Plan snuffs out all small business development in AI

Even copyright owners and the industry groups who represent them recognize the impracticability of voluntary, direct licensing for AI training data. They propose, instead, that Congress pass legislation to implement a statutory or collective licensing model.19 But such legislation would effectively require AI developers to remunerate rightsholders for a use that falls squarely within the protections of the fair use doctrine. The Supreme Court has warned against such attempts to “alter[] the traditional contours of copyright protection” by reducing its speech-protective limitations, like the idea/ expression dichotomy and fair use. Eldred v. Ashcroft, 537 U.S. 186, 219 (2003).

Moreover, a collective or statutory licensing scheme would prove administratively impossible to implement, as the Register acknowledged in a recent Congressional hearing.20 The most obvious problem is scale. Some of the most powerful AI models in existence today were trained on an enormous cross-section of all of the publicly available information ever published on the internet—that is, billions of pieces of text from millions of individual websites. For a very significant portion of those works, it is essentially impossible to identify who the relevant rights holders are, and thus there would be no viable way to get statutory royalties to the proper parties. Moreover, since this technology enables production of creative work at an unprecedented rate, the problem will compound over time. The Office has encountered similar costs and administrative problems with routing royalties to rights holders in administering the Music Modernization Act, where the total number of musical works in the entire ecosystem is something on the order of 25 million. In the context of AI training data, where the relevant quantity of works is almost certainly in the billions, these costs and problems would be multiplied by many orders of magnitude. The amount of “unmatched” royalties would be astronomical, leaving almost all creators with no remuneration at all.

Nor do the economics of any sort of statutory licensing scheme make sense. Again, a staggering quantity of individual works is required to train AI models. That means that, under any licensing framework that provided for more than negligible payment to individual rights holders, AI developers would be liable for tens or hundreds of billions of dollars a year in royalty payments. AI development in the United States would thus become much more expensive than in a jurisdiction with less burdensome restrictions, thereby entrenching the dominance of incumbent technology platforms and discouraging investment in the technology startups that drive U.S. innovation in the AI space. The creation of an impossibly high financial barrier to AI development that only the largest, technology companies have any hope of clearing is a step backward from the current environment in which innovative, new entrants are challenging these dominant platforms. Any participation by small businesses or individual innovators in the nascent AI industry—like the startups that a16z invests in—would be all but barred, and competition would be snuffed out.5