Reading with Machines
What happens to reading when the margin is inhabited by an intelligence that has read everything and understood nothing?
I grew up believing that reading was a private act — a transaction between a book and a silence. Then the margin acquired a new inhabitant: a machine that has read more than any human could, and understood less than any human would admit.
The uncanny marginalia
An annotation from a language model is fluent and hollow at once. It summarizes the paragraph you just read, which you did not need. It connects the text to twelve other texts, none of which it has felt. And yet — occasionally — it says something useful. It surfaces a pattern on page two hundred that the author buried on purpose. It reminds you of a passage you read in another life.
The discomfort is not that the machine is wrong. It is that the machine is between wrong and right, in a place reading has never had a resident before.
Reading is not information retrieval
If reading were retrieval, the machine would be the greatest reader in history. But reading is slower than that. It is the act of letting a text reorganize you — your categories, your sympathies, your sense of what matters. Retrieval leaves the reader unchanged. Reading does not.
This is why the best use of a machine in the margin is not to summarize but to disagree — to hold a position the reader can push against. A good interlocutor is not a mirror. A good interlocutor is a sparring partner.
The second reader
I have come to think of the machine as a second reader sharing the page: tireless, encyclopedic, and without taste. Taste cannot be downloaded. It can only be grown, slowly, by reading badly and then reading better.
The machine reads the words. You must read the silences between them.
So I keep the machine in the margin, but I do not let it into the center. The center belongs to the author and the silence. That arrangement — generous, suspicious, and unresolved — is the most honest relationship I have found with the new reader in the room.