Report from the BBC
In Brief – A federal judge in New York has allowed a lawsuit brought by two voice actors to move forward against LOVO, a company that describes itself as an AI-powered platform that converts written text into human-like speech. The plaintiffs allege that LOVO paid them a combined total of $2000 to make recordings for what the company said was internal research purposes and would not be turned into marketable products, and then illegally used their voice recordings to create text-to-speech voice personas. District Judge Arun Subramanian largely rejected the company’s motion to dismiss the complaint, allowing claims to proceed regarding breach of contract and deceptive business practices under New York state law, as well as the section of the New York Civil Rights Law that protects name, image and likeness. In his ruling, Subramanian did dismiss the claim that the company violated federal copyright law when they used the actors’ voice recordings to train its AI model, the issue that he says carries “weighty consequences” for the “burgeoning AI industry, other holders and users of intellectual property, and ordinary citizens who may fear the loss of dominion over their own identities.” However, he attributed his initial dismissal to plaintiffs not explaining clearly enough how the use of their recordings in training violated copyright law and gave them leave to amend their complaint on that issue.
Context – Deepfakes have long been a top AI policy concern, often focusing on nonconsensual pornography and disinformation. We recently saw a deepfakes crossover with copyright when the Danish government said it was planning to amend copyright law to give everyone copyright over their image and voice so that they could use IP “notice and takedown” regimes to force digital platforms to take down AI replicas. Federal Judge Vince Chhabria opined in a recent ruling on copyright and AI training about harm from “indirect market substitution”, meaning every kind of IP rights holder is harmed by any AI system that can create the same type of content, rather than just actual copies. Think about that in the context of voices.