The commission is seeking to understand the role of the Commonwealth Ombudsman, whose 2017 report identified a number of flaws in the scheme but stopped short of declaring the “income averaging” debt calculation process unlawful.

This report was used by the former coalition government to defend the scheme continuing.

Former senior assistant Ombudsman Louise Macleod told the commission the investigation was triggered by a dramatic jump in complaints from people who had received Centrelink debt notices.

While social security complaints made up two-thirds of the total received by the Ombudsman, Macleod said they almost doubled between October and December 2016 after the scheme came into force.

“Increases in complaints are always a red flag for the Ombudsman’s office,” she said.

The Ombudsman’s investigation into the scheme flagged problems with internal review mechanisms used by the human services department when they received a complaint about a debt.

Macleod said it appeared the department was “circumventing proper administrative process” which meant complainants would not have the option to appeal their matter.

“Beyond that, you’re also talking about a cohort of people accessing the welfare system (who) don’t have the means to take it to the Administrative Appeals Tribunal or to the Federal Court,” she said.

“So we were concerned about that.”

Former Commonwealth Ombudsman Michael Manthorpe is due to front the commission on Wednesday afternoon.

The robodebt scheme ran from 2015 to 2019 and used income averaging of tax office data to calculate and raise debts.

The commission is examining how the scheme was allowed to continue, given significant concerns about its legality raised by early 2017.

More than $750 million from 380,000 people was unlawfully recovered through the program and the automated debt notices have been blamed for contributing to multiple suicides.

Meanwhile, Deloitte partner Elea Wurth – tasked by the commission with producing a technical study of robodebt – said a review found artificial intelligence was not present in the program.

Rather it was a “relatively basic” automation system that could not learn from mistakes or become more accurate over time, unlike other AI programs that have the ability to “self-learn”.

“The robodebt scheme did not have algorithms that allowed for any self-learning over time – it was very defined and specific to achieve a very specific goal,” Dr Wurth said.

“It was extremely rigid. Once those rules have been coded, the system itself stayed in place until a human comes in and changes the rules.”

Dr Wurth said the Deloitte review also did not find any risk management frameworks within the department of human services related to the program.

She said while there were policies in place regarding interactions between the two agencies running robodebt, there was not specific governance procedures for the program itself.

The Deloitte report recommended methods to improve trust with automated programs in future, including “human-centred design”.

“When you have a system that is interacting with Australians, especially those that are vulnerable, systems must be designed with the human at the centre,” Dr Wurth said.

-AAP