How is AI being utilised inside the Australian Government? An Australian parliamentary inquiry has provided a glimpse, with applications ranging from border security to scientific research.

A recent survey by the Community and Public Sector Union (CPSU) found that more than 40% of public servants across 60 agencies were aware of AI use in their workplace, though only 12% reported using AI tools themselves. The majority of current AI usage appears to be self-initiated rather than part of established work processes, the survey found.

Submissions to the Joint Standing Committee of Public Accounts and Audit: Inquiry into the use and governance of artificial intelligence (AI) systems by public sector entities, reveal a varied uptake.

The Department of Home Affairs emerges as one of the most extensive users of AI technology, employing it for critical functions including visa risk assessment, detection of fraudulent documents, and border control through SmartGates used at the border.

It also employs robotic process automation (RPA) solutions for document processing, including Freedom of Information requests.

The SmartGates use a combination of AI-enabled matching of faces to official documentation, which is complemented by rules-based decision making, where anyone not automatically allowed through is referred to human review.

The Department uses some rules-based Automated Decision Making (ADM) systems. However, it emphasized that no negative decisions are made solely by AI systems, with all potential rejections requiring human review.

“The Department is exploring potential use of AI to help staff members locate and summarise Human Resources (HR) policy information to support and inform staff management, for instance a HR Bot. No solution has yet been developed however a set of business requirements is being drafted.

“Increasingly, the Department is observing a trend in enterprise platforms such as ServiceNow, Pega, SAP and SuccessFactors incorporating AI capabilities into product roadmaps.

“Emerging capability in these platforms could provide some benefits, such as enabling users to submit requests in natural language which would support early identification of issues through trend analysis. However, the Department will assess each of these on a case-by-case benefit.”

The Australian Taxation Office (ATO) reported significant success with AI implementation, revealing that its natural language processing systems have helped raise more than $A256 million in liabilities and collect over $A65 million in cash since 2016 by processing 36 million documents to identify potential non-compliance.

The ATO has a large team of more than 100 specialist data scientists, with programming, machine learning and deep learning, and experience in model development, validation, deployment and governance.

“This team develop AI models and processes directly from source code for use across the ATO. With this capability, the ATO is typically able to build its AI solutions in-house, using pretrained models and open-source packages as a foundation. The ATO also has a strong focus on protecting ATO data and privacy when using AI, this means AI solutions are built within the ATO systems, including customisation of open-source packages. This enables the ATO to maintain ownership and possession of the source code for the AI models it uses, and to protect ATO data.”

Approaches to AI adoption vary significantly among departments. The National Archives of Australia (NAA) reported its use of AI is limited to experimental projects and basic use for drafting internal advice from public resources.

“There is currently no use that impacts our external-facing advice, products, or services.” the NAA stated in its submission.

“However, National Archives see potential efficiencies in using AI for collection enhancement such as description, transcription, linking data, collection analysis and discovery. There are also potential uses in records.”

The Australian Bureau of Statistics (ABS) has taken a more cautious approach, blocking access to public AI platforms like ChatGPT from its IT environment and limiting staff to using AI only with publicly available information.

It is conducting a limited trial of Microsoft CoPilot, but did not opt in to the broader Australian Public Service trial led by the DTA owing to legacy issues.

"Given the ABS is only in the early stages of migrating its information holdings from HCL Notes (formerly IBM Lotus Notes) to Microsoft SharePoint, there are less potential benefits from a broader scale trial” it submitted.

“The ABS has explored several applications of AI to better its understanding of the technology and associated risks. The AI use cases being explored by the ABS focus on productivity and reduction of effort, and do not involve decisions on service delivery.”

Use of AI in the Department of Agriculture, Fisheries and Forestry include Natural Language Processing (NLP) to detect specific biosecurity risks from imported cargo goods descriptions; and to analyse free text comments as part of a business quality assurance process. Outputs are not directly executed by software but are provided to inform regulatory officers of any potential biosecurity risk.

“Microsoft 365 Copilot has been trialled as a productivity tool for a limited number of staff; it acts as a digital assistant by drafting content, finding information on the department's intranet, editing materials for style and clarity, and summarising Microsoft Teams meetings.”

Microsoft Copilot remains the only AI system in use within the Department of Finance ICT environment.

“Its use is confined to low-risk use cases, adhering to both internal governance and the Digital Transformation Agency’s (DTA) AI assurance framework. Finance intends to continue this approach while staying receptive to any other safe and responsible AI applications identified in government contexts.”

The CSIRO is working with Google to develop AI-driven assistants for scientists across various disciplines.

It participated in the Australian Government MS365 Copilot trial but had mixed results.

“ ... while some staff have reported significant productivity gains, others have struggled to adopt the technology. CSIRO is optimistic that the benefits of generative AI will grow as the technology evolves and as organisations refine their approaches to integrating it effectively.”

IP Australia has implemented AI tools to help small businesses engage with the trademark system, alongside internal operational analysis tools.

A number of licenses for Copilot for M365 have been acquired to aid staff in their day-to-day work.

“Most of the legislation IP Australia administers (the Designs Act 2003, Patents Act 1990, Plant Breeder’s Rights Act 1994 and Trade Marks Act 1995) have permitted the use of a computer program to make decisions, exercise powers or comply with any obligations since amendments were made in 2018.”

Some agencies are also exploring novel applications, such as the Department of Climate Change, Energy, the Environment and Water's use of deep learning for monitoring terrestrial and marine species through video and image recognition. It is also deploying Generative AI across a range of areas including supporting media monitoring, research and content generation.

Most agencies emphasized their commitment to responsible AI use, with human oversight maintained for significant decisions and careful consideration given to data security and privacy concerns. The government is currently developing a whole-of-government framework through the Attorney General's Department to support automated decision-making systems across agencies.

National Archives is currently developing advice for Australian Government agencies to clarify their information management obligations under the Archives Act regarding records resulting from the use of Al systems and applications.