NDIA Accountability Project

This project is in DRAFT

USE OF AI/ML AND ADVANCED TECHNOLOGIES WITHIN NDIA

‹ PrevPage 1 of 6 · Source p. 119Next ›

DOCUMENT 5 FOI 24/25-1356 - DISCLOSURE LOG

ADDITIONAL ESTIMATES 2023–24

TITLE: USE OF AI/ML AND ADVANCED TECHNOLOGIES WITHIN NDIA

WITNESS: Sam Porter, Chief Operating Officer, DCEO Enabling Services

Strategic Narrative

  • The NDIA is supportive of the safe, responsible, and ethical use of Artificial Intelligence (AI) in line with the advice provided by the AI in Government Taskforce.
  • AI, Machine Learning, and generative AI will provide opportunities for the NDIA in the future to improve the efficiency and effectiveness of delivery to NDIS participants.
  • Machine Learning is a subset of AI that involves the use of algorithms to learn from data and make predictions or decisions without being explicitly programmed.
  • Generative AI is a type of AI that involves the use of machine learning models to generate new data that is like a training set. It differs from other forms of machine learning in that it doesn’t just make predictions or decisions based on data but can also create new data.
  • The Agency is still developing its policy position on the use of AI, especially Generative AI, which is changing quickly in the market. We are listening to various views from across the Australian Public Service to inform the NDIA approach.
  • On 20 September 2023, the Hon Ed Husic MP, Minister for Industry and Science and Senator Katy Gallagher, Minister for Finance announced the AI in Government Taskforce.
  • In January 2024, the Australian Government has released its interim response to the Safe and Responsible AI in Australia consultation, which seeks to ensure that AI is safe and responsible.
  • The response focuses on the use of AI in high-risk settings, where harms could be hard to undo, while allowing low-risk AI use to keep growing.

Page 158 of 189

FOI 24/25-1356 - DISCLOSURE LOG

  • The government is thinking about mandatory rules for AI development and deployment in high-risk settings and is taking immediate actions such as working with industry to develop a voluntary AI Safety Standard and options for voluntary labelling and watermarking of AI-generated materials. This work will be vital to the safe and responsible use of AI within the NDIS.
  • The NDIS will develop approaches to the governance, risk management and adoption of AI at an agency level, based on the outcomes of the AI in Government Taskforce and future update to the interim guidance.
  • As part of the upcoming Microsoft Copilot trial coordinated by the Digital Tranformation Agency (DTA), we will work through uplifting Agency capability to provide a better understanding of how AI may impact Agency security and support the disability sector.
  • Since AI can easily take inputs, NDIA staff will be educated on the risks of disclosing classified information, or personal information. The Privacy Act 1988 (Privacy Act) and Protective Security Policy Framework (PSPF) will be referenced in all training.
  • Aligned with the DTA guidance, the Chief Information Officer will be clear in warning that participant data, plans, or supports should never be included in any interaction with an AI service, whether it is approved or not.
  • In May 2023, the Agency stopped access to Open AI products (see Background) in the Agency Information and Communications Technology Environment.
  • This decision was driven by compliance obligations with the Privacy Act along with the risk of participant data being stored overseas and the re-use of that data by third parties.
  • AI tools offer new and innovative ways for government agencies to improve effectiveness and efficiency of services, but AI also gives threat actors new and improved abilities for targeting ICT systems, which has already been seen with better and more targeted phishing campaigns.

Page 159 of 189

FOI 24/25-1356 - DISCLOSURE LOG

e Al tools will also provide opportunities for better cyber analysis and defence. However, cyber criminals and state-based actors operate at a higher adoption rate than those defending ICT systems.

KEY HEADLINE FACTS AND FIGURES

  • On 17 January 2024, Interim advice from the Al in Government Taskforce was released to guide the safe and responsible use of Al. The advice indicated that mandatory guiderails for Al development and deployment of generative Al in high-risk settings may occur through specific changes to legislation to protect privacy and sharing of confidential information.
  • The DTA updated (22 November 2023) its advice to Government Departments and corporate entities on the use of generative Al to support the management of risk.
  • NDIA welcomes the Al in Government Taskforce created on 20 September 2023 and will consider outcomes from the Taskforce and any guidance on public use of generative Al platforms.
  • NDIA suspended use of Open Al products in May 2023 to prevent input of personal information into Al tools and ensure alignment with Privacy Act.
  • Before May 2023, all NDIA staff had access to Open Al tools including ChatGPT.
  • Prior to access being suspended, only 376 staff had accessed ChatGPT in the Agency.
  • The Agency had been advised that there had been the potential for NDIA staff to use ChatGPT for the development of participant plans, but this has not been able to be confirmed.
  • In the last 90 days, 10 NDIA users have attempted to access blocked Al services.
  • The Agency continues to monitor the rapidly evolving Al and Generative Al technology landscape and block access to tools in the interest of protecting Agency and participant data.

