“The Artificial Unconscious” by Germán Sierra argues that classical psychoanalysis’s linguistic “splitting” of the subject (Freud/Lacan) meets a counter-tradition of multiplicity and machinic unconscious (Jung, Deleuze/Guattari), and that the latter maps more productively onto today’s AI. Sierra distinguishes tightly scoped, rule-bounded systems from open generative models, noting that “training” only optimizes for the probable while complex systems also yield spontaneous patterns the designers did not intend. In this frame, humans function as the models’ environment and feedback loop; representational aesthetics falter in hyperreal sign circulation; and large language models can behave like a cyberpositive, collective artificial unconscious—producing slips, hallucinations, and mythic recombinations—without ever achieving a reflective second-order self, where credibility is measured less by truth than by fluency. Sierra shifts the question from whether AI “desires” to how learning machines evade instructions and generate novel chains of signifiers from human corpora. Generative media are processes rather than knowledge; speculative art becomes inquiry into those opaque processes, edging toward xenophenomenology/xenopoiesis and a nonhuman unconscious (e.g., Siratori’s xenopoetics).