The following is the second statement of EWC on AI. It is intended to inform decision makers during the trilogue phase of the (EU) AI Act Proposal. With it, we establish the ART Principle: Authorization, Remuneration, Transparency.
Statement of the European Writers’ Council (EWC) on the trilogue negotiations of the AI Act Proposal and on the urgently needed reform of the text and data mining exception Art. 4 of the CDSM Directive 2019/790 (EU)
Artistic creation needs protection under the ART principle
“AI has gained some remarkable abilities to manipulate and generate language, whether with words, sounds or images. AI has thereby hacked the operating system of our civilisation.”
–– Yuval Noah Harari, Writer, Historian
“If you have specific concerns or feedback regarding AI development or its impact, I encourage you to voice them and engage in constructive discussions. Ethical considerations, transparency, and accountability are critical aspects of AI development, and public input can play a crucial role in shaping the future of AI technologies.”
–– GPT, Large Language Model, Open AI
Writoid: a robot trying to imitate the style and ways of telling a story
that are exclusive of a human writer.
–– EWC campaign #AgainstWritoids
Generative, analytic, and automated Advanced Informatics (AI), so-called Artificial “Intelligence”, will wipe out many jobs in the book sector, be it in writing, editing, proofreading, production, cover art, illustration, translation, selection and crafting of original and translated manuscripts, and human expertise in promotion and reviewing of books.
It will replace human recommendation with automated endorsements in the trade, and limit visibility of European works, of lesser written and spoken languages, and lesser successful books through algorithmic curation of monopoly portals and large retail companies.
It will flatten bibliodiversity as more and more “personalised” output is produced for market saturation and based on reader habit’s data and sentiments mining, rather than relying on freedom, intuition and the creative will of authors. Human voices of audio book narrators will be cloned against a lump-sum, or used without consent, to read out machine generated stories to children.
The spreading, irrational and uncritical enthusiasm for generative informatics such as large-scale language, image, or audio models, lowers the appreciation of human authors and especially writers and their craft, artistic intelligence, and importance for society – and at the same time it is a blind enthusiasm. Blind to the fact that none of these so-called AI systems would be cost-realistic and fully economically affordable if everyone involved were adequately paid: be it the authors and artists whose works have been stolen and used, be it the linguists and labellers who categorise and process language or images and often work for hourly wages of less than 2 euros, be it the lack of security checks that did not take place or were poorly carried out for cost reasons. Nonetheless, big AI companies are already yielding these informatic software, and are into a race to prevail in a trillionaire business.
If this high-risk generative artificial, miss-called “intelligence” in the cultural and book sector is not regulated under ethical, legal, economic, and contextual aspects within an own, broad and EU wide effort with international impact, and a very own legislative text, AI will establish itself as the greatest and most damaging rupture in the history of civilisation.
Until then, it is urgently necessary to repair at least two predetermined breaking points in the current AI Act proposal to ensure that in the future there is any chance at all acting in a thoughtful and fair manner. Accordingly, the EWC state the following:
EWC’s appeal on the AI Act proposal in the Trilogue:
It is the right time to do the right thing.
In the current draft AI Act (as voted by the EP on the 14th of June 2023), two articles relate to aspects of the cultural and creative sectors: Art 28b(4) and 52.3a, to regulate especially generative AI, which the EWC considers as also high-risk AI.
Neither of the two Articles is sufficient to compensate for the past damages by illegal scraping of authors’ works, and neither of them in their current formulation are adequate to establish a sustainable future for the entire sources of all cultural works: the authors and performers.
Art 28b(4): AI regulation on the aspects of transparency cannot do without taking stock of the past: Generative language and image software tools are based on proven theft.
The EWC commented on 18 of July on the version of the European Parliament and with tabled amendments directly to the co-rapporteurs and shadow rapporteurs as well as several Members of the European Parliament involved in the legislative procedure on the AI Act proposal.
The EWC acknowledges the well-intentioned tendency to be achieved with Article 28b(4):
Transparency, especially about datasets used for machine learning. The aggressive protest reaction of the major AI companies to these newly inserted articles shows that especially the sources of data and their acquisition are most probably not exclusively legal.
In a study by the Stanford University[1] it became clear, that none of the 12 largest foundation model providers including Open AI, Microsoft, or Meta, would comply with any of the requirements laid out in Art 28b(4). Research has proven, that existing large language models and automated text generators are also based on illegally sourced copyrighted material from piracy sites[2]. This text mining for the machine learning of competitive products had been in practice since a decade. Copyright works that may have been obtained from other sources were also used for commercial purposes well before the TDM exception (§4, CDSM Directive 2019/790 (EU)) came into force. Therefore, anyone using automated generators of language, visuals or sounds today, in legal terms, may be using stolen intellectual property.
If the current AI Act proposal ignores this, Europe’s decision makers condone this theft.
For the EWC the intention of Art 28b(4) does not reach sufficiently the “effet utile” to establish sustainable and fair practices: we plea for the essential reference to the transparent indication not only of “data” but of copyrighted works, which are in fact the most important basis for all foundation models. Accordingly, the transparency obligations as laid out in Art 28b(4)c shall include copyrighted works, and the proof of legitimate use.
