The Australian public sector stands at the precipice of a technological revolution, with Generative AI poised to transform efficiency, customer experience, and the very nature of public service delivery.

Yet as government organisations seek to capitalise on this transformative technology, they must also navigate unique barriers and unintended consequences.

Fortunately, lessons from the private sector can provide a roadmap for public sector leaders looking to responsibly implement GenAI. Ahead of the 2nd annual Generative AI for Government Summit 2024, we spoke with Ben Burge, Department of Unresolved and Meaningful Problems at Telstra, Simon Kriss, Futurist and Chief AI Officer at simonkriss.ai, and Michael Kollo, Chief Executive Officer at Evolved AI, to discover more.

Harnessing the benefits of Gen AI

One of the most promising applications of GenAI in the public sector is its ability to simplify the navigation of complex regulatory frameworks. "Take the example of building planning legislation at both a State and Local level,” Kriss says. “There are multiple Acts and overlays to be considered, making the planning and permitting process confusing for citizens and planners alike."

By leveraging GenAI's natural language processing capabilities, government organisations can empower citizens and employees to better comprehend and comply with these intricate regulations. "The word 'dwelling' may appear in different parts of legislation and mean a house in one instance, a granny flat or a farm outbuilding in another," says Kriss. "Generative AI is incredibly good at understanding complex issues and complex text. When linked with sophisticated reasoning engines, AI will simplify the entire process."

GenAI's linguistic prowess also presents an opportunity to elevate public communications and information campaigns. "I think a lot of the problems of the public sector revolve around helping support and assist large groups in society that have very specific needs," notes Kollo. "They're highly contextualised needs; the kind of support and information that's provided is well defined by the laws, regulations, policies, and so on."

Generative models have the potential to enable government agencies to deliver highly personalised and culturally-aware messaging. "They can speak over 100 languages. They can write and read at a linguistic IQ of 140. So as a kind of mass communication device, they're probably better than email," Kollo explains.

Beyond regulatory efficiency and communications, GenAI also holds immense promise for transforming customer service in the public sector. As Kriss observes, "Clearly, like the private sector, key customer service use cases abound – with chatbots often sucking the air out of any room."

Conversational AI-powered chatbots and virtual assistants can provide more efficient and personalised support to citizens, addressing their queries and concerns with greater nuance and contextual awareness.

Unique challenges and barriers to overcome

The public sector faces a unique set of challenges that could get in the way of successful implementation of GenAI.

According to Kollo, the sector has grown used to adopting incremental tools that rely upon the current skill set plus a little bit more. "Anything that comes along that challenges that model – such as AI – represents a big leap forward for capability and technology," he observes. "But it's foreign to us from an adoption perspective."

GenAI, often described as a co-working tool, assists with ideation, summarisation, and task completion in ways that create very different kinds of use cases than we've had before. This shift necessitates the development of a new set of behaviours, potentially a different set of skills, and more critical thinking and management skills that government agencies are unaccustomed to.

Consequently, Kollo notes that "adoption ends up taking a lot longer than we expect because the way that you handle that tool is just something that you're not used to."

Adding to the challenge, Burge cautions against conflating the "impressive” capabilities of GenAI with its "useful" application. "The set of problems that are conducive to AI is smaller than the set of problems being chased by AI solutions," he warns. "AI is just a cooler (faster) way of wasting money unless it is pointed at the right problems."

Burge emphasises the need for government agencies to develop a deep understanding of their data landscape, as "the treasure trove of data on which the public sector sits is a blessing and a curse; it makes AI even more seductive." He encourages proactive efforts to accelerate "the deeply unglamourous data work" necessary to make data more ML-ready and useful for practical AI application.

While many regulatory and data challenges are not unique to the public sector, Kriss identifies one area of particular concern: data sovereignty.

"What may be a barrier for government departments is making 100% sure that data processing only happens in Australia, and this is currently very difficult," he explains. “Even the large storage and computing providers use offshore backup sites and such. This challenge is not unsolvable, but it does take detailed due diligence on providers and, in turn, their providers.”

Learning from the private sector

One key area where the government can learn from the private sector is in creating a greater sense of direct urgency around the adoption of new technologies like AI. "It's about setting up the KPIs of divisions and departments in such a way that people have to stretch and reach for them to achieve their goals. To do so, they need to innovate and think outside the square. That behaviour drives them to reach for new technology," Kollo explains.

He points to examples of smaller, more agile nations like Estonia and Singapore as inspiration, where "there are real KPIs that challenge people to reach and take risks." By fostering this type of performance-driven culture, government agencies can spur more rapid and innovative adoption of AI and other emerging technologies.

