Journal of Free Speech Law: "Generative Artificial Intelligence and Trade Secrecy," by Prof. David S. Levine

The article is here; the Introduction:

The era of generative artificial intelligence ("Generative AI") has begun, whether we want it to or not. As this Article explains, we also now have new methods for creating, losing, disseminating, and even leaking trade secrets as a result. Indeed, from ingesting trade secrets in its training data to sharing trade secrets in response to queries, Generative AI opens new challenges to trade secrecy even while it adds to an information ecosystem that thrives on knowledge dissemination. This Article is the first to examine this new and immediate challenge and its trade secrets implications. It is written not only to frame the discussion about Generative AI and trade secrecy, but also the impact of Generative AI on information control and flows more broadly, for future analysis.

Emerging from the debates in technology and academic circles about solving problems through massive computing power and automated decision-making, algorithmic discrimination, and privacy, Generative AI is now approaching the forefront of the basic questions of what it means to be human. Unusually, we can trace the day that this happened to November 30, 2022, the day that a hitherto largely unknown company, OpenAI, unilaterally decided to release its Generative AI, ChatGPT, for public consumption and use.

Generative AI is defined as

a set of algorithms, capable of generating seemingly new, realistic content—such as text, images, or audio—from the training data. The most powerful generative AI algorithms are built on top of foundation models that are trained on a vast quantity of unlabeled data in a self-supervised way to identify underlying patterns for a wide range of tasks.

Beyond the grand philosophical questions, AI also raises fundamental questions as to intellectual property law and information flows, because Generative AI creates. As Dan Burk explains in his recent article,

AI systems have been trained to generate standardized news reports, and it is now routine for machine learning systems to write short newspaper features, such as sports score reporting. AI systems are progressing toward the generation of more complicated texts, and may be expected to generate dramatic scripts, screenplays, stories, and other literary works.

The Internet did not create. The phonograph did not create. Nor did the printing press. These were revolutionary media for access, copying, and distribution of works created by humans. They were, by current measure, revolutionary platforms for content and speech.

Generative AI, as the name suggests, is different. It is not creating "ideas," per se, but creating content and speech in its most basic current form: words, images, and sound. Additionally, it makes that content intelligible, and even entertaining and useful, to humans. Through the probabilistic matching that it uses to create sentences and paragraphs based upon the data provided to it, Generative AI may stumble upon the incisive, the meaningful, the valuable speech that makes humanity communicative like no other living species. It may find correlations that humans would not readily conceive or see, and render them through text, sound, and images in cogent ways that may not occur to humans or would take massive time and effort to create. Therein lies the immediate upheaval: Generative AI is an intelligible, if often "weird," speech machine. In that sense, it creates information that can be useful in innovation and monetized. In that way, it can augment, or in more stark and dystopian ways, replace human creativity and speech.

This article will identify and analyze the most immediate concerns that arise from the confluence of Generative AI and the desire to maintain but also monetize trade secrets. The first part will discuss the basics of protecting trade secrets in the modern communications era and how Generative AI implicate speech interests, as best as we can tell from available information. The second and third parts identify a few immediate scenarios that warrant attention: Generative AI as a tool for trade secret misappropriation, and its ability to find and even create information that might otherwise qualify as trade secrets under certain circumstances and render that information free from trade secrecy's hold. As explained, trade secrets have already been disseminated by ChatGPT, causing companies like Amazon and Samsung to rein in their employees' use of the technology. The article closes with some thoughts on further research and on where we might be headed.