Soru

Under traditional Anglo-American common law, the doctrine of accession resolves disputes that arise when one party innocently takes raw material belonging to another and, through substantial labor and skill, transforms it into an entirely new and vastly more valuable article. In physical property disputes, if the original material has undergone such a radical physical or economic metamorphosis that its identity is essentially lost, and if the transforming party acted without willful wrongdoing (in good faith), courts typically award title to the finished product to the transformer, requiring them merely to compensate the original owner for the fair market value of the raw constituent elements. The justification for this doctrine rests on preventing economic waste: forcing the forfeiture of the finished good to the original owner would unjustly enrich that owner at the expense of productive labor.

However, legal theorists have noted significant difficulties in transplanting the accession framework to modern digital environments, specifically generative algorithms trained on proprietary datasets. In the digital realm, when a machine learning developer utilizes copyrighted text or imagery without authorization to train a foundational model, the underlying source data is not physically consumed, altered, or deprived of its utility to the original owner. The original text or photograph remains intact and usable by its copyright holder. Furthermore, the generative model does not simply physically integrate the data into an irreversible composite object; rather, it abstracts statistical relationships across billions of inputs to produce novel outputs.

Consequently, traditional accession breaks down on two crucial relational axes in digital contexts. First, because the source material is non-rivalrous—the training process does not dispossess the original creator of their asset—the foundational premise of irreversible physical conversion is absent. Second, compensating the dataset owner solely for the marginal value of a single data point reduces their contribution to negligible fractions of a cent, ignoring the collective synergy that makes large training sets valuable in the first place. Applying traditional accession principles thus risks shielding transformative digital developers from liability while simultaneously depriving creators of meaningful compensation, distorting the equitable balance between rewarding transformative labor and protecting foundational ownership.

Which of the following scenarios is most analogous to the misapplication of the doctrine of accession in the digital context described in the passage?