Challenge: Writer Zakes Mda took tech firm Anthropic to court over its use of copyrighted works to train AI models. Photo: Centre for Creative Arts
Proponents of the Copyright Amendment Bill (CAB) and the “big tech” companies developing artificial intelligence (AI) persistently attempt to convince us about the merits of the claimed benefits that will flow from the bill and AI.
They also do not appear to have concerns about making sweeping statements to support their claims. Maybe this should not come as a surprise — the most troubling provisions in the CAB will facilitate the same tech companies’ use of copyright works in training their AI systems.
Copyright is still largely an arcane area of law, which is not understood by the public or even the majority of lawyers, who have never studied the subject or practised in the field.
Thus, it is difficult to discern whether statements made by, for example, journalists concerning copyright or the CAB are simply the consequence of ignorance (and a case of simply parroting the claims made by proponents of the CAB and the tech companies benefited by the proposals) or in support of the CAB (or big tech companies).
An article published recently in The Citizen claims that, in light of the use of copyright works in the training of AI systems, the CAB “can provide better digital rights protection” for South African authors.
If only that were true.
The article reported on the settlement of a copyright infringement claim brought by several authors, including Zakes Mda, against the tech company Anthropic concerning the use of their copyrighted works in training its AI systems. Anthropic agreed to pay $1.5 billion to settle the case with the authors.
The article also claims authors are not adequately protected under South African copyright law and would not be successful if they instituted a similar claim locally. Both these claims are simply wrong.
First, it must be recognised that the first draft of the CAB was made available for comment 10 years ago. At the time, the kind of challenges posed by generative-AI systems to copyright owners that have emerged over the past three years did not exist.
There is no provision in the CAB that seeks to address these developments in AI. This, in itself, suggests that the CAB might have been overtaken by these developments.
If the CAB was intended to modernise our copyright law, it has, ironically, already been overtaken by technological developments!
To be clear, it is not being suggested that our current copyright law does not need to be updated. Of course, it must. However, this should start with a comprehensive review.
The CAB was not the result of such a process and, worse still, the net effect of its provisions is to hurt the position of authors and copyright owners.
Foremost among the provisions in the CAB that weakens the position of copyright owners (and, thereby, authors) is the proposed introduction of an open-ended, US-style fair-use approach to permissible, unauthorised use of copyright works.
It is obvious why the US tech companies — with their cavalier and dismissive “move fast, break things” attitude — support this approach, which they have relied on to justify their unauthorised use of copyright-protected works for training AI.
In contrast, South Africa — like most countries in the world — has a system of “fair dealing” for determining permissible uses of copyright works. This means that there is a closed list of specified permitted uses, which is required under our international obligations.
On the other hand, the US-style fair-use approach gives judges — ex post facto — the discretion to determine whether a particular use of a copyrighted work is permissible, having regard to four factors.
The US approach allows corporations to claim their actions are permitted, requiring comparatively weak copyright owners and authors to sue them for copyright infringement.
South African law does not allow room for such an argument to be raised. We have a closed list of exceptions to copyright protection and, therefore, copyright owners have stronger — not weaker — protection against tech companies. It is, therefore, obvious why these tech companies would prefer that we introduce a US-style fair-use exception.
Given this imbalance in the US, and due to the costs and uncertainty that litigation under a fair-use system entails, copyright owners (and authors) are likely to agree to sub-optimal settlements. In fact, as The Citizen article acknowledges, Anthropic was partially successful on the basis of so-called “transformative” fair use in the initial litigation.
If anything, proponents of the CAB — particularly those in favour of a US-style fair-use system — have been rather quiet about the spate of fair-use litigation in the US over the past two years concerning the use of copyright works in training AI models.
There could be two reasons for this.
First, it rubbishes any suggestion the US-style fair-use approach to copyright exceptions is more responsive to technological development and results in greater legal certainty.
We have not seen this level of litigation in the countries that have a system of fair dealing.
Second, in supporting the CAB, they made common cause with the big tech companies or at least some of them — the same companies that have been accused of exploiting copyright works for their own ends.
We need to be clear that the compensation writers like Mda will receive from Anthropic is not attributable to the US’s fair-use system, but in spite of it. So much for really seeking to improve the lives of South African creators and copyright owners.
Professor Sadulla Karjiker holds the Anton Mostert Chair of Intellectual Property in the Faculty of Law at Stellenbosch University.