The New South Wales government has bolstered its artificial intelligence guardrails, adding new guidance for public servants using the generative tools like ChatGPT while redrafting its nation leading framework for building it into services.
The new guidelines released this month includes generative AI basics like definitions, guidance on responsible use, and specifics like chatbot prompts and security measures.
The separate AI Assurance Framework — mandatory for any government project using the technology — will get a version 2.0 in the new year to take account of generative AI and the reality that the technology is now “everywhere”, according to the state’s outgoing data chief.
The framework will be adopted as the national benchmark, with the federal Digital Transformation Agency working with the state and its counterparts on the new shared approach to ensure best practice and interoperability.
Dr Ian Oppermann, the New South Wales chief data scientist who is leaving the role to join Victor Dominello’s consultancy ServiceGen, told InnovationAus.com the technology is offering genuinely “enormous” potential.
But also introducing risks as it permeates into government projects and everyday workflows.
“If you’re using SAP or using Salesforce, [you can] flick a switch and suddenly it’s AI enabled,” he said.
“So, people need to be aware of the fact that the tools that they use in existing processes are changing. And if they’re going to change then we need to make sure that people are conscious of the things that are different about it.”
The New South Wales government released an AI Assurance Framework in early 2022 to act as a “soft stick” for the public servants and partners designing and building data driven services.
The AI Assurance Framework helps government teams in the design, build and procurement of AI technologies, while the state’s AI Review Committee gives specific feedback and clearance to proceed.
The framework must be applied to any project that uses AI while the committee needs to also sign off on any that are deemed high priority (large investment) or high risk.
The rapid adoption of generative AI that began later in 2022 meant the framework quickly needed a refresh.
Dr Opperman said version 2.0 will account for the scope of application of AI and to make sure users know when it applies to them. It will also clarify risks are not restricted to operational AI technologies and that the framework must be applied iteratively. Finally, the new version will also be linked up with procurement strategies.
“All in all, it’s largely scope of application, when to think about it, and increasingly backing away from AI as this ‘special thing’. AI is pretty much everywhere and you need to apply [the framework] throughout the lifecycle.”
Minister for Customer Service and Digital Government Jihad Dib is backing the framework that began under the former government and said it is something the state should be proud of.
“The framework is about how do we embrace AI? How do we use it effectively? How do we use it safely? How do we use it properly?” he said at the NSW government’s Future of AI event in Sydney on Monday.
The AI Assurance Framework 2.0 is in the final drafting stages, with Cabinet endorsement next year understood to be the only remaining hurdle.
As the framework is finalised the addition of the new AI Guidelines released this month stretches the advice more widely.
“The guidelines give human level guidance for understanding how to apply AI,” Dr Oppermann said. “The Assurance Frameworks says, assuming you’re going to do it, ‘this is the process you need to run through’.”
The guidelines and framework join the state’s mandatory AI Ethics Policy and New South Wales Government AI Strategy as the strongest guardrails in Australia for government workers using the technology.
Other states have developed their own guidance including the Queensland’s Use of Generative AI for Government and Victoria’s privacy obligations guide for AI. Federal advice on generative AI was also issued in July.
But the NSW AI Assurance Framework is unmatched in terms of offering a practical guide and is being used to tighten up projects like automation at the NSW online planning portal that is using AI to do genuinely new things, not just reduce friction.
“They are specifically looking to see how they can put intelligent automation as well as process automation and artificial intelligence into improving the planning portal,” Dr Oppermann said.
“That’s a great example of many of the different processes within government that could get an efficiency uplift from applications of AI.”
Better planning can mean faster approvals, more housing stock and ultimately better outcomes for citizens, Dr Oppermann said. But it also opens potential for thing like digital twins – a digitised copy of almost anything built in New South Wales.
“Once you get a digital twin, all of a sudden you’re off to the races in terms of the ability to start to do much more sophisticated modelling, understanding what if scenarios and start to explore what happens in certain circumstances,” he said.
“You can start to really do new things in new ways. It’s a different way of understanding the world if you’ve got digital twins.”
