Microsoft Copilot rarely copies full sentences, firm claims

The principle operating here, stated plainly, is this: an act of appropriation is permissible if it occurs infrequently enough to escape being called a pattern. Microsoft’s filings do not argue that Copilot never reproduces the sentences of The New York Times or of book authors; they argue that it rarely does so. Let us ask whether this principle, universalised, produces coherence or contradiction.

Consider what it would mean for every user of another’s labour to adopt rarity as a defence. The pickpocket who lifts one wallet in a thousand transactions does not thereby establish a right to the nine hundred and ninety-nine he declined to touch; he establishes only that he was caught less often than he might have been. A maxim of the form “I may take what is not mine, provided I do so seldom” cannot be universalised without contradiction, because its very coherence depends on most agents refraining while a few defect. If all agents reasoned this way, the defence collapses into its opposite: everyone would take rarely, which is to say everyone would take, and the concept of rare appropriation would dissolve into ordinary appropriation wearing a modest face. Frequency is not a moral category. It is a statistical one, and Microsoft has confused the two, presenting a defence about incidence as though it answered a question about right.

The deeper error appears when we apply the formula of humanity. The authors whose books trained Copilot’s outputs, and the journalists of the Times whose sentences occasionally surface intact, are not merely sources of statistical noise to be minimised in a filing. They are ends in themselves, agents whose labour was undertaken as labour - as the exercise of a rational will directed at a purpose, namely the writing of a sentence that says a particular thing in a particular way. When Copilot reproduces that sentence, however seldom, it does not borrow a fact; facts are common property and no one wrongs another by restating them. It reproduces the specific form the author gave to a thought, which is the one thing in the transaction that belonged to the author alone. To treat that reproduction as tolerable so long as it is statistically marginal is to treat the author as a resource whose objections may be discounted below a certain threshold of frequency - which is precisely to treat a person as a means rather than an end, only with the treatment calibrated by volume rather than announced by principle.

Here the strongest form of Microsoft’s position deserves its due. One might say: no system of retrieval or generation can guarantee zero overlap with prior text, any more than a court reporter transcribing testimony can guarantee no accidental echo of a phrase heard elsewhere. Rarity, on this reading, is not a moral defence but an engineering description of an unavoidable residue. I grant the description. I deny that it settles the question of right. The unavoidability of occasional overlap does not entail permission to build a commercial product whose economic value depends on precisely that overlap occurring often enough to satisfy users while remaining rare enough to satisfy lawyers. A residue that is structurally necessary to the product’s function is not an accident the product suffers; it is a feature the product exploits, and exploitation cannot be laundered by describing its frequency in modest terms.

The duty that follows is not a duty to abolish such systems, which I am not competent to adjudicate as engineering, but a duty of honesty in the maxim offered to justify them. If Copilot’s makers wish to defend the practice, let them defend the principle that a machine may draw upon protected expression to generate value for its maker - and let them test that principle against the will of the author whose sentence, reproduced once in ten thousand queries, was nonetheless the one sentence that mattered to her.