Supervisor: Moira Paterson (Monash UNiversity), Co-supervisor: Pompeu Casanovas Romeu (IIIA-CSIC)

As the adoption of generative AI gains pace governments across the world, including in Australia, have been preoccupied with questions about how to design a regulatory architecture that maximises its economic and societal benefits while mitigating its harms. A common thread of the debate has been to ensure that any regulatory measures that are developed do not stifle innovation.

Initially the Australian government’s approach was to adopt overarching regulation, informed by European AI regulation, based around a series of ‘guardrails’ that would apply to high-risk AI systems. More recently, it appears to be retreating from this position following work undertaken by the Productivity Commission, motivated by prioritising the productivity gains that could be hampered by overarching ‘guardrails’ legislation, and recommending against this approach as a last resort that should only be implemented where other, existing legal and regulatory frameworks are shown to be unsatisfactory. This thesis broadly supports the Productivity Commission’s approach but for different reasons. 

Drawing on rule of law concepts that suggest that a regulatory architecture should enable existing laws to apply to the artificial, this thesis argues that the starting point of Australia’s approach to AI regulation should be constituted by a principle of equivalence where the artificial is regulated in an equivalent manner to human activity. Where there are disjunctions between the two, such as the lack of transparency, accountability and contestability of AI systems, additional regulation should be developed.