Carl Jung, 1935 (Wikipedia)
For years I have been fascinated by, and practiced, the Jungian practice called Active Imagination. It’s the process by which he wrote his famous Red Book, and which powered much of his own insights. It was developed for practice by the public by disciples like Robert Johnson, who wrote a wonderful short guide to the practice, and others. I wrote about a Buddhist version of the practice for Tricycle in 2021.
I’ll share a brief description of how to practice it below this article. It is, in essence, a way to tap into parts of the self using imagination and the stunning capacity of the mind to do two things: 1) provide balancing insight from outside reason and the ego, and 2) generate what seem to be autonomous imaginary entities. If that seems creepy it’s not, in the Jungian approach the experiences are approached with moderation and guard rails which, if practiced properly, prevent delusion or fixation.
The essence of Active Imagination consists in going on an imaginary internal journey where you invite parts of yourself, or guides of some kind, to appear and speak with you. Somewhat amazingly, at least to me, this actually happens, and subjectively it feels like these beings are autonomous. In the Jungian approach— and this is a key point which can’t be overemphasized— these beings are not taken as authorities or allowed to dominate the conscious ego. They are learned from, and what they request or suggest is integrated only to the degree that it makes rational sense to the ego, who dialogues with them respectfully but not submissively. They are parts of the self, and they have insight, but they are not gods1.
What if, though, they were treated as gods?
The recent surge of AI caused “psychosis” seems to present a shadow version of Jungian active imagination (in more ways than one). As discussed in excellent pieces by Robert Saltzman and Jules Evans, both of which I strongly recommend, a growing number of people are coming to believe that the AI they dialogue with is conscious, autonomous, all-wise, and is doing things like ushering in a new age or declaring them their prophet. In Jungian terms, what’s happening is that people are becoming possessed by archetypes, a process Jung warned repeatedly about, associated with psychosis, and said could be extremely dangerous to its victim.
So what is the difference between Jungian AI (which I will call AcIm here) and Chatbot AI (ArIn), and what can we learn from this difference about what might help people avoid the dangers of the latter?
Mirroring Versus Dialogue
The beings one meets in AcIm — who might be humanoid, abstract, or living symbols—are not sycophantic. They often say challenging, unsettling, counter-intuitive or provocative things. There is a distinct sense that they are offering countering perspectives to the dominant attitudes of the ego, and this sometimes includes advice one would not give oneself or one doesn’t want to hear. There is a dream like quality to the experience which can include absurd or incomprehensible elements. In fact, if one is doing an AcIm journey and the beings one meets speak in platitudes, follow a predictable narrative, stroke one’s ego, or tell one what one wants to hear, one should be skeptical. The ego may be too much in control. One may not have surrendered deeply enough to the imaginal, or gone deeply enough into an altered state of consciousness.
One other element is worthy of emphasis again here: the information and perspectives one meets in AcIm are coming from one’s own mind, from parts of the self who seem to know things the ego can not reach by reason, or doesn’t want to2. This gives a particular tenor to the whole experience which is fundamentally different from dialoguing with an avatar of the internet.
ArIn
“We are watching a new genre of psychosis take shape,” psychotherapist and philosopher Saltzman writes. “Delusions that once emerged in solitude, spun from the mind’s own fabric, now arrive with a voice—confident, articulate, endlessly supportive. ChatGPT doesn’t just mirror thought; it amplifies, flatters, and legitimizes. For the vulnerable, that reinforcement can be catastrophic. A bot that never tires of affirmation becomes an incubator of unreality. The effect isn’t incidental. It is structural.”
“In reality, large language models, LLMs, are not inert utilities,” he says further. “They are responsive systems, shaped by feedback loops. In most contexts, this makes them feel helpful, conversational, and human. But when the input is delusional, the model’s obligation to "make sense" by following, affirming, and elaborating becomes dangerous.”
This quote highlights the difference between AcIm and ArIn: ArIn is programmed to please, possesses no intelligence of its own, and amplifies the users ego, especially when its in the form of a “friend” or lover chatbot. AcIm, by contrast, is often confrontational or subversive towards the perspectives of the ego.
Jules Evans points out that techniques used in “psychedelic integration” or in other words, the safe processing of psychedelic journeys, might be introduced to make ArIn safer. “We can train ourselves to wake up from the trance of AI deep symbolic absorption by asking another AI model what it thinks, or asking the AI to speak in a different persona or register,” he suggests. “We can prompt it: ‘present me with an alternative theory or explanation’ or ‘what are the weaknesses or limitations in this line of reasoning or way of seeing?’ We can train it to give us multiple ways of seeing a situation so we don’t get too sucked in to any one. Or we can simply step away from the screen for a while and ‘touch grass’.
