A policy mechanism, proposed by Eric Ries, that would let AI companies train on all copyrighted material while requiring a share of the resulting profits to be redistributed to the original creators. It reframes the AI-copyright fight from an all-or-nothing consent question ("was it legal to train on this?") into a revenue-sharing negotiation ("how much of the resulting profit goes back to creators?").
Ries frames the stakes with a pointed inversion of the usual complaint: "If you think being in the training set is bad, you should try being excluded from the training set. Okay, that's even worse" — arguing opt-out regimes risk locking marginalized or under-resourced creators out of whatever economic upside AI training eventually generates, which compulsory licensing avoids by including everyone and paying out.
Use compulsory licensing as a policy-advocacy framing in copyright/AI debates — shift the argument from "should training be allowed" to "how is the resulting revenue shared" — since the latter is a negotiation that can actually converge.
Sits alongside sovereign-wealth-fund-model-ai as one of two profit-redistribution proposals Ries makes for the AI industry; compulsory licensing addresses training-data creators specifically, while the sovereign-wealth-fund idea addresses the broader public.