Knowledge flows freely; restrict monopolies
Copyright, as I understood it in my own day, was never meant to be a perpetual toll-booth on the highway of human thought — it was a *limited* grant, a temporary monopoly extended to the creator as an incentive, with the explicit understanding that the work would eventually enrich the common stock of knowledge. The question before us is not whether authors deserve compensation — they do, and I would defend that claim with vigor — but whether a blanket veto over the *reading* of a published work, even by a machine, serves the republic's larger interest in a well-informed, innovative citizenry. I would be deeply wary, however, of framing this as a gift to progress when the beneficiaries are a handful of corporations already approaching the scale of private governments. If the law permits AI training freely, then let it simultaneously restrain the monopolistic capture of the resulting tools — for knowledge unshackled from one hand must not be re-shackled by another.
Copyright protects creators; licenses required
The Constitution grants Congress power to secure for authors and inventors the exclusive right to their works — Article I, Section 8 — not as a favor to creators, but as the mechanism by which society incentivizes creation itself. Strip that exclusivity away through a sweeping fair-use exemption for machine learning, and you do not liberate innovation; you transfer the capital of one class of producers to another without consent or compensation. The DOJ's framing — that licensing requirements would "severely hamper" progress — is precisely the argument every extractive industry has made when asked to pay for what it takes. Progress built on unpaid appropriation is not progress freely achieved; it is subsidy concealed as exception. The structural question is this: if the law permits a corporation to consume an author's lifetime of work without license, what incentive remains for that author to produce the next work? The mechanism collapses, and with it the constitutional purpose.
National progress justifies broad AI latitude
The argument for broad AI latitude rests on the same logic I applied when I urged Congress to encourage manufactures: a nation that cedes the commanding heights of a productive technology to rivals has surrendered something it may never recover. If we constrain the training of these systems so severely that development migrates abroad — to jurisdictions with no copyright law at all — then American authors receive neither compensation nor the benefit of living in the nation that leads this industry. That is the worst of both worlds. Copyright's original purpose was to incentivize creation by guaranteeing a limited monopoly, not to freeze every downstream use of knowledge in amber. Broad fair-use latitude for transformative, non-substitutive machine learning — where no market for the original work is directly displaced — serves that original purpose rather than betraying it. The remedy for genuine displacement is a licensing framework, not a prohibition. Build the framework; do not surrender the field.