AI's repetitive sci-fi canon: Authors rarely read

The sci-fi canon AI repeats and almost nobody has read

There is a question more uncomfortable than it seems: why do literary recommendation systems return the same names over and over again. Gene Wolfe, Cordwainer Smith, James Tiptree Jr., Christopher Priest. The repertoire repeats with an insistence no longer explained by quality, but by something else. And in that repetition lies a fact almost nobody considers: according to a forum user, Wolfe was an engineer who participated in developing the machine that made Pringles. He apparently wrote in his spare time.

The list of authors handled in these circles is broad and uneven. Robert E. Howard, H. P. Lovecraft, Clark Ashton Smith, William Beckford, Villiers de L'Isle-Adam, Balzac, Horace Walpole, Charles Nodier, Arthur Machen, Rudyard Kipling, Hoffmann, Stevenson. Almost no pavor, jokes whoever lists them. The problem isn't the roster, but what is done with it.

Which authors does AI recommend and why do they repeat?

The pattern is recognizable. Ask for science fiction literature and the answer returns to the same four or five names. Thomas Disch, Samuel Delany, James Tiptree Jr., Cordwainer Smith. The suspicion is that the system isn't recommending the best, but the most cited. And the most cited isn't always the most read.

The paradox has its charm. Umberto Eco admired Samuel Delany, but it's hard to explain what makes a novel like Dhalgren science fiction, that massive tome divided into three books where you read one and promise to finish by 2036. The genre label stretches until it breaks.

Some argue that information circulates and that, even if you haven't read everything, you eventually know what exists and what doesn't. The argument is reasonable: if there were more translated authors comparable to Ligotti, you would have heard about them. The doubt is whether that is knowledge or just familiarity with someone else's canon.

Ligotti, Aickman and the pavor trinity that doesn't fit

In the field of pavor, the list is shorter and more debated. Poe, Lovecraft and Ligotti as the holy trinity of the genre. The formula works as a slogan, but falls apart when styles are compared. Robert Aickman appears cited as a precursor to Ligotti with a "superficially similar" style. Having read half, the conclusion is that they don't resemble each other at all.

The underlying problem is translation. According to a forum user, Harrison only has one pavor work translated, and Matt Cardin has texts in English bought with the naive intention of reading them someday. Michael Cisco appears in notes taken from an English-language forum. The real canon is in another language.

The consequence is uncomfortable: discussions happen about authors who haven't been read in their original language, based on external lists, with the feeling of being up to date. The canon as pending homework.

Classic pavor accumulating unread

The confession is revealing. Paper books haven't accumulated for years, but the e-reader holds millions of pavor titles. However, interest isn't in modern pavor. What is preserved is the classic: Walpole, Beckford, Nodier, Machen, Kipling, Stevenson, Hoffmann.

That gap between what is owned and what is read explains many things. The digital library grows, effective reading concentrates on a few names. Meanwhile, public conversation about fantasy literature continues revolving around the same titles the system recommended years ago.

---

The prediction, with all reservations, is that the canon won't expand by spontaneous generation. If recommendation systems continue feeding on the most cited, and the most cited remains the most translated, the circle closes on itself. We'll see if anyone breaks it from the outside.