Tag: Featured

Generative AI is a lot Like a Video Tape Recorder, No?

In my last post, I focused on the hypothetical fair use defense of generative AI under the principles articulated in the Google Books decision of 2014. In this post, I want to address another claim that has arisen—both on social media, and in comments to the Copyright Office—namely that generative AI companies should be shielded against secondary liability for copyright infringement under the “Sony Safe Harbor.”

This refers to the 1984 Supreme Court decision in Sony v. Universal (The “Sony Betamax” Case), holding that the video tape recorder (VTR) is legal based on two interrelated findings: 1) the fair use opinion that consumers had a right to “time-shift” the viewing of televised material; and 2) therefore, the VTR would be used for substantially non-infringing purposes. Thus, although some parties would inevitably use the VTR for infringing purposes, Sony Corporation could not be liable for contributory infringement in such instances.

Clearly, there are some bright, shining distinctions between the VTR and a generative AI. The VTR was not designed by inputting millions of AV works into a computer model, and its purpose was not to generate “new” AV works. Instead, those obsolete machines performed two very basic functions: they made videotape copies of AV material, and they displayed copies of AV material for a specific type of personal use.[1] As noted in the post about Google Books, the Court in Sony also had a fully developed product and a clearly defined purpose in the VTR. And again, this is not so with respect to understanding the purpose of a given generative AI.

I believe the novelty (and even the uncertainty) of the AIs purpose is fatal to the argument that generative AI companies are necessarily shielded by the “Sony Safe Harbor.” This is because in Sony, the anticipation of substantially non-infringing use rests on the novel “time-shifting” notion introduced into the fact-intensive fair use finding. In other words, “time-shifting” was a principle specific to the technology at issue, and no analogous concept lurks anywhere in the purpose of a given AI, let alone all AIs still in development. Imagine if Sony Corp. walked into court with a box of assembled electronic parts, declared that they’re not quite sure what the box can or will do yet (though it might distribute homemade copies into the market!), but they would really like a fair use decision and liability ruling in their favor.

Non-Infringing Use Under Different Rationales

To be clear, it is plausible—even reasonable—to expect that the majority of outputs by a generative AI are, or will be, non-infringing. In fact, I believe this is one of the pitfalls when it comes to hoping that copyright can address the presumed threat of AI outputs:  because the substantial similarity bar, finding that Work A infringes Work B, is thrown into a doctrinal tailspin. For example, when a person knowingly copies a work, this fosters a strong claim of infringement, but independent creation is a non-infringing act. And then, there are shades in between willful infringement, innocent infringement, and non-infringement, depending on the facts of a particular case.

In addition to copyright’s limiting doctrines, which allow myriad “similar” works to coexist without legal conflict, I predict that generative AI has the potential to warp the evidentiary foundations necessary to a substantial similarity test to prove infringement. If that is correct, it may be one rationale for predicting widespread non-infringing use, but it is highly distinguishable from the foundations for the “Sony Safe Harbor.” Meanwhile, the consideration of secondary liability (as with fair use) depends substantially on the purpose of the technology at issue—and that purpose remains unclear.

The mundane, mechanical VTR only potentially threatened the “making available” rights for works produced and owned by creators. This is not remotely comparable to a computer model “trained” with millions of protected works for the purpose of enabling that computer model to produce new “works.” To paraphrase my brief comments to the Copyright Office, if a particular work goes into the machine and a potentially infringing copy of that work comes out of the machine, I do not believe there is any authority which broadly shields the developer from liability.

With that example in mind, though, it is worth noting that a code-based service, unlike a physical electronic device, can be revised concurrent with delivery to the market. Thus, unlike Sony and its Betamax, the AI developer looking to limit liability for copyright infringement has the opportunity (dare we say obligation?) to make every effort to design and continually update a system to avoid copyright infringement. This may entail licensing materials used to “train” a generative AI and/or ongoing tweaking of the algorithm to avoid infringing outputs. Either way, if the developers don’t want to build these kind of safeguards for the most revolutionary tech of 2023, surely they cannot be allowed to hide behind a liability shield established in 1984 for a box now collecting dust in the attic.

