To the Members of the AIDA Special Committee on Artificial Intelligence in a Digital Age
Brussels, 08 February 2022
Own-Initiative report – 2020/2266(INI) »Artificial Intelligence in a Digital Age« (PE680.928 – FdR 1224166; rapporteur: Axel VOSS)
European Writers’ Council (EWC) initiative submission on the economic, ethical, legal, artistic, and technical implications of Artificial Intelligence in the book, text, and media sectors with regard to 2020/2266(INI) draft report related to paragraphs 14, 57, 68, 69, 105, 106, 116, 122, 149, 159.
Today we address you in your position as AIDA member, with regard to three necessary comments on the above report, which we would like to make available to you as well as to the AIDA committee members before the meeting on 10 February and discussion of amendments.
The European Writers’ Council represents the interests of 160,000 authors in the book and text sector from 46 organisations and 30 countries who write and publish in 31 languages and in all genres. We have been closely following developments on the legal, technical, economic, and artistic levels of AI-driven software like MT, NLP, and NLG since 2018.
The European book sector will be shaped by the increased use of so-called “weak AI” in the future with three partly disruptive changes: Automated machine translation, text and data mining, and the emergence of competitive language and text software. This is relevant in the context of the debate to grant AI software manufactured products (texts, audios, translations, etc.) an own property right, as well as in view of economically already glaring disruptions.
We are aware that the report 2020/2266(INI) (Reference document: www.europarl. europa.eu) largely omits the cultural sector. It is therefore all the more important that no damage is done inadvertently to the cultural sectors and their sources. We would like to raise awareness of the following three breaking points:
LEGAL ASPECTS, related to 116, 159, 57
We expressly support amendment No 908 on paragraph 116, which you also endorsed (reference document: www.europarl.europa.eu, page 55/173). Rational: AI that is used worldwide is exclusively “weak AI”. Weak AI can only focus on one task; neither can a text generator read or check itself, nor does it, unlike a human translator, assume the liability regime for it’s produced result.
Proprietary rights also mean proprietary duties: Intellectual property rights, especially in text works, always implies an authors’ obligation and responsibility for its contents; this implies fidelity to the facts, respect for the personal and moral rights of others as well as the observance of human rights charters. An AI is not able to cover these duties.
Respecting the power of creation – and the level of creation: weak AI in the text sectors only simulates what we misinterpret as human intelligence, decision-making ability, knowledge, empathy or just: Consciousness. An AI in the textual domain (NLP, NLG) only ever makes copies of what already exists. It does not add a single insight or observation of its own, and is never self-creating or intrinsically motivated, unlike the human mind. The principle of a willful creation and the recognition of the intellectual property of a human personality in its individuality is aptly stated in the Berne Convention. A text-repeating AI is in no way artistically creative.
Outflow of European investment and innovation: The consequences of a copyright, however weighted, for products from AI-supported text software are manifold and fatal for Europe: for example, equipment levies, license fees and royalties for translations, library royalties of the member states, or remuneration from performing rights on the web vis-à-vis platforms and providers would no longer be distributed to European cultural actors, but to mainly non-European companies, who would accordingly claim their shares from joint and individual remuneration. Accordingly, under no circumstances should it be considered to assign comparable IP rights to the results of automated AI processes as to human authors.
With regard to paragraph 57, we would like to note that continental European legislation on intellectual property rights is clear and legally secure. The assertion that companies would find it “easier” in the USA to obtain a copyright for their products lacks any serious basis in legal comparisons; uses of texts, data or works in the USA for AI applications under “fair use” are in principle triggers for legal disputes. Paragraph 57 should be deleted without replacement, as it does not reflect any solid knowledge, and even more so fails to recognise the achievements of the balance achieved between authors’ rights and the rights of the general public.
ARTISTIC AND TECHNICAL ASPECTS, LIABILITY AND RISKS, related to 14, 105, 106
(14) This paragraph underestimates the risks of automatic translation: neither can translation applications read or check themselves, nor do they, unlike a human translator, assume the liability and responsibility regime for their translations.
Example 1: Member State governments already use automatic translators like DeeplL for legislative texts. There is a high risk here that, due to software shortcomings, different interpretations of a law or treaty are in circulation and lead to disastrous consequences.
Example II: Google Translate, which is used by half a billion people every day, doesn’t notice that it produces sexist stereotypes. The University of Porto Alegre in Brazil has put simple sentences with job titles in gender-neutral languages such as Hungarian, Turkish, Japanese, and Chinese through machine translation into English. These languages do not use gender-specific personal pronouns such as “she” or “he” – to see what the web translator makes of them and which professions and characteristics it assigns to whom. Conclusion: engineers, doctors or teachers are men, hairdressers or nurses are women.
