21.1 Introduction
In his 2019 Global Survey of Journalism and Artificial Intelligence, Charlie Beckett concluded that ‘robots are not going to take over journalism.’Footnote 1 While machines may play an increasingly important role in routine journalism labour, Beckett saw the reality and potential of artificial intelligence (AI), machine learning, and data processing in ‘giv[ing] journalists new powers of discovery, creation and connection.’Footnote 2 This conclusion gave hope that the ‘human factor’ – journalistic work by authors of flesh and blood – would remain decisive in news and media productions. With the evolution of generative AI systems, this assumption has become doubtful. Generative AI systems have an increasing impact on news production – in the sense of an increasing potential to replace human journalistic work. Early studies show that substitution effects may be felt quite strongly in the news and media sector.Footnote 3 As generative AI systems are capable of providing content much faster and cheaper, human journalists – writers, photographers, illustrators, filmmakers, and others (collectively referred to as ‘journalists’ in the following analysis)Footnote 4 – may face shrinking market share and loss of income.
A closer look at Charlie Beckett’s 2019 Global Survey reveals that, already at that time, AI was taking over writing tasks. Beckett reported that AI had been employed to generate content for news apps, social media posts, and tweets, including not only text but also illustrations.Footnote 5 AI had also been used to contribute to local news and to write election reports for individual cities in France, based on data covering population, location, mayor, previous election results, wealth, employment rate, and so on.Footnote 6 Further examples concerned the algorithmic creation of football match commentaries, reports on traffic accidents, and weather forecasts. AI was also used to generate headlines and summaries of press releases.Footnote 7 Considering the evolution of generative AI systems, it is now imaginable without much difficulty that AI will conquer more and more territory in journalism that, traditionally, was reserved for humans.Footnote 8 The use of AI for routine journalism labour appears as a precursor of a much broader and increasing replacement trend.Footnote 9 Inevitably, the increasing sophistication of generative AI systems will disrupt the market for human journalistic work.Footnote 10
Considering this trend, the question arises whether copyright law could be employed as a legal tool to support human journalistic labour. Establishing payment obligations for AI use and exploitation in the press and media sector, the protection system could channel money from providers and users of generative AI systems to journalists. The underlying idea is simple: by imposing an obligation to pay remuneration for the use of AI systems and ensuring that this money reaches individual journalists, copyright law could transform AI content revenue into human content revenue.Footnote 11 Ultimately, this approach should enhance the chances of human journalists to continue their socially valuable work.
As simple as the formulation of this policy goal may be, the implementation of the underlying approach raises delicate legal-doctrinal and practical questions. First, copyright law would have to develop remuneration mechanisms that generate a revenue stream from the AI industry to the press and media sector. Second, copyright law would have to ensure that money accruing from AI levies does not simply fill the pockets of press and media companies. Instead of passing on AI revenue to individual journalists, press publishers and other media entrepreneurs may prefer to invest in the automation of journalistic labour.Footnote 12 As already indicated, the increasing use of AI can reduce the costs of news and media production. It would come as a surprise if the prospect of lower costs escaped the attention of press and media businesses. Copyright rules in this area, therefore, must address not only potentially divergent interests of AI and media companies but also the internal tension between press and media entrepreneurs and individual journalists. To the extent that press publishers and media companies employ AI technology to replace human journalistic labour, the obligation to pay remuneration for the use of AI would have to rest on both AI companies offering generative AI systems and media companies using these systems.
From a public policy perspective, there can be little doubt that it is worth while to explore regulatory avenues that could lead to new money streams for human journalistic work. Various arguments offer strong support for the introduction of a remuneration system in favour of human journalists (see Section 21.2): among them are the parasitic use of human press and media productions for AI training purposes; the central function of public interest journalism in society;Footnote 13 and the socio-political objective of helping journalists who face displacement effects. To implement effective remuneration mechanisms in practice, the two steps must be taken that have already been mentioned. First, a money flow from the AI industry to the press and media industry must be set in motion. Second, this money must be directed to individual journalists.
Seeking to achieve these goals, the legislator has the choice between two different reference points. On the one hand, it is conceivable to impose an obligation to pay remuneration at the stage of AI training. The machine cannot produce results resembling human press and media content unless it has had the opportunity to analyse relevant human creations (input dimension). The use of countless human works for AI training offers a starting point for developing a remuneration system that seeks to compensate human authors for the use of their works to build a machine that has the potential to become a serious competitor. In EU copyright law, the provisions on text and data mining (TDM) offer a basis for initiating a money stream at AI training level. However, a closer look at the EU approach reveals that the current configuration of relevant provisions in EU copyright law and the AI Act (AIA)Footnote 14 may make it difficult to ensure that this money finally flows to individual journalists (Section 21.3).
On the other hand, the lawmaker can introduce an obligation to pay remuneration for the supply of generative AI products and services and the use of AI systems by press and media companies (output dimension). Instead of requiring the payment of remuneration at the training level, a ‘levy’ could be imposed on the offer and use of AI systems capable of substituting for human press and media productions. This output-oriented AI levy could then be used to offer financial support for human journalists.Footnote 15 To make sure that AI remuneration is passed on to individual human creators, providers of AI systems and users in the press and media sector could be under an obligation to pay remuneration to collecting societies, which would then use the revenue from remuneration payments to support journalists and their work.Footnote 16 In combination with mandatory collective rights management, this new revenue stream could be used to finance social and cultural funds for journalists of flesh and blood. At the same time, the AI levy could make the use of generative AI systems in the press and media sector more expensive. Adding levy payments as an additional cost factor, the remuneration system could contribute to the reduction of the price advantage following from the fact that media organisations embracing AI systems need not pay honoraria or salaries for journalistic labour (Section 21.4).Footnote 17
Weighing the arguments for and against these different implementation strategies, a legislative approach focusing on the output dimension and seeking to introduce a levy for the offer and use of generative AI systems in the press and media sector seems more promising than an approach taking input and training activities as a reference point for remuneration payments (as will be discussed in Section 21.5).
21.2 Three Good Reasons for the Payment of Remuneration
Generative AI systemsFootnote 18 are capable of mimicking human news and media productions only because human works have been used as training material.Footnote 19 On the basis of existing literary and artistic creations that serve as input data, machine-learning algorithms are able to recognise patterns and similarities. Following this deductive method, a generative AI system learns how to produce output that imitates human works. Its machine-learning algorithm enables the AI system to generate journalistic content on its own – based on the computational analysis of human works that served as training material.Footnote 20 Taking this insight as a starting point, it becomes possible to lay theoretical groundwork for the introduction of a remuneration mechanism in favour of human journalists. Central to this are considerations relating to the parasitic use of human creations for AI training (Section 21.2.1), the contribution of human journalistic work to the improvement of social and political conditions (Section 21.2.2), and the socio-political need to support journalists who face displacement effects (Section 21.2.3).
