TFL Vsebine / Pravni letopis

Za zakonito grajenje velikih jezikovnih modelov, ki temelji na obdelavi velike količine besedil in podatkov, je ključna pravno varna in učinkovita avtorskopravna ureditev. V slovenski avtorskopravni zakonodaji imamo po novem dve izjemi, ki se nanašata na podatkovno analizo oziroma besedilno in podatkovno rudarjenje, ki sta pravna podlaga tudi za grajenje generativne umetne inteligence, kamor uvrščamo tudi grajenje velikih jezikovnih modelov. Izjemi sta bili v slovensko avtorskopravno zakonodajo vneseni z zadnjo novelo Zakona o avtorski in sorodnih pravicah (ZASP-I) oktobra 2022, s katero je Slovenija prenesla določbe Direktive (EU) 2019/790 o avtorski in sorodnih pravicah na enotnem digitalnem trgu v nacionalni pravni red. V prispevku so predstavljene rešitve, kako so se v drugih jurisdikcijah po svetu soočili z zagotavljanjem avtorskopravnih podlag za grajenje umetne inteligence, vključno z ureditvijo v EU, osrednji del prispevka pa se osredotoča na razlago slovenske izjeme za besedilno in podatkovno rudarjenje za namene znanstvenega raziskovanja. Ta izjema je ključna pravna podlaga za grajenje odprto dostopnih velikih jezikovnih modelov za slovenski jezik. V zaključku so poudarjene dileme, s katerimi se trenutno sooča EU, ki sicer namenja ogromne investicije za grajenje umetne inteligence v javnem interesu (Public AI), kakršnega zasleduje tudi izgradnja odprto dostopnih velikih jezikovnih modelov za slovenski jezik, in nakazane možne smeri razvoja.

Legal certainty and the efficiency of copyright regulation are key to building large language models that depend on processing large amounts of text and data. Slovenian copyright legislation has two new exceptions related to data analysis or text and data mining, which also constitute the legal basis for building generative artificial intelligence, including large language models. The exceptions were introduced into Slovenian copyright legislation with the latest amendment to the Copyright and Related Rights Act (ZASP-I) in October 2022, with which Slovenia implemented the provisions of Directive (EU) 2019/790 on copyright and related rights in the Digital Single Market into its national legal order. The article presents solutions for how other jurisdictions around the world have dealt with the issue of providing copyright legal basis for building large language models. The main part of the article focuses on the interpretation of the Slovenian exception for text and data mining for the purposes of scientific research, which is the key legal basis for building an open-source large language model for the Slovenian language. The conclusion sheds light on the dilemmas currently facing the EU, which is otherwise making huge investments in building artificial intelligence in the public interest (Public AI), such as the construction of an open-source large language model for the Slovenian language and indicates possible directions for development.

Copyright Foundations for Building Large Language Models in the Slovenian Legal Framework