Only, if used works are trackable, the necessary licenses scheme could be established. Any sources for machine learning of generative AI must be legally accounted for – and previous programmes, if based on theft, shall be shut down immediately.
On Art 52.3a: There shall be no Human Right for Machines
Art 52.3 of the AI Act proposal (EP version) contains important labelling requirements, which the EWC supports. Art 52.3a, however, erodes this completely, in the context of, quote, “if it is necessary for the exercise of the right to freedom of expression and the right to freedom of the arts and sciences guaranteed in the Charter of Fundamental Rights of the EU, and subject to appropriate safeguards for the rights and freedoms of third parties.”
This paragraph both allows hiding AI information to the public and takes for granted that machines have freedom of expression and dehumanises a right that is exclusively meant for human beings and their very own creation, actions, and intentions.
To apply the cloak of “freedom of expression” or “freedom of art and science” to machine-generated products is taking these human rights away from humans. While humans are responsible for what they create – including hate comments, disinformation, propaganda – should machines not have to be? At the same time, an exception for the labelling of culture-mimicking output renders Art 28b(4) obsolete: if there is no tracking of use through labelling, there can be no effective mechanisms for opting out or for future licensing negotiations.
Accordingly, we call for no exception to the labelling requirement for automatically generated output, even if it appears only to a minor extent within a work by a human author. The risk assessment for such an exception is also completely lacking. Any labelling exception will have legal, economic, ethical, informational and liability consequences, the damage of which will be irrevocable.
We appeal to the trialogue negotiators to urgently include these two points in the mentioned articles. Now is your window of opportunity to correct two major predetermined breaking points before the deadline – it is exactly the right time to do the right thing.
Artistic and authors’ intelligence need the ART principle: Authorisation, Remuneration, Transparency.
Together with numerous other authors’ federations and legal experts, the EWC is of the opinion,
that the Text and Data Mining Exception(s) (§3, §4, CDSM Directive 2019/790 (EU)) does NOT cover machine learning for especially generative AI and large foundation models.
Acts of reproduction undertaken for text and data mining purposes, can only be applied “in certain special cases that do not conflict with the normal exploitation of the works or other subject matter and do not unreasonably prejudice the legitimate interests of rights holders.”
The use of copyrighted material for the development of generative speech, text, image, or audio works shall accordingly stay subject to a licence reservation.
Meanwhile, the “machine-readable opt-out” legally possible for online published works, the so-called licence or usage reservation instead of a consent requirement, is not applied anywhere, as there is neither a technical nor contractual arrangement for this. In addition, the transposition of these articles in national frameworks, excludes any compensation for this usage.
We are therefore in a fatal situation where past use and today’s continued use of human-created labour and copyrighted works are being mis-used to produce competitive AI outputs that replace the very ones that made the generating of these output possible in the first place. The predatory exploitation goes on, without authorisation, without remuneration, and without transparency. Without A.R.T.
This form of plundering is a shameful act that no democracy can tolerate if it wants to continue to recognise its values of decency and dignity as such.
Steps towards a sustainable future
The following steps must be taken as soon as possible:
- Evaluation of the damage caused must be initiated immediately by the European Commission;
- Clarification of the interpretation of Art 3 and 4 CDSM Directive 2019/790 (EU) related to machine learning;
- Reforming this exception should not be wiped off the table; otherwise, the European Court of Justice will foreseeably decide.
Until clarification and/or reform, the machine-readable opt-out, which currently only exists on paper, must become contractual practice within the sectors, and secondly, harmonised sector standards must be enforced, as up to now, for example, in the book sector, no functioning opt-out notices have been included anywhere. If works do not clearly state whether mining is permitted or not, it is indispensable that AI developers make sure whether legal use exists.
The EWC advocates establishing the ART principle and making it enforceable:
Authorisation, Remuneration, Transparency.
The basis of world copyright law in its form as Authors’ Rights, always places the author at the centre. The moral right is an outstanding achievement of civilisation, as it recognises individual human power of creation and considers the personal freedom of decision on the usage of its work of every person and author to be central.
We are at a point today where societies are seduced by convenience, by the desire to game and play with tools, by supposed cost-efficiency, to want to take a shortcut and outsource the most human activity to non-thinking, non-sentient, non-responsible machines.
This involves reverting to pre-Enlightenment attitudes, when no text had an author but a diffuse, non-human entity as originator. We should be much further ahead by now, and not deny ourselves what makes us most human: art and culture. If we allow writoids to replace writers, instead of protecting them, we will simply lose humanity.
The 220.000 individual writers of the book sector represented by the European Writers‘ Council (EWC) and it’s 49 member organisations, deserve a strong, clear, and based on European values action from all legislative and economic decision makers involved in the cultural sector.
Brussels, 24 July 2023
Miguel Ángel Serrano, President, Spain
Maia Bensimon, Vice President, France
Nicole Pfister Fetz, Secretary-General, Switzerland
Nina George, Commissioner, Germany
Further resources Call for ethical AI regulation Artificial Intelligence sources mentioned: Study – Stanford University[1] about the requirements laid out in Art 28b(4) on AI foundation model providers. https://crfm.STANFORD.edu/eu-ai-act Research on illegally sourced copyrighted material from piracy sites[2] AI_Copyright/the-books-used-to-train