Rod Sims: Governments must not give AI companies a free pass on copyright

Article content

At the end of June, Australian Sen. David Pocock alleged that the Albanese government was considering two competing proposals to change copyright policy for AI training.

Article content

Pocock said one proposal would create a copyright “carve-out” for artificial intelligence companies tied to billions of dollars in data centre investment and hundreds of millions of dollars annually for a creative fund. The other proposal instead would expand licensing agreements that would clear the way for AI companies to legally use Australian material to train their proprietary technology.

Article content

Article content

Article content

While the detail is not available, these proposals on the surface are deeply problematic.

Article content

Article content

But first, what other sector refuses to negotiate with suppliers and instead goes to government to bypass such a step? When the chief executive of Anthropic came to Australia recently he spent a lot of time with the government, but it seems no time with those whose content he needs.

Article content

This is a familiar pattern with tech companies. They provide content to users as if it is their own but do not feel obliged to pay for it; Anthropic, for example, recently reached a US$1.5-billion (C$2.1-billion) court settlement of a class action taken by authors.

Article content

In the simplest example, if you ask a search engine or AI what is going on in economics or politics anywhere in the world they will usually distill the content of media organizations to provide the answer. The platform or AI company has no ability to answer the query; they have no resources devoted to doing this. They supply the platform and algorithms, which are valuable but useless without the content they call on.

Article content

Traditionally, they have wanted this content for free. That they often are successful is reflected in the stratospheric valuations of these organizations, while the value of those providing the necessary content is falling to the point that their existence is threatened. Society must care about this anomaly. Quality media underpins our democracy as it provides a vital journal of record, a forum for informed debate and establishes necessary shared facts. Our writers, artists and composers are fundamental to our culture and identity.

Article content

Article content

We cannot know the accuracy of Pocock’s allegations but we do know that discussions have occurred and that the two proposals appear to align with what the AI companies are seeking. We cannot, therefore, be complacent.

Article content

Article content

The flaws in the first proposal seem clear. In essence they remove rights holders’ control over their content and pass this to AI companies. No ability to withhold their content; no ability to understand how their content is used; no ability to control how it is shown; and an uncertain payment. Without these features there is no copyright protection; copyright means payment and control. Further, there have been many suggestions concerning “creative funds” to finance future works. Such funds misalign whose content is being used and what the money in the fund would be used for. The second proposal is unclear. But it can work only if it relies on voluntary agreements between various rights holders and the AI companies.

Article content

AI companies should be treated the same as other companies. If they want a necessary input to production they should negotiate for it.

Article content

Some argue that we must avoid the high transaction costs of AI companies needing to do numerous deals, but the AI companies have enormous resources. Further, it is possible to devise licensing structures in each field to deal with this. In each area there are a few large players and collective bargaining can cover the rest, as it did with Australia’s News Media Bargaining Code and Canada’s Online News Act. Yes, the deals will be complex, but that is the point; avoiding this complexity simply highlights the flaws in a compulsory across-the-board approach.