Page 160 of 189

FOI 24/25-1356 - DISCLOSURE LOG

Background

  • OpenAI products are a collection of AI tools and models created by the research organisation OpenAI, with ChatGPT and similar language-based models being the most popular products.
  • OpenAI and similar products can have an impact on the environment by generating an output (results, recommendations, or options) based on a given set of objectives. It uses a combination of machine AI, generative AI, and/or data and inputs from humans to perceive real and/or virtual environments; simplify these perceptions into models through analysis in an automated way. The output of these models can be documents or images.
  • Access to Open AI products was largely restricted in May 2023 and the NDIA has continued to track the launch of new products into the market.

Improved AI monitoring in the future

  • The NDIA Cyber Security operations team continues to provide regular scanning of the broader technology and AI landscape.
  • In September 2023, NDIA Cyber Operations broadly blocked all generative AI services.
  • This block excludes services like Read Speaker, Adobe Sensei, and Microsoft’s Bing native AI chat function which can be used to support Agency assistive technology requirements.
  • Access to Generative AI services were blocked due to the following risks:
    • Quality of data outputs
    • NDIS data being part of the AI training set which might include Agency protected data
    • NDIA would not have control over NDIS data once it was used in external AI services
    • The use of generative AI products has led to a big rise in malware created by criminal and state-based actors who want to attack systems and services.

Page 161 of 189

FOI 24/25-1356 - DISCLOSURE LOG

AI Opportunities for the NDIA

  • By using Machine Learning AI, Generative AI, and process automation, the NDIA can enhance its performance and productivity in administrative tasks. This can lead to better results for Participants and lower costs to improve scheme sustainability.
  • The NDIA will work with the DTA, the AI in Government Taskforce, and other government entities to establish a cohesive strategy and effective governance model for the management and use of AI in NDIA to ensure:
    • the accuracy of data and outcomes developed through AI
    • the security of the systems and data engaging with AI
    • the development of clear governance structures for the use of AI
    • effective educative supports allow NDIA users and ICT staff to work within whole of government and legislative guiderails when using AI to keep critical data secure.
  • Examples of sanctioned use of generative AI within the Agency ICT environment include:
    • Adobe Sensei: This product allows NDIA business areas to utilise generative AI to assist in editing images quickly within Adobe products, saving critical time in the creation of marketing or educative materials.
    • ReadSpeaker.com: This web based generative AI application is used to allow text to speech services within the NDIA external websites. The AI functionality allows text to mimic natural language and context driven speech patterns and is an advance on robotic text translators.

Page 162 of 189

FOI 24/25-1356 - DISCLOSURE LOG

CoPilot for Microsoft 365

  • In November 2023, the NDIA was invited to be one of 30 agencies to take part in a managed trial of CoPilot for Microsoft 365 (M365) coordinated by the DTA.
  • The trial of Copilot for M365 will provide a safe and integrated AI environment that can be fully controlled by the Agency. The trial will begin in February 2024 with 300 initial licences and will include members of the NDIA Employee Disability Network.

Scheme Actuary algorithms

  • The Agency’s Scheme Actuary and Data Analytics area do not use generative AI within any of the algorithms that currently support the scheme.
  • While machine learning is utilised within draft budgets (or Typical Support Package) for first plans based on key information from a participant’s profiles. The algorithm is only ever used to make recommendations, with decisions made by actual delegates.
  • The machine learning recommendations are used to assist delegates by speeding up the initial analysis to provide quicker resolutions for participants and improved service.

Policy Development Update

  • The NDIA will work with the DTA to develop an interim policy for the governance and use of AI, based on the outcomes of the Copilot for M365 trial and the AI in Government Taskforce.

Page 163 of 189