When AI writes the words, who is the author?
Fielding, Dean of Science at Stellenbosch, submitted an opinion piece to Nature Africa and was told by the handling editor that two AI detectors had scored it as entirely AI-generated. He concedes the detectors were right about the prose and that AI functioned as his ghostwriter, then separates writing from authorship from thought leadership, arguing that expertise, direction, judgment and accountability remained his. Argues against detector scores as verdicts, for richer disclosure that specifies what the tool did and what the author did, and against the assumption that polished student writing indicates dishonesty. Converges with the accountable conversant from the author's side: he lands on accountability, not output quality, as what settles the question. Departs on two points. He treats the exchange as a production site for a document rather than a mediating site, and he assumes the human directing it brings prior expertise, which is the case least threatening to the inherited frame. His own account undercuts him: weeks of iterative pushback is precisely the transactional mode, and he cannot certify from inside it that the thinking was untouched. Most useful as a documented instance of the verification apparatus reading the artifact correctly and learning nothing about the exchange that produced it. Nature Africa comment, 18 August 2026, free to read.