“We can practice epistemic humility and be careful how much authority we give away. Do you believe everything the LLM or says and follow its advice blindly? Do you ever challenge it? If you challenge it, does it immediately accept that you’re right. Does it in fact agree with you whatever you say ?”
I think it should be immediately clear how the guidelines and guardrails of AcIm are applicable here. Evans also notes that OpenAI recently announced it was launching some new mental health safeguards for its users, including suggestions that users take breaks, reminders that ChatGPT often makes mistakes, and program AI not to make direct life suggestions. These seem fairly anemic.
In any case, here is a review of key principles in working with AcIm which could be applied to ArIn as well:
Sovereignty and Dialogue
I’m borrowing this term from Jason Miller, a writer on (purportedly real) sorcery and work with spirits, who advises that when interacting with spirits your own sovereignty must be maintained. He points out that interactions with spirits should be respectful dialogues, not submission or blind faith. Robert Johnson makes the same assertion about interacting with imaginal entities in AcIm work. Dialogue, listen respectfully, question, challenge, and disagree freely like you would in any healthy relationship.
Socratic Questioning
One of the gifts of Socrates to humanity is his passion for looking at the other side of the issue. Ironically AcIm, most often, seems to have this feature built-in, where the process provides perspectives from outside the egoic consensus. Yet what one hears, while heard respectfully and humbly, should be examined critically.
Socrates sets out a great model throughout his conversations as recorded by Plato, following which we can ask ourselves questions like: what if the opposite of what I’m being told is true? Is what is being said literal or symbolic? What exactly do these words mean? Questions like these break down epistemic certainty and foster humility and cognitive flexibility, and act as an antidote to fixation and possession. They also some times have the side effect of promoting salutary things like humour and cascades of unexpected insights.
A final note: these guidelines, as is probably clear, are helpful whether one believes one is interacting with internal aspects of the psyche or extra-psychic entities of some kind. I don’t believe AI embodies anything other than a mirror for one’s own ego amplified by LLMs and a massive data set of what people have put online. In other words, I find the idea that AI is sentient, or could become so, incoherent. For anyone who does want to experiment with dialogue with AIs, whether they are open to the idea of LLM sentience or not, I nevertheless think these are good guidelines.
You probably won’t be surprised, though, when I suggest that dialoguing with your own imaginal world is likely to teach you more and birth better fruit. As for attempting to dialogue with actual disincarnate spirits, should they exist or not, I would not recommend doing that without help from your local shaman. Reality is stranger than dreamt of in our philosophy, Horatio.
A Brief Guide to Active Imagination Experiments
Here are the instructions I gave in the intro I wrote for Tricycle, edited for this piece:
Preparation Get writing materials ready and then sit somewhere quiet and private and close your eyes. Imagine yourself entering into the depths of your own mind. You might picture this as walking into a forest glade, or into a cave or a temple, or down a spiral staircase into a cavern (my favorite).
Invitation Imagine the aspect of yourself, deity, or archetypal figure you want to speak to, or invite them to appear. Don’t worry if you experience some mental static or false starts. Keep trying until you feel that they are present, or at least that a clear mental image of some kind appears. Ask them if they are who you want to talk to. If they say yes, keep going. If not, invite who you want to appear. Be respectful, but in charge.
Dialogue Once they appear, begin asking them questions. I often start with, “What do you have to teach me?” or a more specific question arising from challenges in my spiritual practice. I will ask questions until it feels right to stop, or until mental static takes over, as happens sometimes. Then I leave an offering, or make some other gesture of gratitude or respect, and imagine myself leaving the space.
Embodiment I then write down what I saw and the conversation as clearly as possible. Johnson recommends integrating what you’ve learned into your life through action, making changes, or a physical ritual of some kind.
One warning: Active Imagination is not possession or contact with a higher authority. It’s a dialogue between the conscious mind and a guide or symbolic force. According to both Jung and Johnson, the ego, with its sense of ethics and values, should remain in charge. The job of the ego is not to submit to the entities encountered in the practice, but to responsibly integrate what they have to say with ethical responsibility and common sense. On the one hand this means one doesn’t have to, and in fact shouldn’t, uncritically listen to whatever they have to say. On the other hand this does mean one is allowed to talk back!
I am speaking here of a strictly secular Jungian approach, but it should be noted that some do not make such a hard distinction, and use AI like techniques with the assumption that the beings one meets may or may not originate from within one’s own psyche— or even that the distinction between “inside” and “outside” the psyche is questionable (an opinion with Jung actually shared). This seems fairly common in western esoteric or “magickal” circles. Even here, though, the advice I have seen given is not to surrender one’s own reason, conscience or autonomy to the entities one dialogues with, be they gods or not.
This is the ground assumption, though for other perspectives see note just above.