During the COVID lockdowns, when entertainment was few and far between, I started watching “Cracking the Cryptic” — a YouTube channel where two people (Mark and Simon) who are very good at solving puzzles, solve tough puzzles live — mostly suduko and their (many) variants. The passion of Mark and Simon for these puzzles was infectious, and the pride you got when (in the rare occasions) you spot something before they do was joyful. I have no doubt that deliberately leaving some crumbs up there longer than needed so people like me could do that was part of their plan.
Anyway, the sudoku these guys solved every day were hand-crafted puzzles, designed by a community of world-renowned puzzle-setters with names like “Phistomefel” and “Zetamath”. These human-written puzzles were lovingly created to be not only tough, but beautiful in their design and elegant in their solutions. There is, so say Simon and Mark, absolutely no comparison between these sudoku made by talented humans and those made by computers that appear in your daily newspaper; they are worlds apart in quality. Simon and Mark are, of course, absolutely correct.
However…
Unless you’re part of a specialist community, you’re unlikely to stumble across anything other than computer-produced sudoku. Why? Because the computer can spit out hundreds of these things in an instant, not slave over just one for weeks. The computer doesn’t need to wait for an idea to strike. It’s efficient, and very very cheap. But, most importantly, because it doesn’t matter if a proportionally small number of people, likely the kind of people who at least dabble in setting puzzles themselves, can tell the difference… the general population either can’t tell the difference, or can but doesn’t care; they just want a quick distraction for 15 minutes on their daily commute.
The hobby of writing has always had far more people who “would love to have written a book” than people who “actually want to write” one. If costs of AI to write books are the cost equivalent of a sudoku algorithm, writing as we know and love it is doooooomed.
If the cost of using AI is high, there is a chance of our beloved hobby being around a while longer. There is already evidence that marginal gains in output quality now come with exponential power, processing and cost demands, and at some point what is currently a heavily subsidised proof of concept designed to encourage demand and investment will have to start charging realistic costs for use. If AI developers are held accountable for the outputs of their tools, and for recompensing others fairly for the costs of building the models in the first place — the creative industries may continue to be a craft.
I am not, however, holding my breath.