21.2.1 Parasitic Usurpation
First, it can be said that journalists should be compensated for the parasitic usurpation of the market for human news and media content production. The machine is capable of mimicking human press and media content only after it has had the opportunity to derive patterns for its own productions from myriad human creations that serve as training resources, including journalistic texts, photos, illustrations, and audiovisual material.Footnote 21 Journalistic training material can be of particular importance in the development of AI models that can be used in the press and media sector, as it reflects typical stylistic elements of journalistic work. Newspaper articles, for example, may contain valuable data points that offer information on specific writing styles, word choices, and expressions that readers expect from a journalistic text. From this perspective, it is only fair that human journalists – providing important elements of the source material for AI ingenuity in press and media contexts – receive a remuneration when AI productions finally kill the demand for human journalistic work.Footnote 22 This line of argument plays a central role in the lawsuits that several US newspapers, including the New York Times, have brought against OpenAI and Microsoft.Footnote 23
Admittedly, the parasitic usurpation argument appears less convincing when considered in light of the full spectrum of literary and artistic content that may be used for machine learning. AI models capable of eroding the market for human journalistic productions may have been trained with work repertoires that go far beyond journalistic texts, photos, and videos. The spectrum of writings used for an AI model that generates natural-language text may be much more diverse than newspaper articles. It is unlikely that photo libraries used for the training of an image-generating AI model are confined to press photographs. Hence there might be an incongruence between general training resources (works created outside the journalistic domain) and the specific circle of beneficiaries – human journalists – who should be entitled to remuneration to soften displacement effects. However, this incongruence need not pose insurmountable hurdles when considering further arguments in favour of remuneration for journalists.
21.2.2 Improvement of Societal Conditions
More concretely, there is a second argument which, instead of looking at the source material used for AI training, emphasises the particular importance of journalistic work to society as a whole. From this overarching policy perspective, it can be stated that remuneration for the supply and use of generative AI systems in the press and media sector should be due to human journalists because their work has particular societal value. It would be undesirable if despite displacement effects in the press and media sector, providers and users of generative AI systems could escape the obligation to remunerate journalists by simply going beyond journalistic productions and diversifying the spectrum of literary and artistic material used for AI training. As demonstrated in the cultural sciences,Footnote 24 literary and artistic productions made by authors of flesh and blood provide important impulses for social and political changes. Human expression not only addresses shortcomings of present society and defects of social and political conditions; with critical commentary and alternative visions of society, it also prepares society for the transition to being a better community.Footnote 25 The work of human journalists can serve as an example. AI-generated productions in the press and media field are incapable of providing comparable impulses for the improvement of societal conditions. An AI system may manage to imitate human journalistic productions.Footnote 26 But AI systems do not perceive and experience social, cultural, economic, and political conditions as humans do. They are not affected by societal conditions in the same way as humans.Footnote 27 Problematic developments offer starting points for human journalists to analyse reasons and origins; criticise responsible groups, leaders, and institutions; propose alternatives; and bring all this to the attention of readers, listeners, and viewers in order to change society for the better. Investigative journalism can pave the way for changes by revealing scandals and shortcomings in societal subsystems, such as the fields of political and economic power.Footnote 28 As has been pointed out, AI systems may support the work of journalists. Providing superhuman possibilities of data analysis, they may give journalists unprecedented powers of discovery.Footnote 29 AI systems operating independently, however, cannot free themselves from the data input fuelling their algorithms.Footnote 30 Instead of arriving at valid criticism and alternative visions of society, they simply recombine the data used for training purposes. While, for example, a generative AI system will have little difficulty in producing endless variations of known press and media content, it cannot be expected to interpret its societal impact and evaluate the desirability of potential societal changes. Even highly problematic false, harmful, and extreme messages may be multiplied and amplified without consideration of their societal repercussions.Footnote 31
To preserve the central societal function of human journalistic work, it is thus important to ensure that journalists survive the dethroning of the human author. The introduction of a remuneration system that channels money to human journalists makes sense from this perspective. It prevents the loss of important impulses for the improvement of social and political conditions. By leaving press and media production to AI systems, society weakens its ability to evaluate and renew itself. With the introduction of an AI remuneration system, society can reverse this trend. Offering financial support for journalists, such a system can preserve the important function of human journalistic work in providing ongoing impulses for the improvement of societal conditions. Hence, there is substantial reason to introduce remuneration mechanisms that focus on journalists.Footnote 32
21.2.3 Socio-Political Need to Cushion Displacement Effects
If AI systems in the press and media sector are trained on a broader range of literary and artistic works – not only journalistic texts, photos, and videos – it can be argued that a remuneration system focused on journalists allows authors in the press and media sector to obtain benefits that result, at least in part, from the use of creations made by authors in various other sectors. As already indicated, this objection should not free providers and users of AI systems in press and media contexts from the obligation to pay remuneration. It is an issue of distributing AI revenue – and of solidarity among authors. Arguably, journalists exposed to displacement effects in the press and media sector have a stronger and more direct entitlement to remuneration for AI use in that sector than authors in other sectors whose works an AI crawler has amassed to maximise training resources. Conversely, it can be said that remuneration for AI use in other creative sectors should be due to authors struggling with displacement effects in these other sectors. It should not be paid to journalists even if journalistic works played a role in the training of AI systems that are used in these other sectors.
