Senior public servants think GenAI will boost productivity – but are worried about the risks
- Written by: Helen Dickinson, Professor, Public Service Research, UNSW Sydney
Many bold claims have been made about Generative Artificial Intelligence (GenAI) and its capacity to improve productivity[1] and generate workplace efficiencies[2].
A recent Microsoft survey[3] found 24% of private sector leaders have already deployed GenAI across their organisations. Many are considering laying off staff and replacing them with GenAI systems.
But how much appetite does the public sector have for using artificial intelligence, which doesn’t come without risks[4]?
Our new research[5] explores attitudes in Australian bureaucracy to using GenAI in policy work. Given governments are expected to work in ethical, transparent and responsible ways, we wondered if public servants are more wary of adopting this technology.
No single view
We asked senior bureaucrats from 22 state, territory and federal government agencies about their views on GenAI. We focused on what this might mean for the future of decision-making, policy development and public services.
They expressed a range of views on the transformative potential of GenAI. Some were enthusiasts who saw the potential to conduct government work faster and more reliably.
One interviewee remarked:
Why improve the candle when you could use a light bulb?
Others were less enthusiastic, arguing the technology is overhyped. Critically, they see GenAI as fundamentally inappropriate for use in public policy work and inherently risky on several fronts. These include:
the tendency for AI to hallucinate[6], where tools see patterns in data that do not exist in reality, making outputs inaccurate or wrong
the risk of biases[7] in existing datasets, such as the underrepresentation of some groups or people
the sensitive nature of government data that might be compromised by AI programs.
Regardless of their specific views on GenAI, public servants consistently told us two things.
First, they do not believe artificial intelligence will replace workers. Instead, they are confident these tools will augment their work by freeing them from routine and repetitive tasks. This would allow them to focus on high-value tasks, such as engaging with the public.
Second, the current use of GenAI is largely focused on administration tasks that do not draw on sensitive client data or interact directly with the public.
Robodebt hangover
One of the consequences of the Robodebt[8] scheme is the pace and scale of the adoption of automated tools.
Many interviewees explained public sector organisations are still very cautious about using GenAI technology as a result of the scandal[9].
One interviewee told us the majority of the problems with Robodebt were at a human level, which highlights the importance of individuals
taking their duties, both professionally and ethically, seriously, and interrogating what they get out of AI systems.
Close attention is also being given to the influence of human decision-making in the development of machines that use GenAI.
Incremental change
Our research suggests public service agencies are largely taking a careful and measured approach to applying GenAI in policy work. Senior public servants perceive the public is wary of how governments use these tools. Rebuilding credibility in relation to technology oversight and implementation is imperative.
Public servants described most of their use of GenAI as purposeful experiments. Clear outcomes are set for the use of these tools and evaluation processes are in place to monitor whether they achieve them.
This is seen as important because public sector organisations need to know whether these tools do what they promise – deliver value for money and help guard against any unforeseen risks.
Unauthorised use
Some recent scandals show how GenAI tools can be risky when misused[11]. In response, some public service agencies have banned freely available GenAI models such as ChatGPT[12] and only allow access to officially authorised programs such as Copilot[13].
But this does not mean public servants are not using the technology.
Several interviewees told us they were aware of colleagues using unauthorised programs to enhance their productivity. Personal devices are often engaged to bypass system restrictions. Concerns were expressed public servants might not be receiving guidance on how to use these tools carefully and safely.
New reality
GenAI technology is being asked to perform tasks that require human intelligence and to do these tasks more quickly. However, our findings point to a strong need to align these tools with Australian government values that frame expectations for responsible use of GenAI.
The public service faces a dilemma. Is this an opportunity for innovation in government policy work by tapping into the potential transformative impact of GenAI programs, as promised in other sectors? Or, is a more cautious approach needed to generate trust, both in the technology, and in public sector organisations to use them appropriately?
Elon Musk’s recent work in the Trump administration may suggest the latter. The experience highlighted the significant consequences of tech industry influence[15] and the use of AI tools[16] under the remit of maximising government efficiency.
The Australian public has high expectations of government to solve problems such as the housing crisis[17] and cost-of-living[18] pressures. A combination of machine and human intelligence may offer the power needed to tackle these complex economic and social issues. However, not all agencies have yet decided to flip the switch.
Our research highlights the mix of views among senior public servants towards GenAI. Whether it transforms the public service or simply speeds up business as usual will depend not on the technology itself but on how boldly, carefully, and transparently governments choose to use it.
References
- ^ productivity (mitsloan.mit.edu)
- ^ efficiencies (www.mckinsey.com)
- ^ Microsoft survey (www.microsoft.com)
- ^ risks (hbr.org)
- ^ Our new research (www.unsw.edu.au)
- ^ hallucinate (www.ibm.com)
- ^ biases (www.ibm.com)
- ^ Robodebt (robodebt.royalcommission.gov.au)
- ^ scandal (theconversation.com)
- ^ Lukas Coch/AAP (photos.aap.com.au)
- ^ misused (www.themandarin.com.au)
- ^ ChatGPT (chatgpt.com)
- ^ Copilot (copilot.microsoft.com)
- ^ Mick Tsikas/AAP (photos.aap.com.au)
- ^ tech industry influence (theconversation.com)
- ^ AI tools (www.techpolicy.press)
- ^ housing crisis (www.salvationarmy.org.au)
- ^ cost-of-living (futurework.org.au)
Authors: Helen Dickinson, Professor, Public Service Research, UNSW Sydney
