Jennifer Rothman has posted to SSRN Reframing Deepfakes. The abstract provides:
The circulation of deceptive fakes of real people appearing to say and do things that they never did has been made ever easier and more convincing by improved and still improving technology, including (but not limited to) uses of generative artificial intelligence (“AI”). In this essay, adapted from a lecture given at Columbia Law School, I consider what we mean when we talk about deepfakes and provide a better understanding of the potential harms that flow from them. I then develop a taxonomy of deepfakes. To the extent legislators, journalists, and scholars have been distinguishing deepfakes from one another it has primarily been on the basis of the context in which the fakes appear—for example, to distinguish among deepfakes that appear in the context of political campaigns or that depict politicians, those that show private body parts or are otherwise pornographic, and those that impersonate well-known performers. These contextual distinctions have obscured deeper thinking about whether the deepfakes across these contexts are (or should be) different from one another from a jurisprudential perspective.
This essay provides a more nuanced parsing of deepfakes—something that is essential to distinguish between the problems that are appropriate for legal redress versus those that are more appropriate for collective bargaining or market-based solutions. In some instances, deepfakes may simply need to be tolerated or even celebrated, while in others the law should step in. I divide deepfakes (of humans) into four categories: unauthorized; authorized; deceptively authorized; and fictional. As part of this analysis, I identify the key considerations for regulating deepfakes, which are whether they are authorized by the people depicted and whether the fakes deceive the public into thinking they are authentic recordings. Unfortunately, too much of the recently proposed and enacted legislation overlooks these focal points by legitimizing and incentivizing deceptively-authorized deepfakes and by ignoring the problems of authorized deepfakes that deceive the public.