Based on this premise, a stronger, more direct entitlement of journalists to revenue accruing from the use of AI systems in the press and media sector can finally be derived from a broader socio-political consideration that has particular importance. Inevitably, the displacement of journalists and the disruption of the market for human press and media productions require adequate countermeasures and investment. Journalists who lose their jobs will need financial support. Investment in training activities will be necessary to enable them to obtain new skills and credentials. The introduction of a remuneration system that focuses on journalists is an important and desirable step against this backdrop. If the remuneration for the use of AI in the press and media sector were more widely distributed, the potential of this measure to cushion the disruptive effect on journalists would be reduced. Admittedly, general tax money could be used to enable journalists to adapt to the challenges of increasing AI use. In comparison to a tax-based model, however, the copyright framework offers crucial advantages. It does not raise the spectre of censorship and is more resilient. While the next economic crisis may lead to cuts in the use of taxpayers’ money, copyright measures can weather market cycles. By imposing a journalist-centric remuneration obligation on providers and users of generative AI systems in the press and media sector, the legislator adopts a targeted approach: the industry players – in both the AI sector and the media sector – causing the disruption are required to contribute to the financing of measures that mitigate the corrosive effect on human journalistic work. With collecting societies and their remuneration and repartitioning schemes, the copyright system offers a well-established infrastructure for the appropriate distribution of collected money – in the sense of a distribution scheme that ensures a direct money flow to human journalists.Footnote 33
Hence, there are several good reasons for the introduction of remuneration mechanisms that ensure the payment of remuneration for the use of generative AI systems to replace human journalistic labour.Footnote 34 As already indicated, this insight gives rise to the question of how best to implement remuneration mechanisms in practice. On the one hand, the focus could be on the input dimension: the use of human journalistic content for AI training purposes. On the other hand, the final output – the offer of generative AI products and services in the marketplace and the use of these systems by press and media organisations – could serve as a reference point for payment obligations. To identify the preferable implementation strategy, it is necessary to embark on a closer inspection of both approaches.
21.3 Input/Training Dimension
When remuneration mechanisms are aligned with the input dimension, particular importance is attached to the fact that human works are used to train AI systems. With the evolution of generative AI – capable of replacing human literary and artistic works – this use of human source material for machine training purposes has triggered strong statements accusing the AI industry of the parasitic use of human works. In the legislative process leading to AIA adoption in the EU,Footnote 35 the Authors’ Rights Initiative – over forty associations and trade unions representing authors, performers, and copyright holders in various creative industry segments – stressed in its ‘call for safeguards around generative AI’ that:
[t]he output of AI systems depends on the input they are trained with; this includes texts, images, videos and other material from authors, performers and other copyright holders: Our entire digital repertoire serves training purposes, often without consent, without remuneration and not always for legitimate uses. The unauthorised usage of protected training material, its non-transparent processing, and the foreseeable substitution of the sources by the output of generative AI raise fundamental questions of accountability, liability and remuneration, which need to be addressed before irreversible harm occurs.Footnote 36
Similarly, the European Guild for Artificial Intelligence Regulation (EGAIR) adopted the ‘EGAIR Manifesto’, pointing out that:
the products sold by AI companies are the result of operations on datasets, which contain all sorts of data, including millions of copyrighted images, private pictures and other sensitive material. These files were collected by indiscriminately scraping the internet without the consent of the owners and people portrayed in them and are currently being used by AI companies for profit.Footnote 37
A ‘Joint statement from authors’ and performers’ organisations on artificial intelligence and the AI Act’ warned that
AI technologies increasingly use authors’ and performers’ works and creations to ‘feed’ and train their applications without their consent or knowledge, in breach of authors and performers’ rights granted under international, EU or national laws. In this era of rapidly advancing AI technologies, whose principle consists solely of copying and mixing, we must highlight the urgent need to protect the works and performances of professional authors and performers from misappropriation. Not only to preserve their livelihoods, but also to inform citizens about the use of original works by AI applications.Footnote 38
The final AIA text shows that these initiatives had a deep impact on the parliamentary debate and the trilogue phase in which the European Commission, the Council, and the European Parliament established the definite version of the new piece of legislation. Recital 105 AIA addresses ‘[g]eneral-purpose models, in particular large generative models, capable of generating text, images, and other content.’Footnote 39 Recognising potential corrosive effects on human creativity, it points out that these models ‘present unique innovation opportunities but also challenges to artists, authors, and other creators and the way their creative content is created, distributed, used and consumed.’Footnote 40 The Recital also emphasises that the development and training of generative AI models ‘require access to vast amounts of text, images, videos, and other data. Text and data mining techniques may be used extensively in this context for the retrieval and analysis of such content, which may be protected by copyright and related rights.’Footnote 41
After this statement of the problem, Recital 105 confirms that the use of literary and artistic works for AI training purposes has copyright relevance and involves acts of text and data mining that require the authorisation of rightholders: ‘[a]ny use of copyright protected content requires the authorisation of the rightsholder concerned unless relevant copyright exceptions and limitations apply.’Footnote 42 As requested by authors, performers, and creative industry branches, the EU legislator thus clarified that authors and industry rightholders can exercise control over the use of human works during AI training processes on the basis of copyright protection – unless a copyright exception applies.
Prior to the AIA, the TDM discussion in EU copyright law had already culminated in the introduction of rules that could be understood to confirm the copyright relevance of AI training processes, such as the use made of protected works during the training of generative AI systems. The specific TDM provisions in Articles 3 and 4 of the 2019 Directive on Copyright in the Digital Single Market (CDSMD)Footnote 43 set forth two specific exceptions to copyright, related rights, and database protection that play an important role in the context of TDM projects that require the extraction of data from protected literary and artistic works and/or databases, including journalistic content. Addressing scientific research directly, Article 3(1) CDSMD sets forth an obligation for Member States to exempt from copyright, related rights, and sui generis database protection acts of copying that research organisations or cultural heritage institutions carry out in the context of scientific research that involves TDM.Footnote 44
In addition to this exemption of scientific TDM, Article 4(1) CDSMD contains a general exemption of TDM that is not limited to scientific research. Under this broader provision, anyone, including commercial AI system developers and trainers, may make copies of works, performances, or databases for the purposes of TDM and retain them as long as necessary for the AI training process.Footnote 45 With regard to this broader category of TDM outside the scope of the scientific research rule in Article 3 CDSMD, Article 4(3) CDSMD adds an important nuance by stipulating that rightholders can reserve their rights. The provision contains the following opt-out mechanism: ‘The exception or limitation provided for in paragraph 1 shall apply on condition that the use of works and other subject matter referred to in that paragraph has not been expressly reserved by their rightholders in an appropriate manner, such as machine-readable means in the case of content made publicly available online.’Footnote 46
Arguably, the adoption of specific copyright exceptions for TDM had already established the copyright relevance of TDM and related AI training processes. As the CDSM Directive dates back to 2019, however, it could also be argued that the EU legislator did not have in mind the use of copyrighted material as mere data input for generative AI training purposes.Footnote 47 In the TDM debate, it has been underlined around the globe that TDM copies have a specific nature. They fall outside the concept of reproduction in the traditional sense of making copies for the purpose of consulting and enjoying a work.Footnote 48 After the confirmation of copyright relevance in Recital 105 AIA, however, the power of persuasion of this argument vanishes with regard to the regulation of AI training in the EU. Without distinguishing between use of ‘works as works’ and use of ‘works as data’,Footnote 49 Recital 105 AIA confirms that EU copyright law brings all forms of TDM under the umbrella of the right of reproduction and, thus, requires the invocation of a copyright exception, such as the scientific research rule in Article 3 CDSMD, the broader exemption in Article 4 CDSMD, or the long-standing temporary copying rule in Article 5(1) of the 2001 Information Society Directive.Footnote 50
21.3.1 Money Flow from the AI Industry to the Press and Media Sector
In the case of commercial AI training falling under Article 4(1) CDSMD, this configuration of the right of reproduction also means that EU copyright law brings TDM activities within the reach of copyright holders in the press and media sector seeking to receive a remuneration for the use of human journalistic works in AI training.Footnote 51 Referring to the opt-out mechanism in Article 4(3) CDSMD, the AI Act confirms the intention to give rightholders the opportunity to exercise control over the use of their works for AI training purposes in Article 4 CDSMD scenarios: ‘Where the rights to opt out has been expressly reserved in an appropriate manner, providers of general-purpose AI models need to obtain an authorisation from rightsholders if they want to carry out text and data mining over such works.’Footnote 52 In accordance with Article 4(3) CDSMD, rightholders can exclude TDM via a machine-readable rights reservation. This means that AI trainers must take into account metadata, such as robots.txt files, but also the terms and conditions of a website or online service, such as online portals of press publishers and other media outlets, in order to assess whether TDM is permitted with regard to a particular work.Footnote 53 In principle, rightholders in the press and media sector can thus rely on technical safeguards, such as robots.txt files, to prevent the use of human journalistic works for AI training purposes.
