New Technologies in Old Barrels

The eruption of Artificial Intelligence technologies has created new problems for lawmakers. One of the most controversial functions of AI is the way the technologies train themselves. Currently, AI training processes hundreds of millions of datapoints publicly available on the internet. These include sites likes Wikipedia and Project Gutemberg, as well as articles, blogs, forums, and images. One consequence is that the technology may avail itself to copyrighted material during training. Thus, ownership of the new content may be called into question.

To resolve this dispute, it is helpful to look at the doctrine of good faith improvers from the 1871 case Wetherbee v. Green. In that case, a cooper entered another man’s land under the belief he had a license to. There, he felled some timber that became hoops for the barrels he manufactured. The landowner then sought to recover the barrels. The court held that “the owner of the original materials is precluded, by the civil law and common law alike, from following and reclaiming the property after it has under­gone a transformation which converts it into an article substantially different." The court reasoned that the cooper had acted in good faith because he honestly believed he had a license. Additionally, when trees became more valuable barrels through the cooper’s capital expenditure, it would have been inequitable to restore them to the original landowner. Accordingly, the cooper retained his barrels.

The doctrine of good faith improvers may provide insights into resolving the ownership dispute of original AI content between the underlying copyright holders and the AI platforms.

First, improvers are protected by the courts when acting in good faith. Here, the fair use doctrine provides for use of copyrighted material for learning purposes. Its applicability to AI training has been accepted by lawyers and scholars. AI modules reasonably rely on fair use like the cooper relied on the license he believed he had. Furthermore, AI platforms operate in good faith by allowing other creators to opt out from their content being used for training.

Second, the improvement must work upon the source material a substantial transformation. AI companies have spent years and billions of dollars in research and development similarly to how to cooper did on manufacturing. Subsequently, AI provides answers that are wholly new. Like the original trees that became barrels, the original data components lose their individual identity as they inform new content which users access. This new content, like the barrels, is disproportionately more valuable than any component webpage that might have originally informed it. For example, AI solutions are proving to be lifesaving from agriculture to healthcare. Thus, much as the landowner’s trees joined new markets, copyrighted material may be breaking ground in industries the authors could have scarcely imagined.

The rule that good faith improvers are entitled to their new products is the right rule to settle the question of AI content ownership. While it is uncertain how specific copyrighted material feeds into corresponding AI solutions, it is well established that preventing disruptors to protect established interests harms the common good. The effort to determine this alone, recently proposed by California Assembly in A. B. 412, would be a death knell to the industry. To allow the copyright holders an ownership stake on work that might be informed by theirs would defy the core principle that ideas belong to all, and that it is the creativity in presenting it in novel ways from where ownership rights arise.

Pablo Urioste is a law student at the American University Washington College of Law and a junior staffer at the Business Law Review. In his free time, you’ll likely find him listening to The Rest is History.

Image: Anonymous artist, Böttcher 1880.