Initiative overview
This initiative used artificial intelligence to support the analysis of public consultation responses for a government policy process on regulating AI in high‑risk settings. It was developed to address the challenge of reviewing a large volume of written submissions in a timely and consistent way while maintaining human oversight, transparency, and public trust. The initiative sought to demonstrate that AI could be used safely and responsibly to assist policy development, rather than replace human judgement. The primary objective was to improve the efficiency and consistency of consultation analysis while ensuring that human oversight of submissions was maintained. AI tools were used to help summarise themes and extract insights from de‑identified consultation data with all outputs checked against original submissions. The initiative also aimed to support the Australian Government’s broader policy direction on responsible AI use by acting as an early, practical example of how emerging technologies could be applied in a high‑risk public sector context. Lessons learned were documented to inform future consultations, improve consultation design, and strengthen governance arrangements for AI use across government. Looking ahead, the approach outlined in the methodology and reflections are being refined and reused in future consultations and initiatives. Potential evolution includes better alignment between consultation design and analytical needs, clearer guidance for staff, and reduced reliance on specialist technical expertise. Over time, the initiative’s principles and practices could be institutionalised as part of standard consultation practices and analysis processes, supporting broader and more consistent use across policy areas where large‑scale public input is required.
Results, outcomes and impacts
The initiative showed that AI can assist consultation analysis while preserving human control. AI-supported summaries helped staff identify themes more efficiently, with human oversight maintained effectively, and enabled consistent, verifiable analysis through documented quality assurance and source checking. Results were validated through quality assurance processes, including manual review of submissions, cross‑checking AI outputs against source material, comparison with non‑AI analysis, and supervision by experienced staff. Documented lessons are expected to improve future consultation design, governance, and scalable analysis practices.