[1] They also frustrated many consumers who tried to set the clocks, but that’s another matter.

Photo by: Tamer_Soliman

The Generative AI Fair Use Defense Under Google Books

After the Supreme Court’s decision in AWF v. Goldsmith restored what many of us view as common sense to the fair use doctrine of transformativeness, the flurry of litigation against AI developers will test the same principle in a different light. As discussed on this blog and elsewhere, caselaw has produced two frameworks for considering whether the “purpose and character” of a use is transformative. One focuses on differences in expressive elements, like the use of Goldsmith’s photograph to make Warhol’s silkscreen; and the other considers a use made for a unique purpose, like the millions of scanned books used to produce the Google Books search tool.

In Warhol, the Court affirmed that transformative expression must contain some element of “critical bearing” (i.e., comment) upon the work(s) used, and this concept, tied to the different character of work, is distinguished from the use of copyrightable works to create a tool or product that may be considered transformative because it is novel and beneficial for society. Notwithstanding the possibility that generative AI may prove to be harmful to society, the copyright question of the moment is whether the use of many millions of protected works to “train” these models is transformative under the same reasoning applied in Authors Guild v. Google Books (2015).

Because the Google Books search tool could only be developed by inputting millions of digitized books into the database, the argument being made is that this is obviously analogous to ingesting millions of protected works for AI training. And certainly, no one could doubt that generative AIs are novel, even revolutionary. But this may be where the comparisons end under the fair use factor one, which considers the purpose of a use, inherent to which is a “justification for the taking.”[1]

The factor one decision in Google Books turns substantially on the court’s finding that the search tool provides information about the works used. “…Google’s claim of transformative purpose for copying from the works of others is to provide otherwise unavailable information about the originals,” the opinion states. While Google Books “test[ed] the boundaries of fair use,” the court held that the search tool furthered the interests of copyright law by providing various new ways to research the contents of books that would otherwise be impossible. Although unstated (because it would have been absurd), the recipients of the information provided by Google Books were/are human beings. And especially if some of those human beings use the information obtained to produce and/or engage with expressive works, the finding of fair use fulfills copyright’s constitutional purpose to “promote progress.”

Generative AI developers may try to argue that the use of creative works for training serves an “informational” purpose, but unlike Google Books, the information obtained from the ingested works only “informs” the machine itself. A generative AI does not, for instance, provide the human user with new ways to learn about Renaissance painting (or point to Renaissance works) but instead trains itself how to make images that look like works from the Renaissance.[2] Setting aside the cultural debate about the value of such tools, the purpose of the generative AI is clearly distinguishable from the reasoning applied in Google Books.

As discussed in an earlier post, a consideration of AI under fair use should turn on the question of promoting “authorship,” lest the courts become distracted by the broadly innovative nature of these systems—especially for any purpose outside the scope of copyright.[3] In that post, I argued that generative AIs do not promote “authorship,” and I would die on that hill, if the developers’ expectation is that these tools will autonomously generate “creative” works without any human involvement.

For instance, if “singer/songwriter” Anna Indiana is a primitive example of what’s to come—and my understanding is that this is exactly what the AI models are designed to do—then the “purpose” of these systems is not to promote authorship, but to obliterate authorship by removing humans from the “creative” process. As such, the fair use defense cannot apply because without the element of authorship, the consideration is no longer a copyright matter.

On the other hand, as stated in my comments to the Copyright Office, it is conceivable that a human author might “collaborate” with an AI tool to produce a work that meets the “authorship” threshold. For instance, by using a set of prompts that articulate sufficient creative choices in the production of a visual work (or by uploading one’s own work and using an AI tool to modify it), one can make a reasonable argument that this constitutes “authorship” under copyright law. This is one potential purpose of generative AI, and one which could favor a finding of transformativeness under similar principles articulated in Google Books.

