The results of the Sciencewise and Home Office funded AI in Policing public dialogue have been published today.
58 members of the public have been taking part in a mix of online and in person workshops to explore the considerations around the use of AI by police forces.
A diverse group of the public gathered to hear evidence from those with professional expertise in AI and policing, and to consider in depth what the key opportunities and concerns should be for policymakers in this area.
Results from the dialogue
Those who took part in the work identified four potential strengths for the use of AI in policing:
- Great potential. Rapid AI development has shown potential to improve future policing.
- Improved efficiency. Participants were impressed at the speed at which AI could accomplish tasks.
- Reduction in human error. Participants acknowledged that human error is inevitable in any job, and using AI in policing could mitigate this.
- Preventative not reactive. AI use was seen as a potentially powerful preventative measure.
They also identified four potential weaknesses they felt should be addressed:
- Lack of oversight. Without human oversight, participants felt far less comfortable with the use of AI in policing, particularly for decision – making. They expect the police to review the output of AI systems before taking a decision.
- Loss of human interaction. Participants felt AI was too limited in its ability to replicate human nuance and empathy, which is important in some policing use cases.
- Data security risks. There was concern about data breaches, particularly if their sensitive data is involved.
- Errors seem inevitable. AI has the potential to reinforce unjustified biases and produce errors. Participants expected the police to proactively manage this.
Participants also identified five core questions to be answered to ensure confidence in the use of AI in policing:
- What does ethical look like? Participants were undecided on whether AI would counter or exacerbate human bias – and unjustified bias potentially ingrained within police datasets. While there were arguments that AI would simply bolster already existing biases in police data, there were also counter arguments suggesting AI could work against human biases.
- What’s the evidence base? Participants expressed fear that there was not enough evidence that AI works effectively to confidently involve AI in high- risk areas within policing.
- What about job losses? Participants said AI should not be introduced to policing at the cost of humans losing their jobs. They hoped officers would be re-deployed.
- Who controls personal data? There were some discussions that came up throughout the process around who would own the data AI collects, with particular concerns around private company ownership and the risk of data breaches.
- Is this too much, too fast? There was a consistent and strong feeling of the inevitability of AI across society, and the feeling that AI is happening without the public’s involvement and approval, something the opportunity to be involved in this dialogue highlighted to participants.
Next steps
This is the first time that the Home Office has worked intensively with the public to explore views in this area, and data from the dialogue will be used to inform the policy approach toward when the police adopt AI and how to do it in a responsible way that has public consent.
The CoLab team in the Home Office, which initiates innovative ways of working with the public and developing policy, spearheaded this work with the Sciencewise team, working with delivery partner Thinks to organise the dialogue process.
CoLab is now exploring what other complex issues of policy might be suited to using participatory approaches to establish deeper engagement and inform policy making.
Click here to read a summary report.
Click here to read a more in-depth report of the findings.
