BoI - AI Content: Genius or Garbage

Let me concede the point my opponents most want to win. Most AI-generated content is garbage: the limp listicle, the "rich tapestry of human experience," the email that says nothing in four paragraphs. If the proposition were "most AI output is good," I would sit down now. But nobody has ever judged a medium by its average. Sturgeon's law says ninety percent of everything is crud, and it is the oldest law of culture. We don't call the novel garbage because airport thrillers exist. We ask what a medium can do at its ceiling, and what its failures reveal. Here is my first surprising claim: the sludge is not the machine's failure. It is a mirror. A model is generic because it is the statistical centre of everything we have written. When it produces a cliché, it shows you the cliché's true share of human language. We spent two centuries producing corporate memos, SEO copy and greeting cards by the billion, then acted shocked when a machine trained on them produced more. It did not invent mediocrity. It industrialised ours. Second, the charge that AI lacks originality assumes we know what originality is. The Romantic idea of the lone genius creating from within is barely two hundred years old. Virgil rewrote Homer. Shakespeare lifted his plots from Holinshed and Plutarch, and nobody called it theft. Eliot said mature poets steal. No human has ever written a sentence without a borrowed grammar and a thousand remembered voices. Human creativity is recombination under pressure: a mind compressing an inheritance and pushing something slightly new out the other end. A language model does something comparable. The mechanism differs, and there is no childhood, no grief, no body. But if originality means a novel arrangement of inherited material that surprises and illuminates, the machine is not outside the tradition. It is the most literal version of it ever built: a creature made entirely of culture. "But it means nothing by what it says." Consider what you have ever had access to when reading a poem. Not the poet's mind, only the words, from which you build the meaning yourself. Barthes pronounced the author dead sixty years ago and the literary world cheered. Now the author has died in earnest, and we suddenly demand his presence. You cannot hold both that meaning lives in the reader and that a passage turns worthless once you learn nothing suffered behind it. The passage has not changed. Only your story about it has. Third, history. The lazy version says every technology was feared and then accepted. But sometimes the critics were right. Plato warned that writing would erode memory, and he was correct; we lost something. What critics consistently got wrong was the standard. They judged each new medium by the old one's purposes. Photography was dismissed as mechanical copying, a threat to painting; it freed painters from realism and gave us Impressionism. Early recording was derided as a dead imitation of the concert hall; then artists made the studio an instrument, and we got Pet Sounds, dub and hip-hop. Early film was filmed theatre for thirty years before anyone understood the cut. We are in the filmed-theatre phase of AI, using it to make the blog post and the stock image, and complaining that it does so blandly. The interesting work lies elsewhere: the writer who spends three hundred iterations interrogating a model until it yields a sentence neither would have reached alone; the translator who hears a hundred renderings of one line; the poet who treats it as an alien reader that misreads her in revealing ways. The genius, for now, is in the encounter rather than the output. Fourth, the real danger. This is not purely like photography. Earlier technologies automated the hand or the eye. This one touches language, the thing we thought was ours, at near-zero cost. Homogenisation is a genuine risk: if every draft begins at the same statistical centre, the strange minority voice could drown in fluent noise. And a confident hallucination is worse than a clumsy truth. But these are arguments about distribution, incentive and use. They call for editors, standards, labelling, and the old discipline of refusing to publish what isn't good. They do not show the underlying capacity is garbage. A press can print slander and it can print Spinoza; we did not condemn the press, we built libraries, criticism and copyright. The task is to build that apparatus again, faster. So, the proposition. Garbage teaches you nothing. Genius shows you something you did not know was possible, including about yourself. An instrument that reflects our language back at us, surprises its users with connections no one made, and forces us finally to say what we mean by "creative," is closer to genius than to garbage. It unsettles us not because it is empty, but because it is not. It is full of us. Learn to play the instrument. Thank you.