Australia Blocks AI Use of Copyrighted Content for Training

The Australian Government has decided against allowing tech companies to mine copyrighted content to train artificial intelligence (AI) models, according to a report by The Guardian. This comes after the Australian Government’s review and advisory body, the Productivity Commission, published an interim report in August proposing three ways Australia could approach copyright laws in the age of AI:

• No policy change: Copyright owners would continue to enforce their rights under the existing copyright framework, including through the court system.
• Policy measures to facilitate licensing: Introducing mechanisms such as collecting societies to better enable licensing of copyrighted materials.
• Amending the Copyright Act: Including a fair dealing exception that would cover text and data mining (TDM).

The proposal for a fair dealing exception faced pushback from several organisations. The Association of American Publishers wrote to the commission stating that “un-permissioned and uncompensated use of copyrighted works as datasets for training generative AI (Gen AI) systems is a threat to the professions and livelihoods of authors, publishers, and all who are integral to the publishing endeavour.” Similarly, a rapper told an Australian Parliamentary inquiry in September that attempting to fix the situation after allowing AI companies to use Australian copyrighted data without compensation would be like trying to get a genie back into a bottle.

What did the Commission say about AI and copyright protection?

While acknowledging concerns around unauthorised use of copyrighted material, the commission suggested two primary approaches to address the issue:

1. Addressing concerns before infringement through copyright licensing:
   In Australia, collecting societies handle licensing and negotiate licences for multiple copyright holders at once.
2. Tackling unauthorised use after infringement:
   This involves relying on enforcement measures such as takedown notices, court action, and alternative dispute resolution.

Besides these, the commission suggested the possibility of expanding the fair dealing exception in Australian copyright law to include text and data mining (TDM). It noted that comparable exceptions exist in jurisdictions such as the European Union, United Kingdom, and Japan.

“Such an exception would cover not just AI model training but all forms of analytical techniques that use machine-read material to identify patterns, trends, and other useful information,” the commission said, citing examples from research sectors that use TDM techniques to produce large datasets for statistical analysis.

The Productivity Commission acknowledged that, at present, large AI models are mostly trained abroad and it was uncertain whether a TDM exception would change that trend. However, it argued that such an exception could “make a difference to whether smaller, low-compute models (such as task-specific models) can be built…