A very enjoyable day at the AFR Government Services Summit yesterday.

Minister Leigh’s keynote set the tone: 𝗔𝗜-𝗱𝗿𝗶𝘃𝗲𝗻 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆 𝗴𝗮𝗶𝗻𝘀 𝗵𝗶𝗻𝗴𝗲 𝗼𝗻 𝗲𝘅𝗽𝗲𝗿𝗶𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻, 𝗲𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻, 𝗱𝗶𝘀𝗰𝗶𝗽𝗹𝗶𝗻𝗲𝗱 𝗰𝘂𝗿𝗶𝗼𝘀𝗶𝘁𝘆, 𝗮𝗻𝗱 𝗱𝗲𝘀𝗶𝗴𝗻𝗶𝗻𝗴 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻 𝗯𝘆 𝗱𝗲𝗳𝗮𝘂𝗹𝘁 𝘀𝗼 𝘄𝗲 𝗰𝗮𝗻 𝗴𝗼𝘃𝗲𝗿𝗻 𝘄𝗶𝘁𝗵 𝗰𝗼𝗻𝗳𝗶𝗱𝗲𝗻𝗰𝗲. This resonates deeply with me as a scientist viewing productivity growth and risk management through the lens of evaluation science.

Highlights included David Hazlehurst (CEO, Services Australia) showing how responsible AI is not about “can you” but “should you”—being clever, not creepy—and reminding us that humans are biased too. The challenge is to use AI to remove bias without introducing new kinds.

On my panel on responsible AI and governance, I emphasised—harking back to Minister Leigh’s point:
1. The danger of premature risk labelling without proper evaluation tools
2. Why safeguards must be measured for effectiveness and cost, not just symbolic presence
3. How evaluating AI without perfect ground truth is both possible and necessary—the new frontier CSIRO’s Data61 are working on to unlock AI’s value

Looking forward, explainable and agentic AI raises new challenges. Demanding traditional explainability is like asking for a neuron-level explanation of the human brain. Over-supervising agents with human oversight defeats the definition of “agentic”. The task is to relinquish agency and control judiciously while maintaining trust.

𝗪𝗲 𝗮𝗿𝗲 𝘄𝗼𝗿𝗸𝗶𝗻𝗴 𝗼𝗻 𝗰𝗼𝗻𝗰𝗿𝗲𝘁𝗲 𝘄𝗮𝘆𝘀 𝘁𝗼 𝗲𝗻𝗮𝗯𝗹𝗲 𝗺𝗲𝗮𝗻𝗶𝗻𝗴𝗳𝘂𝗹 𝗵𝘂𝗺𝗮𝗻 𝗼𝘃𝗲𝗿𝘀𝗶𝗴𝗵𝘁, 𝘄𝗵𝗶𝗰𝗵 𝗶𝗻 𝗮 𝗳𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹 𝘄𝗮𝘆 𝗱𝗲𝗳𝗶𝗻𝗲𝘀 𝘄𝗵𝗮𝘁 𝗳𝘂𝘁𝘂𝗿𝗲 𝗷𝗼𝗯𝘀 𝘄𝗶𝗹𝗹 𝗹𝗼𝗼𝗸 𝗹𝗶𝗸𝗲 𝗮𝗻𝗱 𝘄𝗵𝗲𝗿𝗲 𝘁𝗵𝗲 𝘀𝗲𝗰𝗿𝗲𝘁𝘀 𝗼𝗳 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆 𝗹𝗶𝗲.

Finally, I reflected on what’s coming—especially the question I’m sometimes asked: 𝗪𝗵𝗮𝘁’𝘀 𝘁𝗵𝗲 𝗹𝗮𝘀𝘁 𝗷𝗼𝗯 𝗳𝗼𝗿 𝗵𝘂𝗺𝗮𝗻𝗶𝘁𝘆?

It is 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴—of AI and of other humans. AI may surpass us in many tasks, but the joy of learning, improving, and injecting human ingenuity, values, and safeguards remains. Just as no human can beat AI at chess, yet the number of chess players, grandmasters, and their skill levels are at their peak today, human growth itself is the enduring pursuit.