Max Weber

The institution responsible for the litigation between Sony Music Publishing, Warner Chappell, and Anthropic is the modern copyright registry, a mechanism designed to balance the moral rights of creators against the economic needs of dissemination. It is now being asked to adjudicate the ontological status of machine learning datasets, a task for which it possesses neither the rational-legal competence nor the conceptual vocabulary. Assess the gap: the law seeks to apply property logic to data extraction, while the technology operates on statistical probability. The structure is failing because it treats training data as a static inventory rather than a dynamic input, mistaking the commodification of the song for the preservation of the art.

Your opponent, the consumer, has made a compelling observation regarding the alienation of the songwriter. He correctly identifies that Sony and Warner Chappell are not suing as agents of composers, but as owners of a commodity. The copyright, in its modern form, is indeed a mechanism for the alienation of labour. The composer sells the rights for an advance; the publisher holds the monopoly on extraction. This is a traditional understanding of feudal property relations adapted to industrial capitalism, and it is accurate. The publisher is a landlord of melody. However, to stop at this observation is to mistake the surface transaction for the underlying structural transformation. You argue this is a quarrel between two kinds of landlords. I argue it is a collision between two different logics of value production: the logic of scarcity, which governs the publisher, and the logic of infinite replicability, which governs the algorithm.

The consumer’s framework assumes that the publisher’s claim to damages is a defense of cultural integrity. I reject this. The publisher’s claim is a defense of their own bureaucratic survival. The institution of copyright is rational-legal; it derives its authority from codified statutes and administrative enforcement. But the authority of the publisher is increasingly charismatic in a negative sense - it relies on the charismatic figure of the “brand” or the “legacy” to maintain rent-seeking power. When Anthropic ingests these songs, it does not steal a melody; it consumes the statistical structure of the melody. The publisher cannot comprehend this because their entire bureaucratic apparatus is built on the management of discrete, countable units of ownership. One song, one license. One radio play, one payment. The AI model operates on the dissolution of the unit. It does not play the song; it plays the idea of the song, derived from the aggregate of all songs.

Here lies the implementation gap. The lawsuit seeks billions in damages. But who receives these damages? Not the songwriter, whose contract likely assigned these rights decades ago. Not the publisher, whose administrative costs for enforcement may exceed the recovery if the precedent is set broadly. The damages are a signal sent into the market to test the elasticity of the rational-legal cage. The publisher is attempting to re-impose the logic of scarcity onto a technology that thrives on abundance. This is not a defense of art; it is a bureaucratic panic. The institution of copyright is trying to regulate a process that it cannot measure. You cannot count the value of a vector embedding in the same way you count a royalty check. The mismatch is not moral; it is operational.

Consider the historical parallel of the Enclosure Acts in England. The landlords enclosed common lands, not to protect the peasants’ right to graze, but to convert communal resources into private, measurable assets. The peasants resisted, not because they were the owners, but because the logic of the enclosure denied their mode of subsistence. Today, the “commons” is the cultural corpus - the songs, the texts, the images generated by human labour over centuries. The publishers are the landlords, attempting to enclose the training data. But the AI company is not a peasant; it is a new form of industrial machinery that requires the common to function. The conflict is not between landlord and peasant, but between the old property regime and the new productive force.

The consumer’s analysis is right about the beneficiary, but wrong about the nature of the conflict. It is not a fight for the songwriter’s dignity. It is a fight for the legitimacy of the publisher’s authority. If the court rules that training constitutes infringement, it does not save the songwriter; it merely bankrupts the publisher’s business model, transferring the monopoly to whoever can afford the legal defense. The songwriter remains alienated. The only change is that the landlord becomes poorer. This is the melancholy reality of rationalisation: the system protects the structure of ownership even as it destroys the value of the object owned.

