algorithm use in the public sector.

Transparent access to key information such as the data on which algorithms are trained and validated, their levels of effectiveness and bias, their documented effects on individuals and society, and the role that human beings play in the decision-making loop can help build public trust and bring necessary scrutiny to decision making processes. Establishing confidence in the use of algorithm assisted decision making will be crucial in allaying fears about the dangers of so-called “government by algorithm” – the idea that algorithms are increasing displacing human decision making in harmful ways.

Hackathon participants were asked to work through specific policy challenges related to algorithmic transparency across four different stages:

  1. The Early Design Stage
  2. The Development Stage
  3. The Implementation Stage
  4. Redress and Remediation

Reform and Imperial would like to extend their thanks to participants in the Policy Hackathon for their work and ideas to make this document possible.