Fair Learning

By Mark A. Lemley, Bryan Casey. 2020

Introduction The challenge handed to the musician was peculiar and daunting: Take a five-second sample of a randomly selected song and, with just a moment’s notice, transform it into a full-length piece composed in the style of a completely different artist.1 On this occasion, the musician rose to the challenge with such aplomb that it took the internet by storm, earning praise and recognition from media outlets across the globe.2 A Mozart concerto played in the style of Katy Perry?3 No problem. Lady Gaga’s Poker Face in the style of a recent Bollywood hit? Why not?4 Seemingly no pairing of styles, no matter how clashing, proved too difficult.5 The artist capable of such a tour de force? MuseNet—the type of unique, futuristic-sounding name common of many contemporary artists.6 But, on closer inspection, MuseNet’s name wasn’t a reference to the internet (or even

Subfields: AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In AI Ethics, Fairness & Society

Citations in the corpus

SILO Language Models: Isolating Legal Risk In a Nonparametric Datastore (2023, background): “…claims that using parametric language models for generative use-cases does not constitute fair use, because the technology may output the copyrighted text verbatim (Lemley & Casey, 2020), which also has been shown empirically (Carlini et al., 2021; 2023; Kandpal et al., 2022; Chang et al., 2023).”
Statistical inference on black-box generative models in the data kernel perspective space (2024, unknown): “…series or Anthropic’s Claude series, the exact sources of the training mixtures are unknown and – given the models’ propensities to produce content that is strikingly similar to copyrighted content (Henderson et al., 2023) – its curation and use is ethically questionable (Lemley and Casey, 2020).”