As in other cases where copyright holders can refuse permission for a given form of use, this veto right can pave the way for negotiations and licensing agreements.Footnote 54 It is conceivable that the rights reservation option in Article 4(3) CDSMD leads to machine-readable rights reservation protocols that express different rightholder standpoints. One standpoint could be robots.txt that signals an outright exclusion of any use of human journalistic content for AI training purposes. Using this version of robots.txt, rightholders can express their preference for an outright prohibition. They can altogether prevent TDM of their press and media content. An alternative standpoint, however, could be robots.txt that prohibits use for AI training purposes only if the AI trainer behind the crawler is reluctant to pay remuneration. Using this alternative version, rightholders can thus express their willingness to permit the use of human journalistic content with the payment of remuneration. In other words: in an ideal world, the rights reservation option in Article 4(3) CDSMD serves as a catalyst to arrive at generally agreed, machine-readable remuneration protocols in the press and media sector that trigger an automated process for the payment of remuneration.Footnote 55
Unfortunately, it may be quite difficult to achieve this ideal result in the real world. Even if standardised rights reservation protocols – capable of expressing remuneration wishes and modalities – become available, it is unclear whether copyright holders in the press and media sector will manage to create efficient, pan-European rights clearance solutions that offer reliable and well-functioning payment interfaces with the technical safeguards – robots.txt files for example – that express the electronic remuneration caveat. As long as the automated, machine-based identification of rightholders and the automated processing of payments remains complicated or unreliable, the rights reservation option in Article 4(3) CDSMD is unlikely to pave the way for a remuneration system that covers a broad spectrum of news and media providers.Footnote 56 The training of generative AI systems requires the availability of vast amounts of journalistic work. The moment AI trainers are obliged to check rights ownership, observe specific payment conditions, and obtain permission at the level of individual works or databases, the burden of rights clearanceFootnote 57 can easily lead to a situation where licensing takes place only between big players: AI companies entering into umbrella licensing agreements with content majors – the biggest rightholders – in the press and media sector.Footnote 58
21.3.2 Extra Income for Individual Journalists
It is foreseeable that the rights clearance obstacles described here – and the potential predominance of licensing deals at the level of big press and media companies – will make it difficult for individual journalists to benefit from new AI training revenue. If rights clearance solutions become available, these solutions will most probably be the result of industry collaboration: the press and media industry agrees with the high-tech industry on conditional rights reservation protocols that make use of human journalistic content possible the moment the desired remuneration has been paid. As with all types of industry collaboration, this approach raises the question of whether the new revenue stream accruing from AI training will ever reach individual journalists.
In the press and media sector, copyright relating to human journalistic works will normally be in the hands of press publishers and other media content producers. Freelance journalists will hardly ever have sufficient bargaining power to negotiate exploitation contracts that allow them to keep copyright. Instead, these contracts will require the transfer of copyright in its entirety – for the entire term of copyright, with regard to all territories, forms of exploitation, types of use, and so on. This means that AI developers have good chances of acquiring valid permissions for AI training from press and media companies without there being any need to approach individual freelance journalists or their collective rights management organisations (CMO). The only exception would be in copyright systems that contain specific safeguards to protect authors against an overbroad loss of exclusive rights. Copyright contract law in Germany and the Netherlands, for example, follows the maxim in dubio pro autore: a freelance journalist is deemed to have granted press publishers and media content producers only those exclusive rights that are explicitly mentioned in the exploitation contract. In case of doubt, the law assumes that the contract covers only the rights necessary for the envisaged exploitation. All other rights remain in the hands of the journalist.Footnote 59
This configuration of copyright contract law impacts rights clearance for AI training when it is assumed that the use of copyrighted works for AI training purposes is a new, formerly unknown type of use. As already indicated, the discussion on TDM offers several reference points to build this argument, as TDM copies have a specific nature. They fall outside the concept of reproduction in the traditional sense.Footnote 60 As Rosanna Ducato and Alain Strowel have explained in the European TDM discussion:
when acts of reproduction are carried out for the purpose of search and TDM, the work, although it might be reproduced in part, is not used as a work: the work only serves as a tool or data for deriving other relevant information. The expressive features of the work are not used, and there is no public to enjoy the work, as the work is only an input in a process for searching a corpus and identifying occurrences and possible trends or patterns.Footnote 61
Hence there are good reasons to assume that copying human journalistic works for the purposes of automated, computational TDM constitutes a new, formerly unknown reproduction category that falls outside the scope of exploitation contracts concluded prior to the AI revolution in copyright systems following the principle in dubio pro autore.Footnote 62 Rightholders in the press and media sector escape this conclusion only when they use exploitation contracts that cover all types of use – now known or hereafter devised.Footnote 63 In such a case, a journalist’s chances of being remunerated for use in AI training depends on further provisions in copyright contract law that specifically address the grant of rights relating to unknown forms of use. For instance, Section 31a(1) of the German Copyright Act stipulates that an author may revoke the grant of ‘rights in respect of unknown types of use’.Footnote 64 Paragraph 2 adds that ‘[t]he right of revocation does not apply where the parties, upon becoming aware of the new type of use, have agreed on remuneration in accordance with section 32c(1)’. This latter provision clarifies that authors – journalists in the case of press and media productions – are entitled to ‘separate equitable remuneration where the other contracting party commences a new type of use of the author’s work pursuant to section 31a which was agreed upon, but still unknown, at the time the contract was concluded’.Footnote 65
