CSIRO leads Australia's responsible AI implementation

But how these systems are designed is just as important as their output. As AI's influence grows, so do the risks, making it essential that it is developed and deployed in a way that delivers real benefits to all Australians.

This is a key focus reflected in the recent National AI Plan and AI Plan for the Australian Public Service. Both emphasize the importance of developing AI systems that are safe, secure, and reliable.

That's why CSIRO, as Australia's national science agency, is using its expertise to help ensure this transformation occurs safely and responsibly.

Building trusted AI for Australia

When AI systems lack appropriate safety and ethical safeguards, risks are pushed onto users and organizations and individuals remain limited in their ability to prevent or remediate problems. This can lead to biased decision-making, privacy violations, economic harm, and real harm to those who rely on these systems, group leader Dr. Minh Dinh explained.

“Ensuring that AI is responsible by design reduces this risk from the start. It leads to systems that people can trust, and that trust has real value in the market. This is where Australia's AI competitive advantage lies,” Dr Ding said.

A responsible approach to AI

CSIRO's responsible AI approach focuses on building safety, security, privacy and trust into AI systems from start to finish. We emphasize human oversight, allowing people to review and question AI decisions, and empowering users and communities to make informed and safe decisions.

This year, CSIRO's Responsible AI (RAI) team published Engineering AI Systems, a practical guide to navigating the complexities of designing and operating AI systems. This book is based on the RAI engineering principles introduced in Responsible AI, published last year.

These guides demonstrate how responsible AI works in practice, using the same approach that CSIRO researchers and engineers apply when building AI systems, said Dr Qinghua Lu, acting director of research.

Dr. Lu explained that the core of RAI is to create a system that is transparent, reliable, and secure.

“People need to be able to check the output of AI and question decisions that affect their lives, especially in critical areas such as healthcare, finance, and national security,” she said.

Transforming public auditing

CSIRO is partnering with the NSW Audit Office to explore how AI can support government auditing and decision-making. This collaboration combines AI tools with established audit oversight practices to provide a way for people to monitor AI systems and ensure they are used responsibly.

NSW Auditor-General Bola Oyetunji said early signs were encouraging.

“We look forward to exploring further advances in predictive analytics and sensemaking for large documents and are pleased that CSIRO’s approach to AI is underpinned by human oversight,” Oetunji said.

Supporting safe AI implementation

CSIRO has worked with the National AI Center to develop guidance for AI adoption that outlines six key practices for responsible AI governance and use. This guide is designed for organizations and professionals new to AI to help establish basic responsible AI practices.

This guidance provides clarity about AI-generated content and explains when and how to tell people that their text, images, audio, or video was created or modified using AI. Outlines best practice approaches for clear and consistent disclosure to promote transparency and support trust. We discuss approaches such as labeling, watermarking, and metadata.

Dr Lu explained that increased trust in new digital technologies will accelerate their safe adoption by Australian organizations and communities.

“Trust in AI is the new currency. CSIRO is leading the way in science to ensure Australia’s AI future is secure, inclusive and innovative, delivering meaningful benefits for all Australians.”