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

Fifteenth of September, and the newspapers ask who owns the knowledge beneath these thinking machines, as though knowledge were a bolt of cloth one could fold and carry off.

I confess the question vexes me, because I know the ledger better than the askers. Every mechanism now sold for a fortune stands upon forty years of work done in public rooms - professors paid from common coffers, their findings given freely to any who could read them. The algebra of gradients, the architecture of attention, the patient proofs: none of it was purchased at market. It was gifted, on the understanding that the gift would circulate.

And now observe the operation. Step one: the public funds the discovery. Step two: the discovery is published openly, as the scholars’ covenant requires. Step three: private parties take the whole corpus, train upon it, enclose it. Step four: the output is offered back to the public at a price, under patents, as proprietary art. At no step does the original giver consent; at no step may they withdraw. The distance between entry and exit of the common contribution is the whole of the fortune.

My country has seen this before. When the commons were hedged, the apologists called it improvement, and the improvement was real - yields rose, flocks fattened. But the enclosure changed who the land worked for, and that question was settled by deeds drawn quietly, not by debate. So it is here. The answer to “who should own it” was decided years ago, by the accumulation of steps, and the newspaper question arrives only after the fence is built.

I am not so austere as to grudge a man profit on his machinery. Babbage’s engine would have needed patrons regardless. But there is a difference between rewarding the weaver and claiming the pattern was never woven from common thread. The scholars gave their work as one gives to a library, not to a vault. The vault should say so plainly, and pay the library.