Under German copyright contract law, freelance journalists may thus have the chance of obtaining an additional remuneration if TDM is qualified as a type of use that was unknown at the time of concluding the exploitation contract with a press publisher or other media company – a ‘separate’ remuneration that comes on top of the initially agreed remuneration under the contract.Footnote 66 In copyright systems without a comparable cascade of provisions seeking to generate an extra flow of revenue, the success of remuneration claims after the transfer of copyright to an exploiter of the journalistic work depends on more general provisions of copyright contract law. Harmonising copyright contract law in the EU, Article 18(1) CDSMD obliges Member States to ensure that ‘where authors and performers license or transfer their exclusive rights for the exploitation of their works or other subject matter, they are entitled to receive appropriate and proportionate remuneration.’Footnote 67
In the absence of decisions of the Court of Justice of the European Union (‘CJEU’), however, it is unclear what impact new, formerly unknown types of use may have on the assessment of whether the contractually agreed remuneration constitutes an ‘appropriate and proportionate remuneration’ in the sense of Article 18(1). Under Article 20(1) CDSMD, journalists can demand a contract adjustment ‘when the remuneration originally agreed turns out to be disproportionately low compared to all the subsequent relevant revenues derived from the exploitation of the works or performances’.Footnote 68 However, this threshold appears relatively high in TDM cases – despite the intention to relax the disproportionality test in comparison with traditional, so-called bestseller clauses in domestic Member State legislation which, for example, required evidence of ‘gross’ disproportionality.Footnote 69 Even if TDM licenses lead to extra income, it remains to be seen whether this new revenue stream will ever become such an important income component that, factoring TDM fees into the equation, the originally agreed contractual remuneration (which did not consider this new, unknown type of use) appears ‘disproportionately low’ within the meaning of Article 20(1) CDSMD.
In addition to these unresolved issues of copyright contract law, empirical studies show that freelance journalists may fear negative reactions in the press and media sector when they insist on their right to fair remuneration or, more specifically, additional remuneration for formerly unknown TDM use.Footnote 70 Facing a relatively small circle of press and media producers, a journalist may be concerned about seeing their name being added to a ‘blacklist’ of ‘difficult’ persons with whom press and media organisations do not want to work because of past disputes about insufficient remuneration.Footnote 71 If a journalist does not work as a freelancer, but as an employee, copyright rules on fictitious employer’s authorship may moreover stipulate that the employer acquires copyright directly. In national systems with this fictitious authorship rule, press and media organisations become the first owner of copyright – without any transfer or licence.Footnote 72 Extra TDM income, then, depends on the employee’s chances of receiving a pay rise.
On balance, copyright contract law does not offer journalists a particularly powerful arsenal of legal tools. Instead, it is an open question whether freelance journalists will have success when they try to obtain a fair share of TDM revenue that may accrue from the reservation of copyright under Article 4(3) CDSMD and subsequent high-level licensing deals between the AI industry and press and media organisations. Considering the legal uncertainty surrounding claims for additional remuneration, it seems much more realistic to assume that new TDM income will predominantly fill the pockets of large press and media companies that own impressive repertoires of journalistic works.Footnote 73 Individual journalists whose works form part of these repertoires, however, will not necessarily receive higher honoraria or an appropriate share of the TDM income.Footnote 74 Copyright legislation focusing on the input/AI-training dimension may thus generate a new flow of revenue from the AI industry to the press and media sector. The opt-out mechanism in Article 4(3) CDSMD – now flanked by the provisions of the AI Act – can serve as an example. Individual journalists facing displacement effects, however, are unlikely to benefit from this new revenue stream.
21.4 Output/Substitution Dimension
Considering these drawbacks of an input/AI-training-based remuneration regime, such as the system following from the opt-out mechanism in the CDSM Directive and the new rules in the AI Act, it seems particularly important to explore alternative approaches that may strengthen not only the position of press and media companies vis-à-vis the AI industry but also the position of individual journalists. As already indicated, a remuneration mechanism in favour of journalists need not focus on the AI training phase. Instead, the final offer of AI services and products on the market can serve as a reference point for a legal obligation to pay remuneration (output dimension). More specifically, it seems possible to establish a lump-sum remuneration system that channels to human journalists a certain share of the revenue accruing from the supply and use of generative AI systems in the press and media sector.Footnote 75
21.4.1 Towards an AI Levy System
Following this alternative approach, providers of generative AI systems and users in the press and media sector would be obliged to pay remuneration for the production of content that has the potential to substitute for human journalistic works. Surveying the canon of copyright rules, it becomes apparent that a lump-sum remuneration approach is not entirely alien to the protection system. The existing legal obligation to pay remuneration for the use of phonograms could serve as a blueprint for this new remuneration regime. Using Article 8(2) of the Rental, Lending and Related Rights DirectiveFootnote 76 as a model, the rule could take the following shape:
Member States shall provide a right in order to ensure that a single equitable remuneration is paid by providers and users of a generative AI system, if a literary and artistic output generated by the system has the potential to serve as a substitute for a work made by a human journalist, and to ensure that this remuneration is paid to social and cultural funds of collective management organisationsFootnote 77 for the purpose of fostering and supporting human journalistic work.Footnote 78
Admittedly, this text may require further refinement and clarification before it can be adopted as a legal basis for the introduction of an AI levy system that supports journalists. Potential definition hurdles, however, seem surmountable.Footnote 79 As to the question of what output quality is necessary to assume a substitution risk, it must be considered that this new rule would aim at establishing a lump-sum remuneration system. It does not require a fine-grained, precise analysis that determines meticulously what specific amount of remuneration can be deemed appropriate for each and every human journalistic contribution that is affected by the use of generative AI systems. Nor does it require evidence that a specific human journalistic production has not taken place because of the use of generative AI. Instead, all that is needed is a general, abstract assessment of whether, in principle, an AI system is capable of substituting for human press and media content. If such a system is offered or used in the press and media sector, this should be sufficient to confirm a disruptive effect and impose a payment obligation.