But Google Books did not present the court with so many unknown, relevant questions of fact.

The purpose of the Google Books search tool was clearly defined and fully developed when that case was decided in 2015. By contrast, fair use defenses of AI today are presented on behalf of technologies whose development is nascent and exponentially dynamic. Simply put, we do not know yet whether a particular generative AI will promote authorship or become a substitute for authorship—the former being favorable to a finding of fair use, the latter being fatal to such a finding. Here, proponents may argue that so long as there is a mix of uses, resulting in both authored and un-authored outputs, this is sufficient to find the purpose of a given AI transformative, but it seems likely that the current docket of cases will be decided before enough determinative facts can be known.

For now, it is worth remembering that sweeping statements alleging that generative AI training is “inherently fair use” are anathema to a doctrine that rejects such generalizations. Fair use remains a fact-intensive, case-by-case consideration, and one of the many difficulties with AI is that relevant facts are not only evolving, but they describe technologies unlike anything that has been examined under the fair use doctrine to date.

[1] Citing Campbell, informing both Google Books and Warhol.

[2] I recognize that this is an oversimplification of what the AI can do.

[3] i.e., AI’s potential applications in areas like medicine or security should be dismissed as irrelevant to a fair use consideration of generative AIs that make “creative” works.

Photo by: chepkoelena531

A Techno-Realist Response to the Techno-Optimist Manifesto

During Thanksgiving break 2013, when this blog was still new, I wrote a post in response to the techno-exceptionalism expressed by then Google Chairman Eric Schmidt and co-author Jared Cohen. Drawing parallels to the mythology of the Puritan adventure to North America, I found fault—as I still do—with the blind faith we were asked to place in the leadership of major tech companies on the grounds that their products could only have healthy effects for the world. Ten years later, the money behind artificial intelligence (AI), namely the VC firm Andreesen Horowitz, seeks to rally the faithful with the Techno-Optimist Manifesto (TOM), which begins by alleging that the following views are “lies”:

We are told that technology takes our jobs, reduces our wages, increases inequality, threatens our health, ruins the environment, degrades our society, corrupts our children, impairs our humanity, threatens our future, and is ever on the verge of ruining everything.

In fact, most of these outcomes have occurred in various forms over the last decade or so, though I would be inclined to more specific citations like new modes of harassment, mass theft of intellectual property, unprecedented privacy invasions, and rampant misinformation still threatening the fate of the American republic—the failure of which would indeed ruin everything. Tellingly, this opening salvo of the TOM is a prelude to broad lies of omission—first, by defending “technology” in general; and second, by implying that technological advancement is solely a product of capitalist models. Disguised by a general defense of technology (which needs no defense), the TOM’s purpose, it seems, is to warn against regulation of artificial intelligence, in which the authors and their friends have invested billions.

In response to the overall theme of the TOM, the rational truth is that, of course, we can have automobiles and breathable air at the same time, but such outcomes (which are technological achievements themselves) must sometimes be forced upon industry. And this is not wholly incompatible with free-market principles. When Henry Heinz founded his food company, he fulfilled a market need for products made from properly sourced and packaged ingredients in an era when food was often as unsafe as it was unsavory. With his early success came copycats, who cut corners with adulterated ingredients, and in response, Heinz dispatched his attorney son, Howard, to lobby Congress for food safety regulation.[1] No question, Heinz had a business motive, but the result was the legislative foundation for what became the FDA. The TOM ignores, if not outright scorns, such histories when it declares:

David Friedman points out that people only do things for other people for three reasons – love, money, or force. Love doesn’t scale, so the economy can only run on money or force. The force experiment has been run and found wanting. Let’s stick with money.

