Microsoft and OpenAI are putting pressure on US copyright owners to produce information on their “general usage” of ChatGPT and other artificial intelligence tools, while Stability AI is seeking the prompts and outputs generated by counsel for Getty Images during its pre-suit investigation in the UK. On both sides of the Atlantic, the defendants are pushing back on the narrative that AI represents a threat to creatives — the same kind of pushback that led Sony, the maker of the Betamax, to prevail against the entertainment industry at the Supreme Court 40 years earlier.
In a recent joint letter to a US magistrate judge, Microsoft and OpenAI lamented the refusal by the Author’s Guild and a group of individual authors to share information the defendants say they need to mount a fair use defense.The request for a pre-motion conference with Judge Ona T. Wang of the US Southern District of New York argues that if the plaintiffs utilize artificial intelligence to “assist with research and writing” it would show the “transformative and beneficial nature” of ChatGPT and other generative artificial intelligence tools “beyond the cherry-picked uses alleged in Plaintiffs’ Consolidated Complaint” (see here).
The bid feels unlikely to succeed given Wang’s rejection last month of a similar request by Microsoft and OpenAI of the New York Times, who sued for infringement in December 2023 (see here).
The motion by the defendants in that case theorized that “if the Times is using technology similar to OpenAI’s, even if offered by another company, that would undermine the Times’s assertion that tools like ChatGPT threaten journalism,” but Wang wasn't persuaded by the relevance of the information to the question of fair use.
Nonetheless, Microsoft and OpenAI in a joint letter to Wang last week (see here) reframed their request as one for "production of documents and communications concerning the benefits of OpenAI’s technology to the journalism industry."
— Defense, not plaintiff, scrutiny —
While defendants in the artificial intelligence industry have coalesced around fair use as a shield against their potential copyright infringement liability, they've faced a series of setbacks during discovery at fair use factor one, which asks the purpose and character of the use, and factor four, which assesses the effect upon a potential market for the original.
The remaining fair-use factors articulated in Section 107 of the Copyright Act require an analysis of the nature of the copyrighted work and the amount and substantiality of the portion used in relation to the work as a whole.
“Each . . . requires scrutiny of a defendant’s purported use of the copyrighted work(s), and whether that defendant’s use may constitute ‘fair use’ under the Act,” Wang explained in the Times order.
“The factors don't require a court to examine statements or comments a copyright holder may have made about a defendant’s general industry, whether the copyright holder has used tools in the defendant’s general industry, whether the copyright holder has admitted that other uses of its copyrights may or may not constitute fair use, or whether the copyright holder has entered into business relationships with other entities in the defendant’s industry,” she added.
— Pre-suit investigation —
Similarly, Microsoft and OpenAI’s bid to compel discovery on the prompts and outputs generated during the plaintiffs’ pre-suit investigations was also swatted down by Wang, twice — in the consolidated case by the Author’s Guild and authors as well as in the case by the Times, which was later consolidated with infringement claims by eight Alden Global daily newspapers.
That finding aligns with an order by Judge Araceli Martinez-Olguin in August (see here).
There, counsel for a group of authors led by Paul Tremblay agreed to produce the scripts used to generate the examples of infringement provided in their June 2023 complaint but refused to hand over the negative prompts or identify those involved with testing ChatGPT prior to filing suit in the US Northern District of California.
Martinez-Olguin denied a defense motion to compel upon finding that the requested information would reveal “mental impressions and opinions” about the LLM that have no bearing on infringement.
In all three cases against Microsoft and OpenAI — by the Author’s Guild, Tremblay et al., and the New York Times — the defendants have advanced, or are advancing, the same theory during discovery that they raised in their motions to dismiss: that ChatGPT is capable of substantial non-infringing uses.
Moreover, they say they could demonstrate as much by pointing to the plaintiffs’ own actions — if only the district courts would order the plaintiffs to produce the requested information.
— ‘Unprecedented’ —
There was no warm welcome from the entertainment industry upon introduction of video tape recorders, or VTRs, to the market during the 1970s, either. One such device, the “Betamax” by Sony, drew the ire of Universal City Studios.
In a complaint filed in the US District Court for the Central District of California, Universal alleged that because the Betamax could be used by consumers for copyright infringement, Sony, by extension, should face contributory liability.
Universal sought an injunction barring Sony from marketing the Betamax, damages and an equitable accounting of profits. Sony responded by invoking the doctrine of fair use. After its initial win at trial in the district court was wiped out by the US Court of Appeals for the Ninth Circuit, the consumer electronics giant responded by successfully petitioning for writ of certiorari with the US Supreme Court.
There, it would go on to prevail in 1984.
Writing for the 5-4 majority, US Supreme Court Justice John Paul Stevens called Universal’s attempt to impose liability on a distributor of copying equipment “unprecedented.” At fair use factor one, Stevens explained, “Sony demonstrated a significant likelihood that substantial numbers of copyright holders who license their works for broadcast on free television would not object to having their broadcasts time-shifted by private viewers.”
Moreover, Stevens continued, Universal “failed to demonstrate that time-shifting would cause any likelihood of non-minimal harm to the potential market for, or the value of, their copyrighted works,” at fair use factor four.
“The Betamax is, therefore, capable of substantial non-infringing uses,” Stevens declared, clearing Sony of contributory copyright infringement liability.
— Process questioned —
The same scenario playing out for Microsoft and OpenAI in California and New York is also playing out abroad where, in the UK, a copyright case by Getty Images has become mired in discovery disputes. According to Stability AI, the photo agency has been less than forthcoming with requests for information on Getty’s statement of case on infringement, or SOCI.
Stability AI says that whether Getty had a good faith belief in the merits of its copyright claims is unknowable without first knowing the prompts Getty fed Stable Diffusion, an AI-based image generator.
Last month, the impasse spilled over into the High Court of Justice, where counsel for Stability AI said the 17 examples of infringement pointed to in Getty's SOCI are atypical and are most likely the result of “a careful and highly selective process” of ensuring an infringing output by the plaintiff.
“If Getty intends to rely on the examples as representative . . . of a wider infringement, they must particularize a positive case as to how these examples were selected and why they were representative,” Nicholas Saunders KC argued.
Another point of contention for the parties at the two-day hearing involved Stability AI’s request for specifics on who tested Stable Diffusion on Getty’s behalf, and Getty’s assertion that the information is privileged. MLex understands that High Court Judge Joanna Smith, at the conclusion of testimony, reserved a ruling.
Counsel for Stability AI declined comment. Counsel for Getty did not respond to a request for comment.
— Fresh look —
In Sony, Stevens said it was “not implausible” that the benefits for consumers of recording television content for later playback would flow to Universal and other rights holders, because “the Betamax makes it possible for more persons to view their broadcasts.”
The public interest in greater accessibility to copyrighted works, and not less, is not unlimited, he clarified. However, in what AI companies surely hope is a harbinger of things to come, Stevens said that the public interest “supports an interpretation of the concept of ‘fair use’ that requires the copyright holder to demonstrate some likelihood of harm before he may condemn a private act of time-shifting as a violation of federal law.”
Stevens later mused that Congress “may well” need to “take a fresh look at this new technology, just as it so often has examined other innovations in the past.”
Congress is currently mulling more than 100 bills that relate in some way AI, while the US Copyright Office is poised to release the second of three reports commissioned by President Joe Biden later this month (see here).
None will come to fruition in time to provide relief to the current crop of AI defendants.
Please e-mail editors@mlex.com to contact the editorial staff regarding this story, or to submit the names of lawyers and advisers.