Artificial intelligence (AI) has clearly taken over cultural sectors, through algorithms and large corpus of data which fuel them and already offers a host of applications developed by research centres, large companies and specialized start-ups.
Copyright: culture.gouv.fr – Alexandra Bensamoun, Professor at University of Rennes 1, Joëlle FARCHY, Professor at University Paris 1 Panthéon-Sorbonne, CSPLA Qualified Specialists, Chairs of the Mission, Paul-François SCHIRA, Auditor to the French Council of State Mission Rapporteur – “Artificial Intelligence and Culture” 27 January 2020
Depending on the objectives pursued, the techniques as well as the types of data used are not the same. Although the question of AI is today closely correlated with that of “data”, this generic term actually covers very heterogeneous realities. For different historical reasons, certain categories such as personal data, public data and “works” data are subject to very precise legal qualification. This is not so for other categories of data (metadata, use data, etc.) which correspond, in culture industries, to professional practices.
Thanks to this data, an increasing number of concrete applications are emerging throufghout the value chain, from the creation stage through to the production stage and on to the consumption stage. As regards consumption, AI and algorithms are massively used to recommend content to Internet users. Although algorithm-based personalized recommendation has been the subject of debate since the first “filter bubble” works appeared, given the risks of consumers being locked into their habits, in actual fact, a host of means of recommendation exists.
Thanks to algorithms, the use of data also revitalizes the ambition to make appropriate investment decisions and to support, or even replace, the usual human intuitions and expertise with supposedly objective analyses of the determinants of the success of a work or an artist. In addition to analysing market trends, one of the promises of artificial intelligence is to compare, based on the use of historical data, contents which have been successful with those which are currently being produced so as to analyse the keys to success, and possibly to anticipate it. In terms of creation, among the various experiments rolled out in culture sectors, not all have the same degree of maturity; some, which merely accompany the human process of creation, are largely present in the audiovisual and publishing sectors; others, which strive to emancipate themselves from it, are found more so in music and the art market.
From a legal point of view, AI intervenes in the artistic creation phase, which raises questions as regards copyright. Whilst the art market receives creations announced as being AI-generated, the question arises as to the qualification of these new productions. Are they intellectual works, and as such protected by copyright? If so, who is the author and the rightholder? A renewed analysis of the conditions of access to protection (creation, originality, author) could enable these cultural works to fall under copyright. But other solutions are also proposed (special right, absence of private protection, etc.). As such, it is important to test positive law and to be ready to intervene if ever a possible need for regulation arises in the future. In any case, the approach should be carried out within an international, or at least, European framework.
Moreover, in a learning process, creative AI works by ingesting works which are deconstructed and analysed so as to identify common characteristics. This process enables the creation of an inference model whose implementation leads to the generation of an algorithmic creation. As such, the creation of the Edmond de Bellamy portrait was made possible thanks to the development of a training base of nearly 15,000 classical portraits from the 14th to the 20th century. Are these upstream acts to be considered as acts of use giving rise to copyright? The introduction of a “data mining” exception dedicated to AI uses, in Article 4 of Directive 2019/790 of 17 April 2019, seems to validate a positive response. However, this new, very broad limitation also provides for the possible exercise of an opt-out by rightholders, which enables a return to reservation. Consequently, other solutions must be put forward to facilitate the use of protected content whilst ensuring the protection of rightholders. In this context, voluntary general licences could ensure a balance.
Finally, insofar as the quantity and quality of the data which can be called on to fuel the development of AI becomes a factor of competitiveness, the specific issues of data sharing and movement for culture initiatives should be questioned. Although data movement and sharing issues as regards use data and metadata are not always directly related to intellectual property concerns, they are nonetheless significant. Beyond the issue of transparency for a fair distribution of revenues for the benefit of rightholders, it is actually all the balances within the sector which are likely to be called into question by access to use data or, on the contrary, by the loss of control over the customer relationship. For metadata, it is basically a question of qualifying content so as to accompany the processes of artificial creation and to renew the means of forecasting and recommendation proposed by cultural industries. A regulation taking the specificities of each sector into account could usefully accompany the access of operators to the large masses of data held by others.
Download the full AI and Culture report here.