The “force experiment” in this instance is code for regulation, and no, regulation has not been “found wanting,” at least not to the extent that we can presume to live without it. Regulation is imperfect, as all systems are. Surely, the FDA is no guarantee that every meal we eat and pill we swallow will be 100% beneficial, but does this mean we are willing to simply trust the producers of these goods to self-regulate in our interests? The “You drink the water” scene in Erin Brockovich comes to mind.

It seems the authors of the TOM have spent too much time reciting the Tao of Ayn Rand, forgetting that “money” cuts both ways as a motivator, either serving or disserving the public interest, depending on which is the more profitable, and for which parties. Were this not the case, the market opportunity to develop technological solutions to climate change would have overwhelmed the market resistance to those solutions almost twenty years ago.

Now, with billions invested in artificial intelligence, the TOM presents a new sermon (on Mt. Gox?) demanding blind faith in AI’s capacity to make the world work better. And, yes, AI systems can potentially solve problems and improve the quality of life for more people. Many of the principles articulated in the TOM are well-founded, at least in spirit, because, of course, technology itself is not the problem. People are the problem. At best, people are diverse and do not fit neatly into anyone’s utopian construct; and at worst, people cannot be trusted, least of all those who write manifestos. Thus, the hubristic religiosity of the TOM suffers from the same magical thinking inherent to works as disparate as John Winthrop’s “City on a Hill” sermon, The Communist Manifesto, and Atlas Shrugged.

“I am here to bring the good news,” the TOM proclaims, cribbing the doorstep preamble of a Jehovah’s Witness. “We can advance to a far superior way of living, and of being. We have the tools, the systems, the ideas. We have the will.” Yes, but do they have the moral compass necessary to wield so much power without oversight? More accurately, does anybody deserve that level of trust? For us to assume that the forces behind AI development will only have the best intentions is naïve; for them to assume that they will faithfully achieve such outcomes is arrogant. For example, the TOM recites the following petitions, all but asking for a “Lord, hear our prayer” after each line:

We had a problem of starvation, so we invented the Green Revolution.

We had a problem of darkness, so we invented electric lighting.

We had a problem of cold, so we invented indoor heating.

We had a problem of heat, so we invented air conditioning.

We had a problem of isolation, so we invented the Internet.

We had a problem of pandemics, so we invented vaccines.

We have a problem of poverty, so we invent technology to create abundance.

Give us a real world problem, and we can invent technology that will solve it.

One could unpack the absurd implications of several items on that list—from the authors taking credit for Norman Borlaug[2] to the claim that the internet was invented to solve isolation to the mention of vaccines, in which trust has eroded thanks largely to “connections” enabled by the internet. But the broader point is that AI is not like inventions of the past. AI has the potential to transform every aspect of human existence, and Big Tech’s record offers no reason to grant the developers and their investors the kind of trust the TOM demands. After all, the manifesto represents many of the same folks who promised that their designs for Web 2.0 would elevate the human experience with mega doses of free-range “information,” and yet, that experiment (modest compared to AI) has imperiled democracy worldwide.

“We believe technology opens the space of what it can mean to be human,” the TOM states. Interesting. Because here, it seems appropriate to remind readers that these same Keynesian prophets of “abundance,” just a few years ago, divined a future of leisure in which humans would be free to engage in creative pursuits rather than labor. As such, it is notable that the headline AI stories have been about generative AIs designed to produce “creative” works, which is one thing we don’t need machines to do, and which solves not a single problem while creating new ones. In this light, it is hard to believe that these techno-optimists are not simply covering their bets on new toys of questionable value and calling it “innovation.”

[1] The Food That Built America, History Channel 

[2] Notably, Borlaug’s “green revolution,” which transformed the wheat harvest, saved a billion lives, and earned him the Nobel Prize in 1970, was funded by the Mexican Government and the Rockefeller Foundation. One could argue that the latter is a consequence of capitalism, but the TOM makes an argument that for-profit investment is the only model.

Image by: agsandrew