Browsing by Subject "Copyright infringement"

Training AI algorithms can result in copyright challenges. Artificial intelligence and copyright law come together when copyrighted data is used to teach machines to learn, think and act like humans. Downloading and storing copyrighted data for machine learning models may violate copyright law and impose unreasonable liabilities on AI providers. ML models may utilise unlicensed datasets and create digital versions of these copyrighted materials. During the training phase, copyrighted content may be inadvertently or intentionally reproduced, especially when the same data sets are replicated several times during the learning journey. The main analysis will focus on the data training situations of such technologies in terms of copyright infringement; thesis aims to contribute to the ongoing discourse on the use of copyrighted material as "raw data" and "study material" in AI-generated works by examining the complexities of the current legal framework in the European Union (“EU”) and the possibility of copyright infringement. The research also explores the corresponding situation within the United States' legislative framework, analysing relevant case law. The proposed EU AI Act does not specifically address the IP rights of third parties. Certain provisions related to governance, transparency, and technical documentation could be argued to pertain to these rights, but they do not seem to be designed for the benefit of third-party rights holders. Rights holders apparently do not have the right to initiate lawsuits under the proposed EU AI Act for any of these broader obligations. Instead, these are the responsibilities of the "service provider," including the duty to assist relevant competent authorities. However, there are potential arguments related to implicit obligations to mitigate the risk of copyright infringements. Articles 3 and 4 of the “EU copyright directive” further define the scope of the EU's harmonised text and data mining exemption, including the effects of waiving exemptions that are not affected by the proposed EU AI Act. The TDM exception is seen as the starting point for a review of EU copyright law on machine learning. The Directive exempted from copyright infringement mechanical temporary copies, which are necessary for the technological process in question and have no independent creative value, provided that access to the copyrighted content is lawful. In the U.S., the "fair use" doctrine might justify data mining. However, there is still some ambiguity in the law regarding the instances and procedures in which AI can be trained using the works of others.