Ar reikalingi atlygio mechanizmai už autorių teisių objektų naudijimą dirbtinio intelekto sistemų mokymui?
Puida, Mangirdas |
Bakalauro darbe yra analizuojamas atlygio mechanizmų reikalingumas ir šių mechanizmų galimi variantai. Europos sąjungos ir Lietuvos respublikos teisinėje bazėje nėra numatyta pareiga dirbtinio intelekto (toliau – DI) vystytojams mokėti autoriams atlygį, jei DI naudojasi autoriniais kūriniais mokymosi procese. Be to, įvairūs teisės aktai sukelia neproporcingą jėgų balansą DI vystytojų naudai. Pavyzdžiui, teksto ir duomenų gavybos išimtis buvo sukurta prieš kūrinančiojo DI (toliau – KDI) modelio populiarėjimą, šios išimties pagrindinis tikslas yra leisti mokslo ar komercinėms technologijų įmonėms naudotis visais prieinamais tekstais ar duomenimis, įskaitant ir autorinius kūrinius, su tikslu rinkti įvairius statistinius duomenis DI mokymosi metu. Tačiau išpopuliarėjus KDI modeliams, ši išimtis pradėta naudoti ne tik statistiniams duomenims, bet ir autorių kūrinių stilistikos analizėms. Dėl to, KDI modeliai tapo komerciškai pelningi ir populiarūs, kadangi vartotojams pateikdavo įvairius kūrinius, kurių sukūrimo laikas ir reikalingi ištekliai yra mažesni už tradicinių autorių. Todėl įvairios kolektyvinės autorių organizacijos (toliau – KAO) išstojo su įvairias pareiškimais, kuriuose nurodo, kad KDI kuriamas turinys kelia didelę konkurenciją autoriams bei pranašumą įgauną mokydamasis iš autorinių kūrinių. Taip pat išskiriama, kad DI mažina autorių kūrybiškumą ir tolimesnį augimą bei kenkia tradicinėms autorių vertybėms. Tačiau įvairūs DI technologijų kompanijų atstovai neskuba teigti, kad atlygio mechanizmai yra reikalingi, kadangi staigus apmokestinimas autorių kūrinių, kurie yra naudojami DI mokymosi procese, galėtų sulėtinti technologijos vystymąsi, be to nėra pilnai technologiškai išspręsta, kaip atrasti kiekviena autorių, kurio kūriniai yra naudojami ir kaip atlyginti už panaudotus autorinius kūrinius. Tačiau DI konkurencingumas, pilkoji zona teisinėje bazėje ir autorių išskiriama grėsmė tolimesniam autorinių kūrinių konkurencingumui, rodo, kad atlygio mechanizmų reikalingumas yra pagrįstas. Tarp galimų atlygio mechanizmų išskiriami licencijavimo mechanizmai, kuriems nereikėtų daug pakeitimų, kad galėtų užtikrinti autoriams atlygį. Įvairioms KAO reikėtų pritaikyti jau turimus licencijavimo mechanizmus ir susitarti su technologijų įmonėmis tam, kad atlygis pasiektų atstovaujamus autorius. Siūlomi yra ir alternatyvūs atlygio mechanizmai, kurių technologinis pritaikymas yra sudėtingesnis, tačiau jie galėtų tiksliai ir skaidriai skirstyti autoriams atlygį.
This bachelor thesis aims to investigate whether reward mechanisms are needed if Artificial Intelligence (later -AI) uses copyrighted works in AI learning. And if the need exists, what reward mechanisms would be most appropriate. The problem stems from the inadequacy of the European Union and Lithuanian legal frameworks for the learning process of AI. In particular, the Text and Data Mining Exemption (later - TDM) is not tailored to the generative AI (later - GAI). The TDGI exception was designed for statistical data collection, but with the rise of GAI, the use of TDM has moved beyond the collection of statistical information on various data. But also for the collection of information on the technical solutions, stylistics or historical importance of the creations used by the authors. As a result, the GAI, for the user, can create content according to different styles and preferences. And it does so faster, cheaper and with greater financial return. Moreover, the TDM does not provide for what the developer of the AI should do with the data already used if the author opts-out. It also raises questions as to whether all the data used is collected in a transparent manner. Although existing laws require AI developers to provide the data used, not all of them do so. This makes further enforcement of copyright more difficult. Also, the existing legislation does not provide for how to ensure the continued integrity of authors' works if they choose to opt out of the TDM exception by exercising their right. These ideas have also been expressed by various collective authors organizations (later - CAO), who have argued that the TDM does not ensure the continued integrity of copyright data and does not address the issue of the enforcement of authors economic rights. Furthermore, the CAO point out that the GAI threatens the continued existence of authors. In a commercial environment, GAI is much more commercially attractive and has a wider range of applications. This discourages current and future authors from engaging in creative activities. However, various technology companies disagree with this position of authors. They point out that a sudden reward mechanism could slow down the development of the technology and reduce European Union competition in AI technology. It would also create a higher threshold for new AI companies to cross. AI companies also distinguish that they make full use of the legislation when developing the technology and therefore do not violate existing legal norms. Taking both sides of the argument into account, it is clear that the legal norms do not fully regulate the AI learning process and leave a legal grey-area in which AI developers operate. However, in order to ensure the authors future creative perspective, it is important to define what kind of remuneration mechanism need to be provided to the author, in order to ensure a level playing field in the commercial stage. To start with, this problem can be addressed through various types of guidance and, at the same time, the preparation of regulatory standards. The use of licensing mechanisms, which are regulated by law and widely used, would also help to address the situation. Two licensing mechanisms are available. The first reward mechanism supports 5 the continued existence of the TDM and encourages AI developers and authors to agree on contractual terms designed to regulate the AI learning process, but also to compensate for the use of copyright works. The legacy of the TDM would encourage further scientific progress and allow various private initiatives to benefit from the TDM. Additional contractual clauses would include a differentiation of AI according to where it is used and the commercial benefits it generates. This would allow more precise remuneration of authors. The second remuneration mechanism does not support the TDM, as not all developers of AI are honest, and it is therefore desirable to give authors every opportunity to be remunerated for the use of their creations. Another way to ensure that authors are rewarded is through alternative mechanisms, such as a royalty system, which could ensure payments to authors. AI developers could use methods such as determining how much copyrighted information has been used in AI training or how long the data has been used. It is also proposed to charge for copyrighted data used only for content generation. This proposal would be more difficult to implement as it would require a large amount of data which is difficult to collect, but it could ensure the most accurate remuneration and encourage authors to allow their works to be used for AI training. Other proposals even include the use of different decentralized networks and currencies as a form of remuneration. However, such mechanisms are technically complex and the decentralized nature of the system makes their implementation highly uncertain. In summary, it can be argued that a reward mechanism is needed in the context of learning for the AI, as the commercial attractiveness of GAI pushes authors out of the commercial struggle, which does not encourage human creativity and further development. Various remuneration mechanisms can be used to level the commercial playing field, the most appropriate one being licensing, as it already has a clear legal framework and does not require a large financial outlay to enable it to reward authors for the use of their copyright works in AI learning processes.