Uncertainty swirls about what might have been the world’s most imaginative use of artificial intelligence to deliver information and advice to members of the public. Nadia is a programme developed by the Australian Disability Insurance Agency (NDIA) to answer questions on a new national disability insurance scheme (NDIS) with the voice of actress Cate Blanchett and a computer-constructed face. The Australian Broadcasting Company reported on 20 September: ‘NDIS’ virtual assistant Nadia, voiced by Cate Blanchett, stalls after recent census, robo-debt bungles’. However, the next day the London Guardian announced that the ‘NDIA denies Cate Blanchett-voiced ‘Nadia’ virtual assistant is in doubt’. Needless to say, the NDIA’s website is silent on the subject.

The problems are both general and specific. The Australian Turnbull government has got cold feet on technology after the same bad experience with large IT projects as virtually every one else, magnified by somewhat insensitive – not to say incompetent – implementation. An attempt to move the national census online proved a disaster as the website crashed at the critical time, leading to recriminations as to whether it was the technology or cuts to the Australian Bureau of Statistics that were primarily responsible. The creation of an online automated system designed to collect debts due to the government which were detected through discrepancies in information given to different departments raised calls for repayment from 20,000 a year to 20,000 a week. This was not quite the success it might have seemed: many were wrong; sent to old addresses; and led to debt collectors demanding repayment as the first intimidation to taxpayers that they were subject to the attentions of the ‘robo-debt’ machine. It was all too clear that the attempted elimination of human supervision was just too ambitious. The Nadia project is budgeted at a significant Aus $3.5m (£2.04m, US $2.68m).  You can see why the Australians did not want yet another debacle of a similar kind.

There were also specific problems, however, with the Nadia project. This sounds fantastic and has attracted considerable attention. Cate Blanchett has donated her time to the NDIA; the developer is a New Zealand company FaceMe. Its main business is commercial. It is engaged in developing ‘an omni-channel digital employee platform: The embodiment of Artificial Intelligence with a range of defined skillsets, ready to learn and evolve to your business needs to solve real problems – delivering amazing experiences at scale…’ Or to put this more simply, it offers you a chatbot with a human face to answer customer queries  – initially as pre-programmed but with an AI capacity to learn. The NDIA is understandably excited by its collaboration and its head of technology authority Marie Johnson is quoted on FaceMe’s website as saying ‘What we are doing here with FaceMe and people with disability, is creating the new ‘world beyond websites’. So it may be but, interestingly, the FaceMe website emphasises the potential to integrate whizzy virtual presentations with old-fashioned physical interaction through a ‘human in the loop’, something all too often omitted by governments searching for savings.

Alas,  the arrival of this brave new world has been somewhat delayed. It turns out that IBM Watson, on which Nadia’s AI systems are based, is too slow. Answers lag thirty seconds after the question was completed. It now seems that the agency is looking at other providers and, indeed, IBM is racing to update its product. The NDIA’s line is that Nadia is only stalled. Chief Information Officer at the Australian Department of Human Services Gary Sterrenberg told the Guardian ’As the technology matures we will be making the appropriate decisions about it.’

The delay to Nadia raises the same question as the failure of the Dutch Rechtwijzer and disappointment at early versions of chatbots: are the reasons contingent to the individual project or systemic to the concept?

You could argue that this is just one more example of how technology reveals that it cannot deliver. However, more likely is that the project has fallen victim to the hype surrounding AI where future potential is too easily conflated with current performance. Just like the Rechtwijzer, Nadia has been required to fulfil expectations that were just too unrealistic in too short a timescale. In that case, the lessons may be threefold. First, keep the faith. These systems will ultimately deliver on their undoubted potential. And, in the field of access to justice, we need to monitor those projects like MyLawBC which continue to explore the possibilities. The second may be rather paradoxical and at odds with the first. If you are a government funder with one shot at getting it right, consider holding your hand until others have played theirs and you have seen the early difficulties surmounted by someone else. Oh, that our own online court project would have been so prudent. Third, FaceMe’s emphasis on integration with the human in a commercial context is significant. Governments, obsessed with savings, have a tendency to forget that. AI and technology has the capacity to revolutionise legal services delivery but only in co-ordination with human assistance.