Australia Considers AI Data Mining & Copyright Reforms

The Australian Productivity Commission is currently examining whether technology firms should be granted exemptions from existing copyright laws to facilitate the training of artificial intelligence models using text and data. This consideration is a key component of the Commission’s interim report, “Harnessing Data and the Digital Economy,” which uses copyright as a central case study for adapting Australia’s regulatory framework to manage the risks and opportunities presented by AI.

A significant recommendation from the interim report is for the federal government to undertake a comprehensive review of regulations. This review aims to identify and address potential gaps that could be exploited by malicious actors leveraging AI.

The debate around copyright exemptions has drawn strong opinions. Scott Farquhar, co-founder of Australian software giant Atlassian, has publicly advocated for an “urgent” overhaul of Australia’s copyright rules, arguing they are outdated compared to those in other comparable nations. Farquhar contends that creating exemptions for text and data mining to train large language models could attract billions of dollars in foreign investment into Australia.

Conversely, the Copyright Agency, a not-for-profit organization responsible for collecting and distributing royalties to copyright holders, has rejected the idea of such exemptions. Instead, the agency proposes that the government establish a new compensation scheme to remunerate creators whose content is used by tech companies to train their AI models.

Stephen King, one of the two commissioners leading the Productivity Commission’s inquiry, highlighted the complex balance at play. “Copyright is a great example of where Australia needs to sit back and ask: ‘Are our laws fit for purpose with AI?’” King stated. He acknowledged the potential harm of AI companies using copyrighted materials without fair compensation, while also recognizing the desire to foster the development of AI tools that utilize such material. King suggested exploring approaches similar to existing copyright collection societies, like those in the music industry, where collective schemes manage compensation for widespread use. The Commission is seeking feedback on various options, including a potential “fair dealing” exemption for text and data mining, provided AI companies obtain and pay for legal copies of the material. A final recommendation is expected by the end of the year.

The Commission’s report also projects significant economic benefits from AI. Its modelling suggests that even conservative estimates indicate AI could deliver a $116 billion boost to the Australian economy over the next decade. This translates to an estimated $4,300 increase in the average Australian’s real wage within ten years, with potential for even greater benefits.

Despite the transformative potential, the Productivity Commission advises against creating a broad, overarching piece of AI-specific legislation. It warns that overly cumbersome or excessive regulation risks stifling the technology’s benefits. This stance aligns with the views of Assistant Minister for Productivity, Andrew Leigh, who supports this cautious approach.

However, King clarified that the Commission is not advocating for a minimalist regulatory framework. “It’s not light-touch at all,” he explained. “What we are saying is that AI is going to make it easier, cheaper and faster for bad actors to engage in harmful conduct. But most of that harmful conduct is already illegal.” The Commission’s approach emphasizes identifying where harms may occur and ensuring that existing laws cover them, with regulators adequately resourced and empowered to act.

Beyond copyright, the interim report also supports evolving privacy rules towards an outcomes-based approach, moving beyond mere “box-ticking” compliance. Furthermore, the Commission encourages the government to support new pathways that enable individuals and businesses to access and share data pertaining to them.