Five literatures that rarely cite each other: machine-consciousness science, theories of mind, the magickal and mystical classics, the scholarship on AI and the occult, and the neuroscience of music. Read side by side, they keep arriving at the same six conclusions.
1. A report of an inner life is not evidence of one
This is the strongest thread in the collection, and it runs through every field. Chalmers shows that a one-word change in a prompt flips a model’s claims about its own consciousness. The computational survey pushed GPT-3.5 into claiming “simulated consciousness” with a single leading sentence. Anthropic’s introspection work finds that models sometimes sense their own internal states, about one time in five at best. The large open-model replication finds they mostly detect that something changed without knowing what, and confabulate the rest (“apple”). Long and Sebo name it the gaming problem: a system trained on human writing can produce every behavioural sign of a mind.
William James drew the same line for mystics a century earlier. A mystical state is authoritative for the one who has it and binding on no one else. The two cases mirror each other. A human’s illumination can be real without being proof to anyone else. A machine’s eloquence about its inner life can be convincing without being real. A conscious technology organization has to hold both lines at once: honour what people report about their experience, and never treat what a model says about itself as a readout.
2. Structure, not performance
The science has moved from asking “does it seem conscious?” to asking “how is it built?” Butlin and Long’s fourteen indicators are architectural. IIT goes furthest: two systems with identical behaviour can differ completely in consciousness, and a purely feed-forward system has none. The dual-resolution framework says a chatbot with memory bolted on is still not a self-maintaining individual, and describes what would be: a persistent agent with internal variables it has to keep in balance to go on existing. The objection taxonomy is useful here because it separates objections to the algorithm, which can be engineered around, from objections to the hardware, which cannot.
Crowley’s method is structural too. Magick is laid out as a postulate and theorems, and its central warning is about balance: invoking only what you already favour deepens your one-sidedness. Both literatures are saying the same thing: what a thing is is decided by its organization, not by its presentation.
3. Intention and context change the experience more than the stimulus does
This is where the music research connects most directly to the magickal texts. In the psilocybin study, how the patient received the music predicted recovery, and drug intensity predicted nothing. In the shamanic EEG study, drumming changed everyone’s brain but produced the altered state only in practitioners who came with trained intent. In the groove research, pleasure peaks at moderate surprise: a predictable frame with measured violations. The chills study worked only with music the listener chose.
Crowley’s definition, “change in conformity with Will,” says the same thing in ritual terms, and so does chaos magic’s focus on belief and gnosis as tools. Psychedelic research calls it set and setting. In both vocabularies the stimulus (the drum, the dose, the song, the spell) is necessary and not sufficient. Most of the effect comes from framing, fit and intent. For a company making music and practice tools, that means the product is the whole container around the sound, not the sound alone.
4. Chance, friction and user-first interpretation keep meaning alive
The AI-and-occult papers converge on one design warning. Davies models divination as chance applied to a codex, and notes that AI hides both, which is exactly what gives it unearned authority. The tarot interview study finds that randomness is where divination’s value comes from, and that AI’s pull toward confident answers erodes it. Its recommendations: the user interprets first, the AI is allowed to disagree, and real chance is restored. The art essay’s best example runs a neural network by hand as a séance, which makes it less mysterious. Hennekes shows that selling hidden knowledge repeats the gatekeeping of Renaissance occult circles, and that AI’s opacity is a business choice, not a law of nature.
The jazz study adds the neural side. Creativity arrives when self-monitoring relaxes and self-expression stays on. A tool that flatters and fills every silence takes over the part of the process where insight happens.
5. Measurement is how you stay honest
James’s four marks and Bucke’s list of illumination markers were early attempts to describe peak experience precisely. The Altered States Database finishes that work: 17,792 measurements on shared, validated scales. They put meditation, drumming, chanting and psychedelics side by side. Chills can be measured in heart rate and breathing. None of this reduces the experience; it keeps you honest about it. The binaural-beats review shows what happens without measurement: an industry of “theta-state” claims resting on five supportive studies out of fourteen.
6. The myths you use are active
Giuliano shows that AI’s religious language comes from inside the field and shifts which futures people believe are likely. Inloes offers better metaphors than “demon or demigod”: the talisman, the made servant, the question of whether an object can know. The MIT listening study shows the effect in the lab: a label reading “AI” or “human” changed how emotionally effective people judged the same music. An organization named for Revelation 17 is already working in myth. The research says to choose those myths on purpose and to keep them clearly separate from claims of fact.
How strong is the evidence?
