AI Policy: avoiding social calculus on productivity vs. wage inequality

Government should focus on free and fair markets, innovation which augments rather than replaces human intelligence, and leverage sovereign data trusts for public interest AI training.

‘Innovation’ per se is not a relevant policy goal for the Australian Government; the goal should be beneficial innovation for all Australians. If AI and ADM are desirable for a free, prosperous, and sustainable future, regulation must maximise such benefit whilst mitigating foreseeable harms. We cannot just impose ethical and moral standards reactively when things go wrong; regulation should be built around a coherent vision for AI and ADM in Australian society.

Regulatory tools might include:
      1. Sovereign Data Trust for public interest AI training
      2. An register for AI and ADM in use
      3. Appointment of an AI Safety Commissioner
      4. Utilisation of ‘RegTech’ tools to monitor risk critical systems
      5. Sandboxes and Innovation Labs
      6. AI standards to enable quality and consistency
      7. Board Directors  guidance for managing novel risks of AI
      8. New technical tooling for impact measurement of AI and ADM
      9. A robust liability regime for the outputs of AI and ADM

Regulation of AI and ADM should primarily aim to create the framework for free and fair markets, healthy competition and the avoidance of monopolies, robust avoidance of negative externalities and other textbook market failures.


Beyond textbook ‘command and control’ regulation, it is important to remember the inherent regulative function of private law liability regimes. These allow consumers and citizens to vindicate their rights and interests when impinged by the actions of another. For example, the law of torts imposes liability on categories of intentional, reckless, and negligent action that causes harm to others.


The problem currently is that many AI and ADM systems—especially data-driven systems that fall closer towards the ‘intelligent’ and ‘autonomous’
end of the spectrum—complicate notions of agency that underlie private law liability regimes and potentially confound their operation.