Example 3: Correspondingly yesterday’s clichés and bias lead to AI reproducing stereotypes in the text area and exhibiting racist tendencies, or, when it “learns” to translate, as the Microsoft bot “Tay” does on Twitter, in forums or Facebook comments, it formulates in a fascist, anti-Semitic and misogynistic way.
Example 4: Translation programs learn fast – but their word-for-word translation takes no account of allusions, metaphors, irony, the nuances of synonyms or sentence rhythm. The German machine translation “DeepL” can translate shorter texts in one of big nine world languages bases, but not without an editorial and linguistic like narrative intensive revision, around insinuations, bon mots or slang adequately into the other culture area transfer.
For minority languages, there will most likely not be any neural network machine translation, because the transfer data is too small to learn from them. In the long run this means, the European sector is facing a loss of minority languages.
Accordingly, we ask that automatic translation in paragraph 14 not be lightly misinterpreted as “risk-free”. Sector-specific provisions (105, 106) should also be discussed with the sectors first, before draft legislation emerges.
ECONOMIC ASPECTS, related to 68, 69, 122, 149
Paragraph 122 highlights the availability of “data”. Here, not only habits, opinions, shopping behavior or health conditions have to be thought of, but ALWAYS also works in media, books, audio works, visual cultural works. Accordingly, the “availability of data” must be carefully formulated, considering economic, legal, and democratic aspects.
Background: Weak text AI is only as good as the templates it is “trained” from. In order to make the quality of software and results “better”, and to become as diverse as society is, one thing must be clear: they need professional, good, zeitgeisty texts from professionals like book authors, or journalistically brilliant works. This is called Text and Data Mining (TDM), which, besides being an acceptable help and a boon to science and research, is now highly relevant to commercial enterprises. Around the world, Oracle, Alibaba, Google, Microsoft, OpenAI, Nvidia and Amazon are working on text generators and machine translation. Since 2014, also various alliances worldwide have been lobbying for mandated exceptions and limitations in text and data mining on European soil:
a) Software manufacturers such as the BSA Software Alliance, the global political lobbying initiative of companies such as Microsoft, IBM, Oracle, but also Siemens or Adobe, https://www.bsa.org/de/ueber-bsa/aufsichtsrat)
b) Start-up-Alliance, whose permanent advisory board includes the biggest communications companies and intermediaries in the world, such as Apple, Facebook, Google, or Oath, https://alliedforstartups.org/about-us/
c) Institutes like the Centre for Democracy & Technology, whose board includes Mozilla, Microsoft, and global “consultancies” like Brunswick. https://cdt.org/about/board/
TDM takes place on a large scale and leads to lucrative business models – outside Europe:
TDM of European works becomes free gold mining for AI products of international companies, there is a clear need for a new ethics and legal fine graining that does not put the creators of the data and texts in a worse legal position than the users and skimmers of their work.
In fact, however, on 7 June 2021, the Directive (EU) 2019/790 opened the door to those non-EU intermediaries who use book and press works to create competitive AI application products that imitate and partially replace the works of human authors. The only way out is to integrate a “machine-readable opt-out” into your work. But: There is no “rights reservation protocol” installed so far. TDM of European and copyright protected works from international tech companies leads to competitive AI products that put European authors and cultural workers at a disadvantage, resulting in an outflow of investment and innovation of European cultural economies. In many EU countries, during the implementation process of the Directive (EU) 2019/790 the TDM exception is not even remunerated; which leads to the fact, that non-European companies (See above) are profiting of investments to develop AI to compete exactly with the sources their software is trained with. This also includes companies declared as start-ups inside and outside the EU: these should also respect legal principles, be able to pay royalties and not be given special status.
Further Issues in Relation to Data and AI
The right of informational self-determination is a part of European data protection laws. But it ends at international borders. The AIDA Committee may consider advocating for the establishment of this right also towards non-European companies, and to add an additional item to that effect in the draft report.
We stay at your disposal as a discussion partner.
With best regards:
Nina George, EWC President
- https://www.wipo.int/export/sites/www/about-ip/en/artificial_intelligence/call_for_comments/pdf/org_european_writers_council.pdf ↑
- https://republicadelasletras.acescritores.com/2021/06/16/la-inteligencia-artificial-es-bastante-idiota-o-la-ia-en-el-sector-del-libro-un-articulo-de-nina-george/ ↑
- https://medium.com/sciforce/neural-machine-translation-1381b25c9574 ↑