Localisation of the OECD AI Classification Framework (AICF) for the Australian Health Sector – Phase 2

Opportunity

Achieving Responsible AI (RAI) goes beyond addressing algorithmic and data-related issues, it also involves tackling system-level challenges that span multiple layers of governance and the entire AI engineering lifecycle. In response, the Australian Government has introduced a range of ethical principles, frameworks, and guardrails designed to promote, though rarely mandate, good practice across the AI sector. Furthermore, putting AI ethics into healthcare practice requires close collaboration between AI specialists and domain experts from fields such as social science and clinic practice.

As a continuation of the DHCRC Project – DHCRC localisation of OECD AI Classification Framework to support governance of investment and commercialisation of AI systems amongst Australian SMEs, this second phase will focus on enhancing the prototype OECD AI Classification Framework web-based tool, refining the question set, and developing a quantifiable matrix to support the creation of a logic model. Our research aims to develop a risk matrix, AI-based decision support or classification logic to extend the potential for the OECD AICF to deliver more granular classification of AI solutions in healthcare.

This project aims to develop an internationally recognised framework for classifying and validating AI tools in
healthcare, supporting consistent approaches to procurement, governance, and evidence-based adoption. The first
step will be to refresh and analyse current regulatory and practical needs, then establish a comprehensive framework
that captures these requirements. The next phase will make the framework’s classification logic operational and
implementable, followed by validation to ensure reliability and applicability. The final stage will focus on the development of a prototype incorporating the Data Input and AI Models to be considered in the OECD AI Classification Framework.

Project Objectives

• Refine the OECD-based questions by incorporating insights from previous research on the responsible use of AI in decision-making, in particular the design of automated guardrails, benchmarking methods and compliance mapping approaches. This process will ensure that the questions are both contextually relevant and aligned with best practices in AI ethics, transparency, and trustworthiness.
• Develop a mathematical framework and corresponding risk assessment method informed by the AICF data input variables and AI model dimensions. This mathematical framework will form the foundation for a logic based classification model capable of interpreting and categorising diverse response patterns to the identified questions. The resulting model will facilitate consistent and evidence-based assessment of AI risk and trustworthiness across multiple application contexts.
• Test and validate the proposed logic-based classification model using real-world case studies. This evaluation will assess the matrix’s effectiveness, reliability, and robustness in capturing and classifying AI-related risks within practical application settings.