Legal basis for using copyright objects in machine learning
Anton Tarasiuk · Theory and Practice of Intellectual Property · 2024
In this article, I characterize the legal basis for using copyright objects in machine learning.In this regard, the definition of machine learning from the perspective of copyright law isprovided. In the work, legal relations that occur in the context of machine learning usingcopyright objects are studied.The two key legal grounds for the usage of copyright objects to train models/neural networksare defined: license agreement, fair dealing, and fair use doctrine. Specific terms of thelicense agreement on copyright objects for machine learning are defined.In the analysis of the doctrine of fair use as a legal ground to use copyright objects withoutpermission/license of the author, four actual and ongoing disputes in the USA regarding possibleunlawful usage of copyright objects for machine learning without getting a license orproviding payment to the authors are examined. As a result of this study, the author concludesthat in order to determine whether it is possible to use the fair use doctrine as a legalbasis, there must be a separate examination for each data used, and the analysis should becomplex, including the way the new product based on the respective trained neural networkwill be used and the effect it may have on the market.As a result, key indicators of the pros and cons of the possibilities of applying the fair usedoctrine are defined. Pro indicators: (a) obtaining information from objects about the essence,structure, and ideas embedded in these objects; (b) a transformational and innovative goal; (c)lack of reproduction of the full/partial expression of objects for the end user/the possibility ofcreating modifications of such objects; (d) lack of competition with authors. Con indicators: (a)the use of extremely valuable/highest quality copies of copyright objects in the industry thatare a fundamental asset of a particular company (an organized collection of photographs,etc.), while it was possible to get objects in the public domain for the stated goals; (b) a realimpact on the market of authors due to the creation of competitive end products that causereal harm to them (real/hypothetical); (c) partial/full reproduction of original copyright objectsfor end users of the final product; (d) causing reputational damage to authors.