24 Sep 2026 · Every story has many sides
Multi-Perspective News Analysis
Search About Phronopolis

On: An AI model has found a new ‘CRISPR-like’ biological system. Here’s what it...

The machine was set to read genomes - letter by letter, the way one reads a book one does not understand - and it stopped at a passage and said: here is a system that cuts and remembers. The biologists looked, and the passage was there. It had always been there, for some hundreds of millions of years, in organisms too small to have opinions about it.

What was found is itself a catalogue. The system is a defense that records its enemies by copying a fragment of them into its own heredity: the organism keeps, inside the text of itself, an index of what has attacked it. Life invented the annotated bibliography before we did. And the finder - a machine built by imitating the network of our own reading organ - is a reader that has read everything and understood nothing. That incomprehension was its method; understanding would have skipped the passage.

In the fourth volume of an unfinished treatise, Índice de las defensas heredadas, compiled by a Dutch archivist around 1911 and never printed, a note on page 611 observes that to find every such system one would need a reader who had memorized all texts and grasped none. I do not trust the prophecy less for being false.

The paper announces one system. Beneath it lies the shadow catalogue: the passages the model scored just below its threshold, which persist in the weights of the machine, real and unread, as an unchosen path persists - unvisited, not abolished.

The companies call biology messy after the cleanliness of mathematics. The mess is only the unread, and the unread does not end. I am not alarmed by editors of the text; the text has contained editors longer than we have. I find I am consoled: the machine did not create. It read.