Sanctions on nearly 2,000 Queensland drivers may be reconsidered as road transport authorities urgently review a “design fault” that led an AI-powered road safety camera system to erroneously issue double demerit points for alleged non-wearing of seatbelts.

Some 1,842 drivers were incorrectly given double demerit points after the image-recognition systems built into the mobile phone and seatbelt detection cameras – which have been deployed as fixed and mobile units in high-accident zones throughout the state for several years – detected that passengers had not been wearing seatbelts.

Queensland road regulations allow for double demerit points to be applied if drivers are caught driving while using mobile phones or driving without seatbelts, twice within a one-year period.

The errors occurred over 22 months between 1 November 2021 and 31 August 2023 and the issuance of double demerit points resulted in 121 drivers being suspended, 632 being put on Good Behaviour Bonds, and 11 suspended interstate or overseas licence holders being banned from driving in Queensland.

Some 1,036 Queensland drivers had been formerly sanctioned for the seatbelt offences, while 42 of the detected offenders had never been licensed.

The camera images are subject to human review before penalties are applied, the state Department of Transport and Main Roads (TMR) explains, noting that images flagged by the AI system are forwarded to “authorised personnel within the Queensland Revenue Office,” “[who] view these images to decide if an offence has occurred, and if an infringement notice is to be issued.”

In this case, the imposition of double points was traced to a flaw in the system, Minister for Transport and Main Roads and Minister for Digital Services Mark Bailey said in initiating the urgent review of the state’s Queensland Camera Detected Offence Program (CDOP).

“This should never have happened,” Bailey said in apologising to “every person impacted by this” and promising that TMR would “reverse the incorrectly issued double demerit points”.

“All offences were correctly recorded and financial penalties have been correctly issued,” he said, promising a “full and urgent independent review of this matter.”

“The design fault is solely in the application of double demerit points in specific circumstances.”

Facing the risks of automation

AI-based mobile phone and seatbelt cameras – which have been rolled out nationwide and around the world – have been flagged for a range of errors, including ticketing an Indian priest whose long beard was obscuring the overhead view of his seatbelt.

A review of the cameras’ rollout in Victoria found they had caught 1,162 drivers using mobile phones in a single month, as well as 2,993 seatbelt offences.

Due to be complete within months, Queensland’s new CDOP review is the sixth such evaluation of its operation and efficacy, with the latest – covering 2018 and 2019 and published in February – concluding that the program reduced overall crash numbers by 11.1 per cent in 2018 and 10.9 per cent in 2019.

The new review marks the first time that Queensland’s mobile phone and seatbelt cameras will come under scrutiny since they replaced manual police traffic stops in late 2021 – reviewing the efficacy of both the AI-based detection algorithms and the manual processes for reviewing their output.

Yet it’s not the first time the cameras, which photograph drivers’ laps, arms, and chests, have come under scrutiny: privacy advocates were pushing for a review of CDOP earlier this year, concerned that the indiscriminate photographing of car interiors was threatening the “sexual privacy rights” of female drivers – and that there were not clear enough privacy protections to ensure the images were appropriately protected.

Particularly in the wake of the Robodebt debacle, the use of automated systems for legally binding decision-making raises questions about privacy and accuracy.

With increased enforcement seeing CDOP take in an extra $200 million in fines during the 2022-23 financial year – a 70 per cent increase over the previous year – the accuracy of the program has become more important than ever.

“Regular recalibration of the evaluation models for the mobile speed camera program is important to derive the most accurate associations between levels of mobile speed camera enforcement and associated crash risk changes,” the last review noted.

“There will also be a need to carefully integrate [mobile and seatbelt cameras] into the CDOP framework.”