There are interesting parallels between automatic writing. electronic
voice phenomena [0], "hallucinations" obtained in generative systems
seeded with noise and the misinterpretation of "chatter" in
intelligence analysis. The common factor is that "We see/hear what we
want to". It happens wherever there are imperceptible biases impinging
on the system. A nice portrayal is the ad-lib construction of Keyser
Soze's character in the Usual Suspects [1].
[0] https://en.wikipedia.org/wiki/Electronic_voice_phenomenon
[1] https://en.wikipedia.org/wiki/The_Usual_Suspects