AI and Competition in the Press Sector : Spiil pleads for ensuring a level playing field between all types of AI models

Generative artificial intelligence technologies have ushered in a new area for the press industry. Opportunities in terms of innovation and efficiency gains are manifold. But so are the threats to quality information produced by professional journalists and to the press market sustainability.

As generative artificial intelligence is no longer limited to a simple technical assistance tool but is increasingly involved in the generation of content with little or no human supervision, it is necessary that the protection of copyright, the fair distribution of value and the fight against misinformation be guaranteed.

The compromise recently reached by European co-legislators on the AI Act represents an important step forward. Despite some significant advances, this compromise nonetheless fails to address all the issues of concern to publishers. In particular, Spiil makes the case that the consequences of AI use in terms of competition in the press industry must also be considered. Part of the European Commission’s consultation on competition in generative AI, this position paper touches upon two main points:

1) The need for ensuring a level playing field between all types of AI models (general purpose vs specialized, open source vs proprietary) and their different use cases (commercial vs non-profit as part of public-private partnerships) to make sure publishers’ copyright is respected under any circumstances.

2) The case for promoting greater scrutiny of AI generated contents to prevent an unfair competition with articles written by professional journalists.

1) The need for ensuring a level playing field between all types of AI models and their different use cases to make sure publishers’ copyright is respected under any circumstances. 

The AI Act and the related legislations (especially the 2019 copyright directive) take into account neither certain types of AI models nor some purposes for which they may be used. Spiil makes the case that these exceptions de facto grants a significant number of AI providers and private actors a competitive advantage, which may be used to circumvent obligations under copyright law.

Of particular concern are:

a) The exclusion of specialized AI models from the AI Act

While providers of generative AI tools will have to provide a sufficiently detailed summary about the content used for training their models, specialized models are excluded from the AI Act and thus exempted of any transparency requirements of this kind. Consequently, the exploitation of publishers content and the unfair competition with journalists working for specialized press outlets will continue. Certain specific generative AI tools are currently trained using very specialized data coming from specialized press outlets and will compete directly with the press outlets they are trained on.

b) Exceptions granted to general purpose AI models released under free and open sources licences

According to the compromise text, exceptions as regards the transparency related requirements imposed on general purpose AI models will be granted to models released under free and open source licenses, unless they are commercialized or can be considered to present systemic risk.

Even though these exceptions shall not exempt providers from respecting copyright obligations (i.e. the implementation of a policy to respect Union copyright law and the drafting of a sufficiently detailed summary about the content used for training of the general-purpose AI model), Spiil thinks that proprietary and open source generative AI models shall be placed on an equal footing with regard to all transparency requirements. Indeed, it is doubtful that open source models will effectively abide by copyright rules if they are otherwise released from other meaningful obligations, such as adherence to a code of practice, designed to ensure compliance with those rules.

c) Copyright exceptions for text and data mining as part of public-private partnerships.

According to article 3 of the 2019 copyright directive, public-private partnerships can avail themselves of the copyright exception to perform text and data mining for research purposes, provided they perform their activities on a non-profit basis. In addition, the conditions under which private companies gain access to contents are to be defined in a dedicated agreement. However, Spiil contends that these precautions fall short of an effective protection against copyright infringements. Unless the directive is revised, it is not clear how a private company could be prevented from taking advantage of the partnership to exploit data for commercial purposes. For example, a big tech company could be tempted to enter into a contract with a public institution of a member state just to escape their transparency obligations, without having any general interest purpose.

2) The case for promoting greater scrutiny of AI generated contents to prevent an unfair competition with articles written by professional journalists. 

Spiil is concerned that AI could further distort the press market and lead to a heightened competition between publishers who are genuinely committed to producing valuable articles to inform citizens’ judgment on the one hand and players who use AI to produce contents with little journalistic value or which deliberately spread misinformation on the other hand. In the current state of the French legislation, an online press service could indeed have access to press subsidies even though almost all its contents are automatically produced thanks to IA, provided this online press service employs at least one journalist.

In line with the proposals made by the Geste in a draft working document, Spiil finds it necessary to revise the criteria that are to be met to benefit from press subsidies. Furthermore, it cannot agree more with the proposition to enhance the value of analysis and curated information produced by professional journalists and make it more visible to the public (e.g. through watermarking technologies or other relevant labels).

Spiil is the French association of independent news publishers. Spiil gathers 270 press companies that publish more than 350 press titles with all sorts of economic models. Spiil advocates for a fair and sustainable press market. Spiil is a member of News Media Alliance, a European association of press associations.