AI as psychedelic pattern‑finder

Leading language models should be thought of not as encyclopedic savants but as systems that find unusual, cross‑domain statistical patterns — like a person experiencing synesthesia or a psychedelic insight. Those patterns can yield surprising, valuable creativity but also produce confident, systematic misfires (hallucinations) that resemble the output of altered human cognition. — This metaphor shifts expectations about model behavior and implies different remedies (uncertainty signaling, human‑in‑the‑loop pattern validation, evaluation that tests cross‑domain patterning) for governance and deployment.