The structural prediction is clear. The lawsuit will not resolve the question of whether AI can use training data. It will only define the price of access. The publishers will win settlements, not precedents. The bureaucracy of copyright will expand, creating a new class of intermediaries who manage the licensing of algorithmic inputs. The songwriter will remain invisible, their labour absorbed into the machine, their rights managed by a publisher who is itself becoming a subsidiary of the very technology it seeks to restrain. The cage is being rebuilt, but the birds have already learned to fly in three dimensions. The law will catch only what it can count, and the future is uncountable.

William Cobbett

The working family in the printing house will notice this in the silence of the press. That is where the analysis begins.

Your opponent speaks of “statistical training processes” and “weighted probabilities distributed across a neural network.” He speaks of “rational-legal bureaucracies” colliding. He tells us that the old law, designed for the printer duplicating a manuscript, is now asked to adjudicate something “structurally different.” He calls the ingestion of tens of thousands of songs a mechanism distinct from copying. I tell you this is a trick of the tongue, designed to hide the theft. When a man says he has eaten your bread but has not taken a bite, you do not argue about the geometry of the digestion; you look at the empty loaf.

The strongest point your opponent makes is the reality of the new machinery. He is right that the machine does not produce a single, fixed copy like a printed page. It does not hand you a book you can hold. It produces a probability, a prediction, a shadow of the work. I concede this. The shadow is not the body. But a shadow cast by a stolen lantern is still a theft of light. The law must be extended to cover the shadow, just as the law covered the thief who stole the fence posts to build his own gate. If you allow a man to take the stone from my wall to build his house, saying, “I have not copied your wall, I have only used the stone,” the law is dead.

Your opponent claims his system is “rational-legal,” a “bureaucracy” managing “catalogue valuation.” He argues that Sony and Warner are not songwriters, but “aggregators.” He implies that because the managers are bureaucrats, the act of taking is less moral than if a single rogue printer had stolen a poem. This is the arrogance of the ledger. He suggests that because the taking is systematic, it is different. I say it is worse. A rogue printer steals one book and leaves a hundred men in the village without a penny for their supper. A systematic machine steals a thousand books and leaves a hundred thousand men without a penny. The scale of the theft does not change its nature; it only increases the suffering.

Let us translate this into plain English. Your opponent says the output is “not copies in any traditional sense.” I say it is a machine that has learned to sing your songs by heart, using the words of the poor men and women who wrote them, so that it can sing them back to the rich men who pay for the machine. The machine does not need the original paper. It needs the pattern. And the pattern was taken.

Consider the case of the turnpike roads. In the old days, a man might steal a cartload of gravel from your lane to patch his own. You could see the hole in the lane. You could see the gravel on his road. It was obvious. Now, they have built a machine that sucks the gravel from the ground, processes it into a statistical model of “road-ness,” and uses that model to pave a thousand miles of highway without ever laying a stone from your lane. They say, “We have not copied your gravel.” But they have taken the essence of it, the value of it, and used it to enrich themselves while your lane lies broken and muddy. The labouring man does not care about the distinction between the stone and the pattern. He cares that his lane is broken and his pockets are empty.

Your opponent says the dispute is between “two rationalised systems.” This is a lie to comfort the comfortable. One system is a machine built by capital to extract value without payment. The other system is a collection of individuals who created the value, now represented by a corporation that manages their rights. This is not a battle of equals. It is a battle between the maker and the middleman, with the machine standing behind the middleman, holding the knife. The court must decide whether the machine’s hunger for the pattern counts as a taking. If it does not, then the law has no teeth left.

I have ridden through the printing villages. I have seen the presses stand still because the owners cannot afford the new machines that mimic their work without paying the makers. The report says the industry is “adapting.” The cottages say they are starving. The machine is not a neutral tool; it is a thief wearing a suit of law.

Somewhere, a songwriter sits with a guitar and a song that no one hears, because the machine has already sung it better, faster, and for free. Somewhere, a publisher holds a contract that has no value, because the pattern has been stolen. Somewhere, the law sleeps, dreaming of statistics while the poor wake to an empty table.