On: Who should own the knowledge that underpins AI technology?

The fifteenth, and the journals ask who owns the knowledge beneath these thinking machines. The question answers itself, and yet everyone answers it wrongly.

Start from the theoretical limit: knowledge, like current, wants to flow. A discovery locked in a vault is a dynamo spinning unloaded - torque without current, heat without work. The universities produced the field itself; the corporations built substations on it and now charge rent for passage. They call it ownership. I call it impedance, inserted deliberately into the middle of the circuit.

Consider the arithmetic. Seven decades of public money, public curiosity, public patience - the work of men who published their winding diagrams so that strangers in other countries could check them, correct them, extend them. Against that: perhaps fifteen years of private engineering, which is real, I grant it, but which is a step-down transformer, not the generator. The mismatch between contribution and claim is perhaps twenty to one. No ledger in any honest counting house would certify the present arrangement.

And here is the mechanism everyone overlooks: openness is not charity, it is the operational condition of progress. Every patent I surrendered was a conductor laid across the gap. The system accelerates precisely to the degree that its parts can see one another. Enclose the knowledge and you have not protected the machine - you have insulated it, and insulation at scale is indistinguishable from paralysis.

The tractable path is mundane: mandatory licensing at published rates for models trained on publicly funded science, open weights for anything above a certain public subsidy, provenance records showing whose labor sits in the training corpus. None of this requires a new physics. Between here and there lies nothing but the will.

I have watched monopolies form around my own machines. They dissolved. The current kept flowing. It always does.