Everyone except the government knows childcare isn't predictable
Advertisement

Everyone except the government knows childcare isn't predictable

Mother-of-two Farrah Millar was diagnosed with breast cancer on the day her second son was born. And then her partner left her.

Faced with the prospect looking after two children under the age of five while undergoing chemotherapy, the 39-year-old turned to childcare and the government's subsidy schemes to help pay the costs.

The only problem was that she didn't meet the activity tests, which require parents to undertake at least four hours of work or study or volunteer in order to qualify.

She couldn't. She had to spend her time at the hospital.

Farrah Millar who was diagnosed with breast cancer on the day her second son was born.

Farrah Millar who was diagnosed with breast cancer on the day her second son was born.Credit:Sharon Gerschwitz, ShaBo STUDIO

Advertisement

The system doesn't like people like Millar. It is directed at parents who can tick an online box saying they expect to work three days a week or help out at a school canteen or Saturday sport.

Many fit this mould. Millar did not.

After days on hold and a fiery public blog post she was eventually approved for a full subsidy because of “exceptional circumstances". The government points out it does have a $1 billion childcare safety net in place.

But the risk is others in her situation may not have the know-how to lobby for the same result in a system that is becoming increasingly automated, unless they know about blog posts and are able to spend days battling bureaucracy.

Under sweeping childcare subsidy changes to come in from July 2, casual workers will need to predict their hours three months before their rosters come out - they can update them at any time - but some won't, leaving them with a debt to the Commonwealth when the Tax Office realises they have claimed childcare subsidies when they haven't been at work.

The Turnbull government's otherwise admirable reforms will work fine for many mothers but, like many government systems, won't be good at dealing with exceptions, even those that ought to be predictable.

Over time, it might become better. It was a prospect held out by then social services minister Christian Porter when he announced the measures that brought about Centrelink's robodebt in 2016. He said it would use technology that "learnt from previous experience".

It ought to be possible, given the unprecedented amount of data on work, study, health, and incomes being amassed on the MyGov website.

"Prediction takes information you have, often called 'data', and uses it to generate information you don't have," economists Ajay Agrawal, Joshua Gans and Avi Goldfarb write in their new book Prediction Machines.

“At low levels, a prediction machine can relieve humans of predictive tasks and so save on costs.

“As the machine cranks up, prediction can change and improve decision making quality. But at some point, a prediction machine may become so accurate and reliable that it changes how an organisation does things."

That transformation is now fundamentally reshaping our public childcare subsidies, as it has Centrelink and the National Disability Insurance Scheme before it.

But there's a problem: “Machines are bad at prediction for rare events," the Harvard Business Review economists write. And their biggest weakness is that they sometimes provide wrong answers that they are confident are right.

The Turnbull government's child care changes come in from July 2.

The Turnbull government's child care changes come in from July 2. Credit:Erin Jonasson.

Robodebt scanned the tax records of millions of Australians and matched them to their predicted earnings in order to create thousands of incorrect debt notices. It pushed many many of Australia's most disadvantaged people into unwanted financial distress.

If it had worked well, it could have emulated the reoffence prediction machines in the United States, which perform better than judges in deciding whether to grant bail to offenders who might reoffend.

An Australian version could combine the Centrelink, Childcare, Medicare and Australian Tax Office data now stored on MyGov to pre-empt the need for families to apply.

Agrawal, Gans and Goldfarb see data as the "new oil". And like oil in the early days of exploration, it should be almost free.

Until we develop such a machine here, many such as Millar will have to continue to navigate bureaucratic minefields and blindspots, suffering trauma along the way. Even when prediction machines eventually get most things right, the path for exceptions such as Millar won't be smooth.

"One major benefit of prediction machines is that they can scale in a way that humans cannot," the economists write. "One downside is that they struggle to make predictions in unusual cases for which there isn't much historical data."

That's where humans and machines can work together, the machines getting better at prediction all the time, and the humans providing them with information about exceptions.

Agrawal, Gans and Goldfarb say we are not only going to be seeing a lot more predictions, we are going to be seeing it surprising places: our shops, courts and schools chief among them.

Ross Gittins is on leave. 

Eryk Bagshaw is an economics reporter for the Sydney Morning Herald and The Age, based in Parliament House