The Scientific Paper as a Human Artifact

The scientific paper is a compression technology built for one reader: the human nervous system, with its bounded attention, limited memory, and need for low-dimensional input. The same constraint shapes scientific practice, from the one-variable-at-a-time experiment to reducing raw data to a few figures and a short narrative in a standard format. This review asks what happens to that artifact as two of its premises dissolve. Digital infrastructure now lets science store and share raw data, code, and executable analyses rather than prose summaries alone. More radically, artificial intelligence supplies non-human readers and writers: agents that can work over far larger and less pre-compressed records than any person, and that can generate text, experiments, and entire papers. The human-facing paper was already strained before AI, as output outgrew any reader, texts grew less readable, the unit of publication fragmented, and the narrative omitted what was needed to verify it. Drawing these literatures together, this review sets out seven design challenges that AI poses to scientific writing. It concludes not that the paper is obsolete, but that its monopoly is ending, and that its human-facing functions, namely certification, explanation, and understanding, must be deliberately preserved as compression passes to machines.