Ok, here I am really giving away something that I’ve thought of as a thought worth money. I’ve been reworking an essay on this subject for years. Who knows? Maybe I’ll publish it at some point, but my hard drive is bursting with 3/4 finished long-form writing and the basic thought is not doing anybody any good sitting there. Plus, honestly, Christopher Miller’s 2018 book got at one major part of what I was doodling around with, and a lot of what I have to say here today echoes his work.1
You know why we can be influenced by textual AIs? (Well, one reason.) It’s because we’ve been able to be convinced for centuries that textual representation is someone real who is actually fake. Maybe since writing, maybe since the invention of the novel, but for a long time.
It turns out that it is surprisingly easy to write as someone else than yourself and be taken for that person. Here we have to think of the power of two crafts—almost technologies—of written representation that enable writers to be imposters with particular skill.
The first is fiction, and our training as readers in fiction. The novel is the parent of deceptive AI. Certainly there were many texts written that attested to experiences that were fabricated or unreal prior to the appearance of the novel, and there were premodern traditions of critical readership that mulled over the truthfulness of chronicles and traveller’s tales, but they largely did not suppose that the author had created a false self, a persona, merely that the creator of the text had invented or exaggerated what they claimed to have seen or experienced. Those critical traditions often did not get hung up on a point that we today often consider, in fact: a chronicle that attested to an event or an observation that was told to the writer as if the writer had directly witnessed it might be completely legitimate.