The general conceptual contours of the proposed lump-sum remuneration approach would be as follows. The system would serve the overarching purpose of creating a new revenue stream to support the work of human journalists. Revenue accruing from the payment of remuneration for the offer and use of generative AI systems in the press and media sector would be channelled to collecting societies that would use the money to support journalistic work. If this revenue system is combined with inescapable, mandatory collective rights management, the repartitioning schemes of CMOs can ensure that, in compliance with their statutory purposes and rules for social and cultural funds,Footnote 80 individual journalists can benefit directly from the extra income accruing from remuneration payments. To achieve this result, they do not have to invoke the complex rules of copyright contract law discussed in Section 21.3. Perhaps more importantly, they do not have to run the risk of blacklisting.Footnote 81 The distribution machinery of CMOs ensures that journalists receive a share of AI revenue automatically.Footnote 82
21.4.2 Extra Income for Individual Journalists
In contrast to the industry collaboration that is likely to arise from an input/AI-training-based approach and the reservation of rights under Article 4(3) CDSMD, this output-oriented remuneration approach does not give rise to concerns that collected levies will hardly ever reach individual journalists.Footnote 83 By imposing an obligation to pay remuneration on both ‘providers’ and ‘users’ of generative AI systems, the proposed remuneration mechanism addresses the internal tension between the interests of the press and media industry on the one hand and the interests of individual journalists on the other. Press and media organisations may embrace AI to reduce production costs.Footnote 84 Individual journalists, by contrast, have a strong interest in compensation for the reduction of their market share and income.Footnote 85 Taking these divergent interests into account, the explicit reference to generative AI ‘users’ in the proposed provision solves this tension in favour of journalists: the moment a press publisher or media company employs generative AI technology to replace human journalistic labour, the proposed remuneration system ensures that this press or media organisation is bound to pay remuneration for the use of AI. Under the remuneration rule developed here, not only the offer of generative AI systems by the high-tech industry (providers) but also the use of such systems in press and media productions (users) triggers the payment obligation. As a result, the obligation to pay remuneration rests on both AI companies offering generative AI systems and media companies using these systems in the press and media sector.
21.4.3 Focus on Public Interest Journalism
More concrete guidelines for the use of collected revenue can be derived from the three objectives described here. Following the argument that the remuneration system offers compensation for the parasitic use of human works for the purpose of enabling AI systems to kill the market for human creativity in the press and media sector (first argument), collected money could be used broadly to support human journalistic productions. A more targeted approach, by contrast, follows from the insight that AI-generated press and media content leads to a loss of human journalistic works that uncover societal problems and provide impulses for improving societal conditions (second argument). To allow human journalists to continue their socially valuable work, the establishment of funds seeking to promote human journalistic productions – in particular in the area of investigative journalism – seems warranted. Finally, the general socio-political goal of supporting journalists who lose their jobs due to competing AI content (third argument) justifies the establishment of social funds that serve as insurance against displacement effects caused by generative AI systems. The three rationales developed here offer a basis for different measures ranging from the establishment of a general repartitioning scheme to more specific, targeted investment in social and cultural funds.
Importantly, the repartitioning schemes of CMOs may even offer room for using AI revenue to subsidise socially valuable public interest journalism.Footnote 86 In Amazon – a case about the payment and repartitioning of private copying levies in Austria – the CJEU confirmed that EU law offers considerable flexibility with regard to the use of collected funds for social and cultural purposes. One of the questions asked by the Austrian Supreme Court was whether a collecting society lost its right to the payment of fair compensation if, in relation to half of the funds received, the collecting society was required by law not to pay the levy income to the persons entitled to compensation but to distribute it to social and cultural institutions.Footnote 87
Answering this question, the CJEU held the view that EU law did not contain an obligation to pay all the lump-sum remuneration collected via levy systems, such as the levy system for private copying, directly to rights owners in cash. Instead, a Member State was free to provide that part of the lump-sum remuneration be distributed in the form of indirect compensation through social and cultural institutions set up for the benefit of authors and performing artists.Footnote 88 The fact that the remuneration had to be regarded as recompense for a specific harm did not constitute an obstacle to the establishment of such an indirect payment mechanism through the intermediary of social and cultural institutions.Footnote 89 The Court also stated that a system of indirect distribution of collected funds was conducive to ensuring that European cultural creativity and production received the necessary resources. It further safeguarded the independence and dignity of authors and performers.Footnote 90 The Court made it a condition, however, that the social and cultural establishments involved actually benefit those entitled to the lump-sum remuneration (journalists in the case of the AI levy system proposed here). Moreover, it was necessary that the detailed arrangements for the operation of social and cultural institutions were not discriminatory. Benefits had to be granted to those persons entitled to remuneration and the system had to be open to nationals and foreigners alike.Footnote 91
Arguably, this decision makes it possible to adopt measures to offer extra support for journalistic work with particular societal relevance. If it is legitimate to use half of the revenue accruing from AI remuneration for social and cultural purposes, it also seems possible to devote particular attention to the furtherance of public interest journalism when taking decisions on the distribution of this substantial share of the collected money.Footnote 92
However, the decision of the CJEU in Amazon sheds light on two potential obstacles. First, the Court made it clear that the use of funds by social and cultural institutions had to constitute an indirect form of payment for those entitled to the collected remuneration. In the Amazon case, the remuneration was the result of private copying legislation providing for the payment of fair compensation for the damage caused by acts of private copying.Footnote 93 Against this background, the question arose as to what extent the partitioning of collected funds had to relate directly to the losses suffered by individual groups of authors. A similar question can be formulated with regard to AI remuneration: Is it necessary to align the distribution of collected AI levies strictly with the degree of displacement effects in different branches of journalism? If a strict link to the loss of job opportunities in a given segment is necessary, it seems difficult to spend a higher proportion of the collected remuneration on programmes supporting public interest journalism unless this type of journalism is also affected by replacement effects to a greater extent.
In Amazon, however, the CJEU referred to the fact that it was very difficult, if not impossible, to calculate the individual damage that an author suffered because of private copying. Considering this difficulty, the Court underlined that Member States enjoyed wide discretion in determining the form, the detailed arrangements, and the possible level of lump-sum remuneration.Footnote 94 In the exercise of this wide discretion, Member States were free to establish a system of indirect remuneration via social and cultural institutions.Footnote 95 Hence the Court itself did not insist on a system that distributes collected money meticulously on the basis of the individual harm suffered by an author because such a detailed calculation of individual damage was hardly possible. A parallel between this aspect of the Amazon case and output-based AI remuneration can easily be drawn. As in the case of private copying, it is hardly possible to calculate the exact damage that the offer and use of generative AI inflicts on an individual journalist. Hence the criterion of substitution effects underlying the proposed remuneration system for AI-generated output need not constitute an insurmountable hurdle to setting up social and cultural programmes with a particular focus on public interest journalism.
