Sui generis solves for publishers best – Europe proves

The sui generis database right is also viewed as akin to publisher’s right. Publishers are neither authors nor related rights owners, but they organize the supply of information via their creative role in designing the flow of information and their financial investments. In 2000, Clark noted that the sui generis right in the Database Directive is as near as publishers were ever likely to get to a publisher’s right.39 In 2019, the neighboring right for press publishers was introduced in the Digital Single Market Directive.40 Hugenholtz observes that the new neighboring right seems to have its justification in the investment in content aggregation, quite similar to the sui generis database right.41 However, same as the previous neighboring rights, the publisher’s right lacks a threshold test, which, on the one hand, will lead to overprotection and legal uncertainty and, on the other hand, again demonstrates that the EU database right appears to be more justified.

Because of this similarity between sui generis right and neighboring rights, the database sui generis right has been transposed as a special neighboring right in Germany, France, Spain, Portugal, and Italy. In Belgium and Netherlands, the right has been transposed by a special act, illustrating its sui generis status.4

The Database Directive provides for standard exceptions to the sui generis right, such as extraction for private purposes, teaching and scientific research, public security, administrative or judicial procedures (Article 9). The term of protection is fifteen years from the date of completion of the database, starting from the first of January of the year following creation of the database. In case of a substantial change that is considered a substantial new investment, a new term of protection will start for the resulting database (Article 10). This means that a regularly updated database is granted permanent protection.43

Unlike copyright protection that, under the Berne Convention, is also provided to nationals and residents of third countries, the sui generis database right is granted only to nationals of Member States and companies having their registered office and principal place of business within the Community (Article 11). However, reciprocity agreements may be concluded.44 Since the enactment of the Database Directive, the European Commission has published two evaluation reports in 2005 and 2018.45 In the first evaluation report, the Commission observed that the economic impact of the database sui generis right is unproven and the empirical evidence casts doubts on its necessity for a thriving database industry. However, it was also noted that most respondents in the online survey had found that the sui generis database right “has brought about legal certainty, reduced the costs associated with the protection of databases, created more business opportunities, and facilitated the marketing of databases” (Section 1.4). Therefore, most European publishers and database producers preferred to retain the sui generis protection. Although its impact in stimulating the production of databases is limited, the Database Directive does not impose significant administrative or regulatory burdens. Thus, the Commission concludes that one of the policy options is leaving the Database Directive unchanged (Section 1.5).

The sui generis system PICs out of generative AI that is solely fact based as well as generative AI tools used by legitimate news organizations but still eliminates generative AI analysis – this turns the journalism industry internal link

Díaz-Noc 20 [Javier, full professor of Communication at the Universitat Pompeu Fabra Barcelona. Main researcher of the Newsnet project (News, Networks, and Users in the Hybrid Media System. Transformation of the Media Industry and the News in the Post-Industrial Era, 2019-2022). Member of the DigiDoc research group and its coordinator (2020-2022). Artificial Intelligence Systems-Aided News and Copyright: Assessing Legal Implications for Journalism Practices, https://mdpi-res.com/futureinternet/futureinternet-12-00085/article_deploy/futureinternet-12-00085-v2.pdf?version=1590572036, ALB]

2. What is Automated Journalism? A Typology of AIS-Aided Created News The already short development of AIS-aided, or automated news, appears, generally speaking, as a practice of presenting or producing news items out of previously gathered and structured data, normally using templates and applying some more or less complex algorithms. We follow in this respect the definition provided by A. Graefe in 2016: “It is the process of using software or algorithms to automatically generate news stories without human intervention—after the initial programming of the algorithm, of course. Thus, once the algorithm is developed, it allows for automating each step of the news production process, from the collection and analysis of data, to the actual creation and publication of news” [7] (p. 9). The history of AIS-aided news production is scarcely ten years old. The (initially British, now global) newspaper The Guardian started with software in 2010 to produce some news on sport statistics and graphics, and in 2014 experimented in a similar way with Guarbot, a tool to produce news on financial information. The real effective experimentation can be traced back to 2014, when a journalist then hired by The Los Angeles Times, Ken Schwencke, designed an algorithm to produce some news on a low-intensity earthquake that happened that year, based on data from the United States Geological Survey service. One year later, the main French daily newspaper, Le Monde, used another algorithm designed by Data2Content and Syllabs companies to produce some news on election results, using numeric data as well. One year later, in 2015, a Chinese tool, Dreamwriter, was created by a videogames company, Tencent, to produce news on consumer prices that was 916 words long in just one minute, with apparently no mistakes. From then onwards, many other ones have appeared: Heliograf (used by The Washington Post since 2016), Quill, Soccerbot, Wordsmith by Automate Insights, used since 2014 by the Associated Press agency, Recount, StatsMonkey, Media Brain, Kognetics and RADAR are some of them [2]. RADAR is a rather interesting case. It was created in 2017 by a news agency, the Press Association, which in three months produced, using this software, more than 50,000 items. The software was developed by Urbs Media and financed with EUR 150 M by the Google Digital News Initiative Innovation Fund. It uses open access datasets on topics such as transport, education, health, crime and education, and it is able to produce several versions of every item adapted to the necessities of their clients. Behind the RADAR working flow, there was a team of six journalists who identified interesting topics and conducted the production of automated news. Some topics and sections seem more appropriate to use AIS-aided newswriting. Finances, election results and especially sports coverage have appeared as the most widespread topics in which algorithms are used to produce news. Media have sometimes used chatbots to communicate with users, and these tools are able to write their own sentences, based on patterns and on topics and terms recognition. In 2017, the Innovation Lab of the Spanish native-digital journal El Confidencial created a software named AnaFut which creates football chronicles of the lower categories. Sport coverage also combines documentation and bots [12], in the case of BeSoccer. Most of those systems use as a primary source data provided by official institutions: the Spanish public broadcast service, Radio Televisión Española, decided in 2020 to experiment with data extracted from the Spanish Football Federation to offer short news on results of the lower leagues, “interpreting them and presenting a text in natural language, related to the selected event with no personal intervention”, using HTML format and as a mere news, and not penalizing the SEO positioning of the source itself. Similar systems were used by The Washington Post to cover the Olympic Games in 2016. Actually, according to Beckett [13], artificial intelligence systems can help journalists and the media in three phases: gathering, production and distribution. This can lead, however, to a wide variety of results and, which is our point from a legal perspective, intervention by human journalists. These results can be:

1) Mash-up news items can be produced, aggregating several previously published works. This results, using legal terminology, in derivative works which are obliged to mention the works and authors on which the new items are based. As an example, this is the case of Adrian Holovaty’s mashed up infographics produced for the website Chicagocrime.org, [14,15]. Producing intelligent infographics is also the method developed by Intelygenz and Prodigioso Volcán in Spain from 2018 onwards (see http://losdelvolcan.com/grafia/web/): while the journalist creates his or her item, a machine-learning comprehension software scans the words and, relating them all, it creates some graphics with no human intervention—so it can be defined, in legal terms, as a derivative work—to complement them, so as to produce, once again using legal terms, a collaborative work.

2) Automatization of processes can help journalists to provide more context, data and even links (internal or external) to their items. It is common practice to search in the documentation service of the newsroom to find related news to be used and linked. Contextualization of news seems to be exclusive of human journalists, although interfaces and search engines can help in extracting that which is needed from massive databases [16] (p. 179).

3) Another relevant use of AIS in news production is the verification of information, automatically tracing sources and facts. One example is Truthmeter, “a tool that automatically scores the journalistic credibility of social media contributors in order to inform overall credibility assessments. The Truthmeter computes credibility scores based on data made available through the Twitter API” [17].

4) Content curation is another purpose in which AIS can help journalists in their search for scoops on relevant topics. This is one of the ways used by RADAR (‘Reporters And Data And Robots’), a software system used by the British Press Association, which combines humans and machines to create localized stories at scale on topics such as “crime figures, hospital waiting times and pupil absences from school” [13] (p. 25).

5) An interesting application of AIS is the adaptation or customization of messages to different users, producing several versions—thus, and from a legal point of view, derivative works; any one of them is protectable under copyright law. The Swedish newspaper Svenska Dagbladet, for instance, designed a tool to create different homepages for users, measuring the number of clicks, time or permanence or preferences introduced by the user [18]. Out of our focus, but to be considered for its commercial interest, AIS can help with subscriptions, which is in 2020—until the coronavirus crisis, at least—a common movement in media industry, following models like that of the New York Times’ paywalls. This is a called a “dynamic paywall”, in which companies like Deep Bi are working.

6) Most usually, artificial intelligence systems-aided or automated news production is based on database exploitation and automatization of raw data using patterns, which results in what some authors have named database journalism [19] (p. 5). The results of those practices can be considered protectable by copyright laws as sui generis rights. Patterns are provided by humans; only when the system is able to learn, improve and create new patterns may a non-human authorship appear.

By now, then, artificial intelligence systems-aided journalism is reduced to factual content coverage, while in principle only humans are able to elaborate more complex and contextualized pieces. Factual content coverage using AIS is, however, appealing for media companies, because it provides “a cost-effective way to create high-quality factual content that does well in SEO terms” [20].

The key of all those systems, and the ones that could be improved in the next future, is whether they need post processing by humans or not. Some of them, like Monok or RADAR, do not seem to need it to produce simple news items, with no great context. Artificial intelligence systems are not able to generate text of complex or non-predictable nature, a hypothesis placed by Ufarte and Manfredi [21]; or, as Belz says, “with a lot of unpredictability in the output” [20].