The Think Family Database in Bristol, England, aimed to predict criminal risk with vast datasets. Yet, its scorecards for citizens were often unreliable, raising questions about the accuracy and ethics of data-driven policing.

Launched by Bristol City Council and Avon and Somerset Police in 2016, the database gathered sensitive information from housing records to mental health files. Machine-learning models assigned risk scores to residents, but some models proved too inaccurate for use, leading to public mistrust.

John Pegram, a local police accountability leader, was baffled by his inclusion in an Offender Management App until he sought legal help. His case is emblematic of the broader issue: citizens often lacked transparency and control over their data.

The UK may soon follow suit with wider AI adoption in criminal justice. However, if even models like Avon and Somerset’s falter, will public trust survive? The answer might be as elusive as a risk score in a big data bucket.