Second, the CJEU made it clear in Amazon that a system of indirect remuneration via social and cultural institutions must not be discriminatory. This further requirement could also be seen as an obstacle to the establishment of a system favouring public interest journalism. Stronger support for journalistic work with particular societal value could be regarded as an unfair discrimination against other forms of journalism, such as the rainbow press with glamorous and entertaining celebrity news. This conclusion, however, need not be the last word on the matter. Social and cultural funds can arrive at a distribution scheme that offers stronger support for public interest journalism while basing their sponsoring decisions on objective criteria, such as the extent to which extra financial support is necessary for a certain form of journalism in the era of generative AI systems that displaces human journalists while enhancing the risk of misinformation and disinformation.Footnote 96 Arguably, the need to foster human content production as a countermeasure to displacement effects caused by AI is stronger in the area of public interest journalism than in other fields of journalism, such as the tabloids.
Moreover, it must be considered that even if discrimination in favour of public interest journalism was found, this discrimination could be justified. Given the fundamental importance of public interest journalism – revealing societal problems and providing impulses for the improvement of societal conditions (the second remuneration argument developed in this chapter) – there is a sound justification for lending stronger support to journalists working in this area.Footnote 97 In fact, there is a strong tradition of positive discrimination for the benefit of public interest journalism in European media policy.Footnote 98 Following this tradition, the CJEU’s Amazon decision can be understood not to preclude the partitioning of collected AI remuneration in line with specific social and cultural objectives, such as the aim to ensure the survival of public interest journalism in the era of generative AI.
21.4.4 Foundation in Copyright Law
In addition to these conceptual considerations, the proposed AI levy system raises the legal-doctrinal question of whether copyright law offers a sufficient basis for a remuneration claim relating to AI output. Press and media content produced by a generative AI system need not display protected traces of a human journalistic work.Footnote 99 Compared to the AI training (input) perspective, the situation is different. During the AI training phase, protected human works are used as learning resources for the AI system. Hence there is a direct link between the machine-learning process and the use of protected works. Qualifying copies made for AI training purposes as relevant reproductions, the lawmaker can create a legal basis for a remuneration claim in copyright law.Footnote 100 With regard to AI output, the copyright basis for remuneration is less clear. Instead of reproducing individual expression – protected free, creative choices by a human journalistFootnote 101 – AI output may merely reflect unprotected news of the day, facts, concepts, and styles.Footnote 102
The absence of protected human expression in AI output, however, does not pose an insuperable obstacle. In fact, a copyright concept that, by analogy, can be invoked as a legal-doctrinal basis for the introduction of a lump-sum remuneration system focusing on AI output was developed in the last century. In the discussion on the so-called domaine public payant, Adolf Dietz explained in a landmark 1990 article that, in addition to traditional exploitation and remuneration rights of individual authors, it was consistent and advisable to recognise in copyright law a new right to which a different rightholder – the ‘community of authors’Footnote 103 – was entitled as a collective. Dietz pointed out that this step could be regarded as a corollary of a modern understanding of copyright law ‘as part of a more comprehensive concept of culture law’.Footnote 104 Once this broader role and responsibility of copyright is taken as a starting point, the law is no longer condemned to accept ‘harmful discrepancies’Footnote 105 between substantial profits made by exploiters of public domain works on the one hand, and precarious working and living conditions of current authors on the other.Footnote 106 Instead, copyright can be employed as a legal tool to introduce a remuneration right for the community of living and creating authors as a means of redress:
What we finally propose is simply to introduce another right owner, namely the community of living and creating authors, among several kinds of right owners already existing in copyright law. This community of authors should have the direct right to participate in the income from exploitation of works of dead authors after the individual term of copyright protection has expired.Footnote 107
As this statement indicates, Dietz developed his concept of a new right for the community of authors with a focus on the exploitation of works in the public domain. He placed his proposal in the context of the discussion on the domaine public payant that had gained momentum after the Second World War.Footnote 108 From his perspective, soaring prices and income from the exploitation of public domain works in the field of literature, music, and art should, ‘at least partly and proportionally, also serve the living and creating generation of authors’.Footnote 109 Evidently, the introduction of a new – collective – right to participate in revenue accruing from the exploitation of public domain works begs the question of how this new right of the community of living and creating authors might be exercised in practice. Dietz solves this problem by relying on the well-established system of collective rights management in Europe:
[T]here must be a natural or legal person or body ready to interfere and, in particular, to control the market and claim the participation right, if necessary in a lawsuit. In addition, this body must be able to distribute the incoming money according to statutory purposes and rules, preferably under government supervision. … We should not forget, however, that these kind of bodies already exist, and have done so for decades, in the form of collecting societies.Footnote 110
Before turning to parallels between this remuneration concept and the AI remuneration system discussed here, it is noteworthy that in the second half of the last century, the proposal of a domaine public payant did not remain a mere theoretical option. In Germany, it formed part of the official government proposal for new copyright legislation that was discussed in 1965.Footnote 111 Although the German legislator finally refrained from introducing a new remuneration right for the community of authors in the 1965 Copyright Act,Footnote 112 the fact that the domaine public payant was included in the government proposal shows that the concept and the underlying objective of improving the working and living conditions of authors had broad support in Germany.Footnote 113 An international UNESCO/WIPO survey conducted in 1982 also brought to light several starting points for implementing the domaine public payant in copyright law.Footnote 114 In more recent debates on recalibrating copyright, Rebecca Giblin confirmed the concept’s continued relevance and importance. In a critical assessment of the term of copyright protection, she qualified the domaine public payant as a useful reference point for her proposal to draw a clearer distinction between incentive and reward goals and introduce an opt-in ‘creator-right’ that would give authors access to remuneration systems in return for the registration of their works after an initial term of protection.Footnote 115
The parallels between the domaine public payant and the proposed output-based remuneration system in the area of generative AI are striking. Both concepts concern creations that fall outside the scope of the exploitation rights of copyright holders: literary and artistic works that never enjoyed, or no longer enjoy, copyright protection in the case of the domaine public payant; general news of the day, facts, concepts, and styles in the case of AI output that does not reproduce the individual protected expression of a human journalist. The precursor of the domaine public payant thus shows that potential legal-doctrinal concerns need not thwart the introduction of a remuneration system focusing on AI output. Even if AI output merely reflects unprotected news information, concepts, and styles, it is still possible and consistent to incorporate a lump-sum remuneration right in copyright law as a collective right of the community of journalists: a new right that is subject to mandatory collective rights management. As a new rightholder in copyright law, the community of journalistsFootnote 116 should be entitled to benefit from payments made under this new system. As explained, the collective remuneration right should be administered and enforced by CMOs that distribute collected money through repartitioning schemes and social and cultural funds.
