Data Is Nothing Like Oil: Dusting Off Copyright’s Originality Doctrine for the AI Age

We have all heard it said that ''data is the new oil.''And perhaps, for those who dream of striking it rich by mining, commodifying, and exploiting data in an AIdriven economy, the analogy holds.For everybody else, however, it behooves us to recall that data is nothing like oil.Both data and oil can generate immense economic value and power, to be sure, and it turns out their rampant extraction and exploitation can do untold harm to our environment; but that is, by and large, where the analogy ends.Unlike oil, data is not tangible, it is not finite, it is neither rivalrous nor excludable.In economic terminology, that makes it a public good.Even if we are willing, for the sake of argument or analogy, to overlook the physical and fundamental differences that distinguish tangible and intangible things, when seen through the lens of copyright law, data should look nothing like oil; if anything, as I will explain, it should look to us more like the ocean.Of course, we all know by now that vast quantities of data are essential for training AI systems, and that the data sets assembled for this purpose almost invariably comprise myriads of copyright-protected works.Copyright infringement cases are consequently spawning around the globe, testing the lawfulness of using these works as content for training data without the consent or compensation of copyright owners.Government consultation papers and policymakers repeatedly ask how current copyright laws apply, whether new laws are needed, and if so, what they should say and how they could work.Meanwhile, large AI firms and content industry actors are busy privately negotiating lucrative licensing arrangements to secure exclusive access to high-value data troves.In this dawning era of AI, when so much remains in flux and the future trajectory of this world-changing technology is being hashed out in courtrooms, boardrooms and back rooms, the metaphors we use to understand our predicament could prove