The CDEI’s Review into bias in algorithmic decision-making
The CDEI has commissioned RUSI to carry out independent research into the potential for bias to occur in predictive analytics technologies being developed by police forces, as part of our wider review into bias in algorithmic decision-making.
The CDEI and RUSI held a roundtable last week which brought together police forces, civil society organisations, academics and policymakers to discuss the use of algorithms in policing, the potential for bias and how the regulatory and governance environment can be improved. This followed a roundtable held earlier in the week, in partnership with TechUK, to explore the role of technology providers in this space.
Attendees at the roundtables discussed a wide range of issues including:
Benefits and risks of predictive analytics in policing: One of the benefits discussed was the potential for predictive analytics to help police better-manage and gain insights from their data. However, the potential for biased outcomes against certain groups (if the algorithm were trained on historic police data) was noted as a risk.
Impact on individuals’ civil liberties and human rights: We discussed the important implications these technologies can have, if unchecked, on individuals’ civil liberties and human rights. There was consensus on the need to conduct meaningful public engagement before rolling out these tools, in particular with the groups most likely to be affected by them.
Challenges of sharing data across agencies: We discussed the need for police to work with local authorities in order to have access to greater datasets and develop tools to provide a better picture of trends, such as the drivers behind youth violence. Nonetheless, many attendees stressed concerns around how this could lead to further surveillance, data protection issues and be challenging to implement in practice.
The need for a consistent, nation-wide approach: The importance of consistency in developing this technology was emphasised. There were different views in the room about which body would need to drive this approach, but there were strong calls for clearer oversight and governance in this area.
The CDEI will use the findings of this research to co-develop, with the policing sector, a code of practice for the trialling of predictive analytical technology in policing. The code will seek to mitigate algorithmic bias and address wider ethical concerns.
RUSI will publish initial research findings in September, alongside the CDEI’s draft code of practice, which will be circulated for consultation. The CDEI will publish its final report on the review into bias in algorithmic decision-making, including recommendations to the Government, in March 2020.