Alternatively, it is possible to forge a link with the input/AI-training dimension and focus on the use of human training material as an indispensable precondition for AI output that resembles human press and media productions. As already pointed out, generative AI systems are capable of mimicking human press and media content only because the works of human authors have been used as training material at some stage.Footnote 117 Even in the case of AI systems trained on synthetic, machine-made literary and artistic material, the system’s capability to mimic human journalistic work can only be explained by the fact that human training resources played a role somewhere in the whole chain of training processes leading to the generative AI system producing output that resembles a human press or media production.
Considering this connection between input and output, it can be argued that remuneration for journalistic AI output must be paid because, directly or indirectly, this output is the result of the use of human works, including journalistic texts, photos, and videos, for AI training. AI input and output are two sides of the same coin: the payment of remuneration at the output level simply constitutes a deliberate choice of the legislator. Instead of placing heavy administrative and financial burdens on AI trainers,Footnote 118 the lawmaker can leave the training process (input dimension) unencumbered and take measures to compensate journalists when final AI products and services are offered in the marketplace and produce press and media content (output dimension). Using AI output as a reference point for remuneration, the legislator can also establish a strong link with the press and media sector: the remuneration for human journalists can be aligned with the extent of AI use in the sector. Even if an AI model has been trained on a broader spectrum of literary and artistic works – not only journalistic creations – the generation of AI output in press and media contexts ensures a clear connection with displacement effects that affect journalists. As discussed in Sections 21.2.2 and 21.2.3 (second and third arguments), this connection justifies the development of a remuneration system in favour of journalists – instead of aiming at compensating all authors whose works may have been employed for AI training purposes.
The detachment of the act triggering the payment obligation – the generation of AI output – from the act that provides the legal basis for the compensation claim – the use of human works for AI training – is not unusual in the area of lump-sum remuneration systems. In the context of private copying, for instance, the CJEU has explicitly recognised that EU Member States are free to impose an obligation to pay compensation for reproductions made by private users on manufacturers and importers of relevant copying equipment, devices, and media. Even though the act with copyright relevance – the private copying – will occur only after the equipment, devices, and media have reached end consumers, the payment obligation can be imposed on manufacturers and importers:
given the practical difficulties in identifying private users and obliging them to compensate rightholders for the harm caused to them, and bearing in mind the fact that the harm which may arise from each private use, considered separately, may be minimal and therefore does not give rise to an obligation for payment …, it is open to the Member States to establish a ‘private copying levy’ for the purposes of financing fair compensation chargeable not to the private persons concerned, but to those who have the digital reproduction equipment, devices and media and who, on that basis, in law or in fact, make that equipment available to private users or who provide copying services for them.Footnote 119
In the light of this existing configuration of levy systems in the area of private copying, it does not seem unusual – and perhaps even less unusual than a legal-doctrinal solution based on the domaine public payant – to simply delay the remuneration payment and take the AI production of press and media content as a reference point for compensating journalists for displacement effects caused by AI training with human works that enables the machine to become a competitor. This alternative legal-doctrinal approach also forges a link with proposals in the AI remuneration debate that seek to establish a lump-sum remuneration system at the AI training stage.Footnote 120 Arguing for the payment of remuneration later – when AI products and services are finally brought to the press and media market – it offers a practical solution that integrates all global remuneration approaches focusing on the use of human creations as training material for generative AI systems. It shows that there is common ground for all lump-sum remuneration proposals that seek to compensate human authors for the use of their works in AI training.
In sum, there are thus two legal-doctrinal avenues that can lead to the introduction of a lump-sum remuneration system focusing on the offer and use of generative AI systems in the press and media sector. On the one hand, the concept of domaine public payant offers a basis for establishing a collective right of the community of journalists – a new right that is subject to mandatory collective rights management – to receive remuneration for AI output that has the potential to replace human journalistic work. On the other hand, the focus can be on the use of human journalistic material as an indispensable precondition and training resource for AI systems capable of producing output that resembles human press and media productions. Following the example of levy systems in the area of private copying, it is possible to uncouple the act triggering the payment obligation from the act that provides the legal basis for the remuneration claim. Hence the legislator is free to delay the remuneration payment and take the production of AI press and media output as a reference point for compensating journalists.
21.5 Conclusion
Generative AI systems are likely to replace human journalistic work and usurp the market for human press and media productions. This development has a broader societal dimension. To enable journalists to fulfil their ‘watchdog’ function, draw attention to societal problems, and provide impulses for necessary changes, it is advisable to introduce remuneration rules that offer financial support for human journalistic work and, in particular, human public interest journalism.
In the EU, the rights reservation option following from Article 4(3) CDSMD – now flanked by AIA provisions – could serve as a basis for a remuneration system focusing on the use of human journalistic content for AI training. In practice, however, this new AI training income is likely to fill predominantly the pockets of large press and media companies that own impressive repertoires of journalistic works. Individual journalists whose works form part of these repertoires will not necessarily receive higher honoraria or an appropriate share of the new source of revenue.
Against this background, it is preferable to follow an alternative path and introduce an output-oriented remuneration system that imposes a general payment obligation on all providers and users of generative AI systems capable of replacing human press and media productions. This output-based approach imposes an obligation to pay remuneration on both AI companies offering generative AI systems and media companies using these systems in the press and media sector. The moment a press publisher or media company employs generative AI technology to replace human journalistic labour, this configuration of the system ensures that the press or media organisation is bound to pay remuneration for the use of AI. Moreover, the output-based approach can be combined with mandatory collective rights management to ensure payment directly to individual journalists in accordance with the repartitioning schemes of CMOs. The remuneration can also be used to finance social and cultural funds that support journalistic work. When distributing AI remuneration, social and cultural institutions are free to prioritise public interest journalism as a countermeasure to AI-generated misinformation and disinformation.
Open access