Doctor unveils AI governance test for human parity

Dr Andrew Rochford has published a research paper outlining a method to test whether artificial intelligence decisions match the standard of qualified human professionals, as scrutiny grows over automated decision-making in Australia.

The paper sets out what Rochford calls Above-the-Loop Governance, a framework designed to compare a sample of AI-made decisions with judgments made independently by qualified experts on the same cases. It is aimed at organisations using AI to make decisions with material consequences, including lending, hiring, insurance claims and public benefits.

The work comes amid a broader policy debate over how governments and companies should oversee automated systems. That debate has sharpened in aged care, where the Albanese Government is under pressure over an assessment algorithm used to help determine support and care entitlements. Critics say there is no clear evidence showing whether the tool reaches the same conclusions as a qualified assessor.

Rochford argues that existing oversight methods do not answer that central question. Many organisations either place a human reviewer within each decision process, rely on occasional spot checks, or operate with limited practical oversight once systems are deployed.

His method proposes a different model. Instead of checking every decision, it would draw a statistical sample from an AI system's output and send those cases to qualified human professionals, who would assess them blind, without seeing the AI result. Their judgments would form what the paper describes as a living benchmark against which the AI's performance would be measured over time.

The comparison would be expressed as a single score for each class of decision, using agreement statistics drawn from fields such as clinical research and quality control. The framework also proposes retaining sampled cases, expert judgments and any resulting actions as an audit record.

If an AI system repeatedly diverges from the human benchmark in a particular decision category, those decisions would return to human control while the cause is examined. The paper also says the model should identify situations where human experts themselves do not agree, which may indicate that the decision criteria are unclear or that the matter is not suitable for automation.

Rochford, a medical doctor and media broadcaster who also founded the technology firm Trakked, said the framework was developed to address the gap between regulatory principles and day-to-day operational proof.

"This methodology has been developed because organisations are handing consequential decisions to AI faster than they can prove those decisions are sound. We don't trust a surgeon or an accountant to make decisions just because they're clever. We trust them because of everything standing behind them: the training, the registration, the accountability when it goes wrong, and because that trust can be taken away. AI arrived and is rapidly making the same kinds of decisions, where people's lives and livelihoods are on the line, with none of these systems behind it," Rochford said.

He linked the issue directly to board accountability and legal risk when automated decisions affect individuals.

"Right now, when these decisions go wrong, no one can tell the person affected why or how it has happened. And accountability for this lands straight onto the board. As a director myself, I take that very seriously. The Chief Justice of New South Wales has said it is extremely doubtful a director who leans on an AI's recommendation without applying their own reasoning can rely on the protections they depend on. You cannot delegate the thinking and keep the shield," Rochford said.

Policy context

The proposal comes as Canberra and regulators tighten their focus on AI governance. The federal government has established an Office of AI and committed to a national framework, while the Australian Prudential Regulation Authority and the Office of the Australian Information Commissioner have both pressed industry on transparency and oversight in automated decision-making.

Changes to privacy law due to take effect later this year will also require organisations to disclose in their privacy policies where automated decision-making significantly affects individuals. That adds pressure on businesses to show not only that AI is in use, but also how it is supervised and when it can be challenged.

Rochford said regulation alone cannot tell an organisation whether its own system is meeting the required standard at a given moment.

"AI regulations can set the standard, but they can't tell an organisation whether it's meeting that standard today, in its own systems. Above-the-Loop Governance closes that distance. I trained in medicine, where we never test a treatment on every patient; we test a properly drawn sample and monitor it for as long as it's prescribed. That's the same statistically sound method behind clinical trials, financial audits and quality control, and it has been keeping people safe for seventy years," Rochford said.

The paper was reviewed by academics at the University of Sydney's Centre for AI, Trust and Governance, with an independent review of the statistical method by a lecturer in business analytics at the University of Sydney Business School.

Professor Terry Flew, co-director at the University of Sydney's Centre for AI, Trust and Governance, said: "Trust is rapidly becoming the defining challenge of AI adoption. Organisations have traditionally measured success through efficiency, productivity and financial performance, but the widespread deployment of AI means they must also measure its impact on institutional trust. Public confidence is earned when organisations can demonstrate that AI-enabled decisions are accountable, transparent and consistently aligned with professional standards. In that sense, trust should be treated as a strategic asset and governed accordingly."