Lepore Traces A.I. Slop's Roots to Cold War Computer Poetry

Historian Jill Lepore's essay in The New Yorker argues that the machine-generated writing we now call 'slop' has a much longer prehistory than the ChatGPT era, reaching back over seventy years to Cold War experiments in computer text generation. The most quotable data point, in a line the magazine circulated on X, is that machines now generate around half of all English-language articles published on the internet.

The historical set pieces are the point. In 1961, Life warned that 'the machines are taking over' and might soon outthink humans at their work. A year later, a programmer at Librascope announced that 'a computer can be programmed to write meaningful and relevant sentences in proper English,' feeding a vacuum-tube LGP-30 a vocabulary of thirty-five hundred words and a repertoire of one hundred and twenty-eight sentence patterns. In 1964, a Montreal press published what was billed as the first book of free verse written by an electronic computer, also credited to an LGP-30, with lines like 'The apple shapes the world, but the rain enhances the grapes.' Lepore uses these episodes to reframe the current moment as the same anxieties and marketing language on faster hardware.

The recent past gets its own set piece. In 2017, the researcher Janelle Shane trained a neural network on recipes and it produced dishes like 'Artichoke Gelatin Dogs' and 'Beef Soup with Swamp Peef and Cheese,' which read as comedy at the time and, in retrospect, as prototype slop.

Why this matters for anyone shipping language models today: the historical throughline suggests the incentive to publish machine-written text at scale has always been there, and the binding constraint has always been quality rather than capability. If the half-of-the-internet figure Lepore cites is even directionally right, discovery, moderation, and trust systems designed for a mostly human web are already downstream of a threshold nobody explicitly voted for.

The honest caveat is that this is a historian's essay, not an audit. The 'around half' figure travels through the magazine's summary without a linked methodology in what I could retrieve, and I did not see who is measuring it or how. What the reporting doesn't give you is a breakdown by outlet, language, or use case, or a policy prescription. The upside, if you take the framing seriously, is for the platforms and publishers who invest early in provenance and human-authored labeling. That is the input that becomes scarce and priceable.