You are not permitted to download, save or email this image. Visit image gallery to purchase the image.
Dunedin barrister and researcher Warren Forster is ``frustrated'' ACC plans to introduce a new claims approval process without being fully transparent about the predictive modelling being used.
Controversies involving AI predictive analytics have recently hit government bodies, including ACC and Immigration New Zealand, where an AI approach prioritised overstayers for deportation.
Predictive analytics uses many techniques, including from artificial intelligence, to analyse current data to make predictions about the future.
University of Otago Associate Prof Colin Gavaghan said in May a new university Centre for Artificial Intelligence and Public Policy, which he co-directs, would discuss artificial intelligence (AI) issues with the Government.
Mr Forster commended ACC for taking steps to test their model, but believed more work was required ``before this model is rolled out nationwide''.
ACC recently released some documents about its planned new approval process, including a technical briefing paper.
But Mr Forster said ACC had not fully taken human rights and ``ethical considerations'' into account in preparing to implement the new system within a matter of weeks.
He believed that ACC claims about cover declines were based on an ``inaccurate dataset'', and he challenged the suggestion that ACC claimants had consented for their claims data (2010-2016) to be included as part of creating the model.
ACC spokesman James Funnell said the process being adopted by ACC was ``designed to immediately accept simple and obvious injury claims - cuts, bruises, burns, broken bones, etc.''.
Anything more complex was referred to a person to assess, and the new system ``cannot decline any claims,'' he said.
Earlier approvals meant clients could ``focus on their recovery''.
The model was built by experts, and then tested thoroughly.
``That included testing for bias in the data, and the testing satisfied us that the model outputs were not biased,'' he said.