New definitions. Could they apply to all media?
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5561522
The New Art Forgers
The “substantial similarity” between a copyrighted work and an unauthorized derivative has formed the bedrock of copyright infringement jurisprudence since the mid-nineteenth century. Recent technological developments, however, are destabilizing these conceptual foundations. In May, the Copyright Office suggested that the use of copyrighted works to train AI models may constitute infringement even if model outputs are not “substantially similar” to model inputs if they nevertheless “dilute the market” for similar works. One month later, Judge Chhabria of the Northern District of California argued that AI outputs do not have to be “substantially similar” to copyrighted training data in order to be infringing. The plaintiff’s incentives are sufficiently harmed, Judge Chhabria argued, when the market is flooded with “similar enough” AI-generated works.
These developments should be read as early warning signs of a disturbing doctrinal shift from “substantial similarity” to a new and dubious threshold for actionable infringement: “substitutive similarity”, where the substitutability of the defendant’s work, rather than the similarity of protected expression, provides the cause of action. This novel theory of harm, if widely adopted, would impose dangerous restrictions on downstream creativity. Any new work that was “similar enough” to existing works would be treated as potentially infringing, despite the absence of substantially similar expression. This would corrupt what is essentially a question of fact – whether the defendant copied “enough” of the plaintiff’s work to constitute unlawful appropriation – with deontic considerations of the wrongfulness of free-riding.
At the same time, artists are understandably rattled by the speed and scale of AI generation. AI models can produce “new” works in the style of established artists in a matter of seconds, dramatically undercutting the market for their work. AI style mimicry makes it difficult for artists to control their personal brands and for consumers to locate authentic works by their favorite artists. Copyright is responsible for protecting artists’ creative incentives, but its legal tests were not designed to handle the scale of imitation enabled by AI.
This Article offers a way out of this jurisprudential morass. Instead of lowering the burden of proof for infringement, Congress should strengthen the attribution rights of existing creators. Low-protectionists have long advocated for attribution rights as a way of protecting authors’ interests without expanding the scope of their economic entitlements. Proper attribution allows creators to capture the full reputational benefits of their labor without stifling downstream creativity. For example, Congress could enact an AI-specific attribution right that requires the disclosure of copyrighted training data in output metadata. This would mitigate the labor-displacing effects of generative AI by directing consumers to the original creators of a popular style or aesthetic.
Generative AI places copyright jurisprudence at a critical crossroads. Indulging Judge Chhabria’s novel theory of harm would effectively inaugurate a new standard for infringement – “substitutive similarity” – that would stifle not just AI innovation but human creativity more broadly. The stakes for protecting free expression through careful guardianship of longstanding doctrine could not be higher. This Article guides readers through this critical inflection point with new terminology for the jurisprudential lexicon as well as practical proposals for reform.
Interesting idea.
The Upcoming Moral Crisis in Primitive Artificial Intelligence
As we continue to develop artificially intelligent systems, there is an increasingly high chance that we will develop a system that is both conscious and capable of suffering. Furthermore, it is likely that the development of this conscious machine will be entirely unintentional. While this machine will have moral status, identifying it will be extremely difficult, leading to it being treated the same as its inert predecessors. For these reasons I believe that a crisis in ethics is looming. This paper aims to argue that it is possible for a machine to have moral status, that the first such machine will likely be produced unintentionally, and that identifying this machine will involve significant difficulties.
At least they are thinking about it…
New York court system sets rules for AI use by judges, staff
The New York state court system on Friday set out a new policy on the use of artificial intelligence by judges and other court staff, joining at least four other U.S. states that have adopted similar rules in the past year.
, opens new tab
The interim policy, which applies to all judges, justices and nonjudicial employees in the New York Unified Court System, limits the use of generative AI to approved products and mandates AI training.
