Over the past 13 years, I have repeated variations on the theme that strong copyright rights are essential because a healthy democracy requires a diverse, professional creative sector. Typically, I have advocated this perspective to refute the claim by the copyleft that copyright rights are in conflict with the speech right. Now that we are in climate in which creators find themselves asserting their rights against forces that wish to limit their copyright rights, policymakers should take note that America’s diversity of expression is a critical advantage in developing world-class artificial intelligence (AI).
AI developers want policymakers to believe that respecting copyright rights is a barrier to innovation and America winning the AI race with China. A key point animating that claim is that because China doesn’t respect IP, America must likewise cheat to avoid being at a strategic disadvantage. This is misguided. As a baseline principle, it should not be the policy of the United States to risk $2 trillion worth of economic value in the creative industries as a chip on the AI roulette wheel, let alone to play the game on China’s terms.
“Silicon Valley watchers worry that enthusiasm for AI has turned into a bubble that has increasingly loud echoes of the mania around the internet’s infrastructure build-out in the late 1990s,” reports the Wall Street Journal. That AI investment and hype are incubating a bubble is almost certain, but development will persist, and the question remains how to build world-class AI products that can fulfill some of the grand promises of the industry. One answer to that question is that America must not squander the advantage gained by its foundational commitment to IP, which has fostered a rich supply of creative and cultural wealth unmatched anywhere in the world.
The Science Requires Symbiosis
Although I do not personally like to think of creative works as raw materials comparable to iron or coal or fossil fuels, I shall indulge the AI developers in this analogy to stress the point that respecting copyright rights is not an option. Simply put, if development of AI severely degrades the incentive of people to create new works, then both the creative economy and the AI models will collapse. The issue is described in a July 2024 paper by a team of computer scientists, explaining that as LLMs train on “recursively generated data,” the data becomes poisoned and the models collapse.
It’s not a complicated idea. Just like a power plant cannot burn the same lump of coal twice, or a farmer cannot grow anything in soil depleted of nutrients, LLMs need a steady supply of new material. “To sustain learning over a long period of time, we need to make sure that access to the original data source is preserved and that further data not generated by LLMs remain available over time,” the paper states. [emphasis added]
Thus, while the current generation of LLMs train on billions of available works, the next generation will cannibalize unless there is new creative work on which to continue training. In other words, to achieve the best ambitions of AI, the American plan cannot envision a future without authors of copyrightable works—neither a future in which AI replaces too many creators nor one that harms the incentive to create, which lies at the heart of IP.
If nothing else, the science of model collapse insists that to develop world-class AI, it is essential to uphold the American traditions of strong IP and the absolute certainty of the speech right for all creators. These two forces have long been a source of strength that no other nation can claim. Simply put, if we attack the copyright rights of authors from one side and/or speech rights from another, then we will cede ground that is ours to lose.
Copyright Rights Support Safer AI Products
While there may be ideological differences about the value of diversity in creative expression, there is allegedly no political divide when it comes to AI and harm to young users. Senators Hawley and Durbin in a recent Judiciary Committee hearing, repeated the theme that Congress is in lock step when it comes to Big Tech’s utter lack of responsibility to young users in particular. And although it is popular to say that we shouldn’t talk about AI as a monolith, this principle should not fail to acknowledge that the LLM trained on unlicensed creative works may be the exact same product that lacks guardrails and safety features.
Chat GPT is Chat GPT, whether it is used to expedite scholarly research or by a sixteen-year-old to encourage and plan his suicide. And one ethical consideration is clear enough: no authors of literature volunteered to have their work “train” a machine that would help a child hurt or kill himself. And perhaps the policy considerations are more closely aligned than they appear. Although licensing creative works for AI “training’ will not in itself foster safety design in the products, a licensing regime will necessitate one thing that both IP owners and countless parents are demanding—transparency.
Consumers, lawmakers, and the courts should decline to accept that AI development must be allowed to proceed apace in the “black box” without public scrutiny over what goes into or comes out of the box. Members of Congress on both sides of the aisle who today find themselves regretting past grants of unconditional immunity to the Tech Bros should heed the warnings of the creative community and this time, reject the culture of moving fast and breaking things. As Joseph Gordon-Levitt describes in a recent post about AI, we can move fast but also steer. In fact, I don’t think steering is optional.
Image source by: TRAIMAK
