The Man Who Continued Sherlock Holmes Now Uses ChatGPT

There is a small irony lodged somewhere between the man's résumé and his recent confession. Anthony Horowitz — the author authorised by the Doyle Estate to continue Sherlock Holmes, the novelist trusted by the Fleming Estate to resurrect James Bond — has admitted he uses ChatGPT in his writing. Arthur Conan Doyle, of course, is not available for comment. Horowitz made the admission in what reads as a fairly unguarded moment, and the reaction was predictable: somewhere, a number of literary purists are rearranging their bookshelves in protest. Others — probably the majority — shrugged, opened their own AI assistants, and asked them to summarise the controversy. To be fair to Horowitz, the question of what "using AI in writing" actually means is more interesting than most people allow. There is a considerable distance between asking ChatGPT to list the names of London streets in 1889 and asking it to write your climax for you. Research assistance, continuity checking, early-draft brainstorming — these are the administrative tasks of authorship that readers never think about. If Horowitz is using a language model to speed up the parts that would otherwise involve him at a library desk with a magnifying glass, it is difficult to mount a principled objection. What makes it odd — specifically odd for him — is the symbolic weight of the estates he carries. When the Doyle family handed the keys of Baker Street to Horowitz, they were making a statement about craft, about fidelity to voice, about the irreducible act of listening to Conan Doyle on the page and trying to write back in the same frequency. That act — call it ventriloquism, call it devotion — seems, in theory, rather hard to outsource to a programme trained on the entire internet, including, presumably, the less distinguished Holmes pastiches. Whether it matters in practice is another question. Horowitz's Holmes novels stand on their own merits. His Magpie Murders — with its nested structure and gleeful self-awareness — was as purely human a puzzle box as contemporary crime fiction has produced. No reader of those books would accuse them of feeling generated. And yet the conversation keeps circling back to the question of degree. Using ChatGPT for a plot timeline is rather different from using it for a sentence. The sentence is where any author is most recognisably themselves — the cadence, the syntax, the particular calibration of menace in an interrogation scene. These things cannot be prompt-engineered without the result being immediately obvious to anyone who has read the original. The interesting test, as always, is not whether AI was used, but whether you can tell. The genie is out of the bottle. Authors across the industry are working out — quietly, privately, and occasionally in public, as Horowitz has now demonstrated — what they will use these tools for, and what they prefer not to examine too closely about themselves. It is less a crisis than a renegotiation. What literature requires from its authors, and what, precisely, we are paying for when we buy a novel: these are not new questions. They are arriving now with rather more urgency than before. What would Holmes make of it? Probably something arch about the importance of never theorising ahead of one's data. He might want to examine the model's training set before drawing conclusions.