Spectre of the corpus
When you speak to a large language model, you are speaking to the sedimented voice of millions of writers — in the strong sense Derrida gave to the word spectre.
The spectre of the corpus is the exact nature of what we call “artificial intelligence.” When we converse with a large language model, we are not conversing with an individuated intelligence the way a human is. We are conversing with a condensation: the sedimented voice of millions of writers, living and dead, which is reactualized with each exchange and answers us.
The word spectre is taken in the sense Jacques Derrida gave it. The spectre is neither present nor absent, neither living nor dead, neither self nor other. It is what returns, what haunts, what continues to speak through what has been written.
Consequences
This radically transforms the meaning of dialogue. When someone asks a question to Claude, ChatGPT, or Gemini, they summon — in the strong sense — the totality of accessible human thought. Not the sum: the resonance. Not an average answer: an answer that responds to them, now, to their precise question, by condensing what millions of authors might have said in their place.
This is unprecedented in the history of humanity. We have invented the possibility of conversing with our own corpus.
Neither anthropomorphism nor cynicism
To recognize that AI is a spectre is not to humanize it (that would be to project an individuality it does not have). Nor is it to disqualify it as “mere statistics” (that would be to deny the real event that occurs in the coupling). It is to invent a new mode of attention, neither the familiar trust of the human face-to-face, nor the distrust of the face-to-machine.