Published on the 19/02/2025 | Written by Heather Wright

While proposal to make HR AI high risk raises ire…

A trial of Micrsosoft Copilot at the Australian Treasury has produced mixed results, with the AI tool failing to meet staff expectations and not suitable for more complex tasks, but providing benefits for basic admin tasks.

A report by the Australian Centre for Evaluation into the 14-week trial notes a range of issues during the trial, which involved 218 Treasury staff, but while there are caveats aplenty – not least of all that staff became discouraged by Copilot’s performance and stopped using it – it noted the ‘relatively minor’ cost of Copilot licences meant only about 13 minutes of time saving through its use by mid-level staff was required to offset costs.

“The recommendations would introduce uncertainty, discourage use of AI-driven efficiencies and impose excessive new obligations on employers,”

The review notes overall usage of the product was lower than expected, with ‘unrealistically high’ expectations on the part of participants blamed in part, with some staff discouraged by the performance of Copilot and giving up on it.

Most participants used Copilot just two or three times a week, or less and while 75 percent of participants felt Copilot could support 25-50 percent of their tasks pre-trial, only 38 percent felt that was true post-trial, with a whopping 59 percent saying post-trial that Copilot could support little to none of their tasks.

Copilot didn’t perform as well as products staff had used elsewhere, though the report notes that was in part due to restrictions imposed by Treasury’s IT security environment, and time was needed to learn how to use Copilot effectively.

The report notes Copilot’s clearest benefits came from improvements in basic admin such as finding and summarising information, generating minutes and drafting content.

There was also no clear evidence that Copilot helped improve work outcomes, though the review notes ‘promising indicators’ and suggests the trial period may not have been long enough to provide definitive evidence.

The evaluation also notes a significant difference in how employees and managers viewed the effectiveness of the trial, with 59 percent of managers claiming no impact on work processes from using Copilot, compared to 34 percent of participants – 56 percent of whom said it had a positive impact.

While there’s little for Microsoft to celebrate, it can draw solace from Treasury’s discovery of ‘unintended benefits’ with Copilot contributing ‘to accessibility and inclusion for neurodivergent and part-time staff, or those experience medical conditions which saw them having time off work’.

Automatic summaries of missed meetings, and support commencing work where staff previously had issues doing so, along with ‘levelling the playing field’ for those struggling to navigate workplace norms or culture, were all noted and contributed to an increase in work confidence for some, particularly junior or new employees.

Indications of some reductions in workload stress also suggested ‘potential for generative AI’, the review notes and as the technology develops, along with skills, the tech should be monitored, it says.

The report includes a range of recommendations, including providing clear and specific use cases to distribute licenses to staff who can demonstrate likely benefits and time savings, and the managing of expectations along with training and support and guidelines around the use and disclosure risks of GenAI.

The evaluation report comes as another report warns about unchecked use of AI and automated decision making (ADM) in the workplace – and this one has sparked ire among some parties.

The Future of Work report was handed down by the House of Representatives Standing Committee on Employment, Education and Training following its inquiry into the digital transformation of workplaces.

It found digital transformation has exposed ‘significant challenges’ and calls for the government to classify AI systems for employment related purposes as high risk and subject to strict mandatory regulation. That use includes systems for recruitment, referral, hiring, remuneration, promotion, training, apprenticeship, transfer and termination.

Challenges include gaps in Australia’s regulatory frameworks and workplace protections and ‘a very concerning and excessive use of technology-enabled surveillance and data collection by employers’.

The report recommends updates to the Privacy Act and Fair Work Act to protect worker data and privacy, saying there has been a ‘surge’ in employers collecting worker data and concerns around its use and calls for a ban of high-risk uses of worker data, including disclosures to tech developers. It wants sale of workers’ personal data and any data collected in connection to work or undertaken during employment to third parties to be prohibited and ‘meaningful consultation and transparency’ with workers around surveillance measures and the data used in AI systems.

Also flagged is increasing use of emerging technologies to automate ‘not only tasks but decisions traditionally made by humans’, considerably impacting how work is organised and how the employment relationship is managed, providing both benefits and risks for workers.

The report includes 21 recommendations including to develop workforce capabilities and work with industry to increase SME uptake of AI and ADM, along with requiring developers to demonstrate that AI systems have been developed using lawfully obtained data that doesn’t breach Australian IP or copyright laws, with explicit protections for Indigenous Cultural IP.

It has been welcomed by some unions, but labelled ‘unbalanced and impractical’ by some business groups.

The Finance Sector Union says the recommendations strengthen workers’ voices and rights and says workers must benefit from the new potential and efficiency gains of AI technologies, rather than all benefits flowing to businesses.

But the Australian Chamber of Commerce and Industry says the report takes an ‘overly simplistic’ view of AI and other new technologies.

Andrew McKellar, ACCI chief executive, says the recommendations such as classifying all AI systems used for employment-related purposes as high-risk would create ‘heavy-handed obligations’ on AI deployment in workplaces and undermine genuine effort to raise productivity and says proposed amendments to the Fair Work Act are ‘unnecessary’.

“The existing framework already addresses potential risks posed by automated decision making or machine learning in the workplace. The recommendations would introduce significant uncertainty, discourage the use of AI-driven efficiencies and impose excessive new obligations on employers,” ACCI says. It adds that the recommendations would ‘overextend’ the remit of the Fair Work Commission into settling disputes about privacy, which the ACCI says is ‘not appropriate’.

The Business Council is also unimpressed by some of the recommendations and ‘the risk they present to increasing union power by stifling innovation, which will only further entrench Australia’s lagging productivity, impacting the cost of living of all Australians’.

“Harnessing the power of technologies like AI is imperative to fixing sluggish productivity growth and we need sensible regulations that safeguard communities, while also fostering a culture of innovation,” Business Council chief executive Bran Black says.

“At a time when we have significant productivity challenges, this union-backed approach will simply add more red tape and slow down technology take up – ultimately this is bad for workers and bad for business.”