However, Kriss cautions that the government is not necessarily lagging behind the private sector in AI adoption. "I don't believe that government is at all behind the private sector today. In fact, in the USA some of the largest adoption is happening in government with recent announcements by the IRS and the State Dept as great examples."

Where the private sector can offer valuable insights, Kriss says, is in the concept of "organisational ambidexterity" when it comes to AI adoption. "On one hand is AI Exploitation (using AI to be better, faster, stronger, cheaper). On the other hand is AI Exploration (using AI to derive new services or completely new ways to deliver services)."

Exploitation, he explains, is the easier path, as it involves simply implementing proven AI solutions. But for government agencies to truly harness the transformative potential of AI, they need to also develop the capacity for AI Exploration – the vision and blue-sky thinking required to uncover new service delivery models and ways of working.

"Government needs to develop ambidexterity to look at Exploration (which requires vision, planning and blue-sky thought) at the same time as they are Exploiting this new technology," Kriss advises.

By embracing both the exploitative and exploratory dimensions of AI, government organisations can strike a balance between improving operational efficiencies and unlocking novel, citizen-centric service delivery models – ultimately delivering greater value to the communities they serve.
 

Attracting AI talent to the public sector

The demand for AI expertise has far outstripped the available supply, creating a talent crunch that threatens to slow the public sector's progress.

"We are sovereignly low on AI talent in Australia," says Kriss. "But so is Canada, America, Europe and others. We see this in roles such as Prompt Engineers that are commanding US$350,000 and more. However, this is something that AI is also starting to solve. More and more solutions are beginning to self-prompt and self-design. We are going beyond no-code to almost no-design."

The scarcity of AI talent is not just limited to highly technical roles, such as prompt engineers. There is also a pressing need for "Business AI folks” (aka non-tech), as Kriss notes, who can bridge the gap between AI's capabilities and the specific needs of government departments.

"We need thousands of people who understand business and understand AI's capabilities to act as both a 'translator' and as 'cat herder', trying to control the requests to deploy AI in every corner of the Government department," he explains.

To address this challenge, the public sector may need to rethink its traditional approaches to talent acquisition and retention. As Kollo suggests, "If you're a government department manager, you can hire from a consultancy basis or a contractual basis or project basis. You don't need to hire AI talent as employees to get the benefit."

This flexibility in hiring models can help the public sector tap into the broader talent pool, including those who may not be willing to commit to full-time employment. By leveraging a mix of permanent staff, contractors, and consultants, government agencies can assemble the diverse range of skills and expertise required to successfully implement AI-driven initiatives.

Moreover, as Burge points out, the public sector's advantage in attracting AI talent may lie in its ability to offer a different kind of value proposition. "The public sector's advantage lies in attracting the great minds who are naturally drawn to the opportunity to be deliberate about how this upending of the economy will re-shape society," he says.

Risks and unintended consequences

A growing chorus of experts is warning of the significant risks and unintended consequences that can arise from hasty and ill-conceived deployments.

One of the key issues highlighted is the "Fear of Moving Forward" (FOMF) that often paralyses public sector organisations. As Kriss explains, many government departments and local councils are "crippled with fear, predominantly based on low organisational knowledge of AI." The solution, he argues, is to prioritise education and ensure the entire organisation is equipped to understand and manage this transformative technology.

Another common pitfall is the desire to outsource AI implementation to external providers before building internal expertise. Kriss advocates for an "inside-out approach" where the first AI use cases are focused internally, allowing organisations to learn from their mistakes in a more controlled environment before scaling outwards. "Mistakes will happen," he cautions, "because this is an evolving and transformative technology and it is most often deployed in an experimental way."

Rigorous testing is also crucial, Kriss emphasises. He recommends subjecting AI systems to intense scrutiny, with 10 staff members deliberately trolling the system and trying to get it to do things it should never do. This kind of adversarial testing can help uncover vulnerabilities and biases before deployment.

Burge highlights another critical distinction that public sector organisations must understand - the difference between classification and decision-making. "Machine Learning excels at classification but models very rarely output 'actions' or 'decisions'," he explains. Mistaking these noisy estimates for definitive decisions can lead to costly mistakes, underscoring the need to integrate AI outputs with human expertise and economic considerations.

Perhaps most concerning are the potential risks of AI-enabled profiling. Kollo stresses the importance of public transparency, arguing that if an algorithm is going to "insinuate or identify anything that's unknown or uncertain, it really has to be able to tell people why it's done so, and how it has done so."