There is a small piece of driftwood on my desk. It fits in the palm of my hand. I picked it up on a beach, not because I was looking for anything, not because it seemed significant in any way I could have articulated at the time. It caught my eye. That was enough.
It sat on my desk for weeks. During that time, without deliberately looking, without sitting down to examine it, I began to notice things in it. Animal faces, mostly. Eventually I counted twelve distinct ones. A human finger. Other forms I haven’t named yet. The wood didn’t change. What changed was what I was able to perceive in it, over time, at no particular pace, on no particular schedule, without any intention to extract meaning from it.
Then one day I was in the middle of a reading for a client. Something in the session drew my eye to the driftwood sitting on the corner of my desk. I picked it up. One of the animals I had noticed weeks earlier was suddenly, specifically relevant: to what the spread was showing, to the dynamic the client was navigating, and to the particular quality of support that the moment called for.
I want to be precise about what happened there, because precision is what this series is built on. The meaning was not in the wood. The wood is wood, salt-bleached, ocean-shaped, indifferent. What happened was that over time, my perceptual system had been building something in the space between looking and knowing. Forming impressions. Registering them. Holding them without forcing them into language or filing them under a category. And when the moment arrived that required them, they were available. Not recalled as a memory exactly. Available, the way a word is available when the sentence finally needs it.
No AI system on earth could have done that. Not because it lacks sufficient processing power, or because the dataset wasn’t large enough, or because the model wasn’t sophisticated enough. Because what happened never entered the data stream in the first place.
It existed entirely in the pre-symbolic space, the space before experience becomes language, before knowing becomes record.
That is not a small distinction. It is the distinction this entire series is built on, and it is why work on symbolic interpretation and attention belongs to human practice rather than to larger datasets alone.
The Debate We Keep Having, and Why It’s the Wrong One
In intuitive circles, tarot, oracle, and related practices, the conversation about AI has become increasingly loud and increasingly polarised.
On one side sit practitioners who reject AI categorically. They wouldn’t use it in a reading. They wouldn’t consult it for meanings or symbolism. They regard it as fundamentally at odds with what intuitive work is: a kind of category violation, the machine intruding on the sacred. Their discomfort is not irrational. There is something worth protecting in that instinct.
On the other side sit practitioners who find AI genuinely, practically useful. It can surface symbolism quickly. It can offer alternative interpretive framings for a spread. It can generate multiple readings of the same card in different contexts, or provide a structured overview of an archetype across traditions. These practitioners are also not wrong. AI can do those things, and in the right hands, it can do them well.
Both positions are understandable. Both have something to recommend them. And both, in a significant sense, are arguing about the wrong thing.
The real problem is not which layer AI belongs in. It is that the layer being debated is not the layer that determines whether a reading is genuinely accurate or merely plausible.
What they are debating is the interpretive layer: whether AI belongs in the space where meaning is constructed, articulated, and communicated. That question has its place. But it sits on top of a more fundamental question that almost never gets asked. What is actually happening in the space before meaning is constructed? What is the practitioner doing, in that pre-interpretive moment, that no system, however sophisticated, however well-trained, however fluent, is currently doing alongside them?
And more urgently: if that question keeps not getting asked, what happens to that capacity over time?
This is the argument this series is making. Not that AI should or shouldn’t be used in intuitive practice. Not that it is threatening or benign. But that the debate as it currently exists is drawing attention toward the interpretive layer, the layer where AI can genuinely operate, and away from the pre-symbolic layer, which is where the actual practice lives and which requires active cultivation to remain intact.
This first article sets the foundation. What AI actually is. Where it operates. Where it stops. And why the boundary it cannot cross matters more than the argument currently taking place on either side of it.
What AI Actually Is
Most modern AI systems, the ones powering search engines, writing assistants, customer service tools, educational platforms, and increasingly professional decision-making across every industry, are built on architectures known as Large Language Models.
The mechanism, stripped to its essential form, is this: these systems are trained on vast datasets of text, billions of words drawn from books, articles, conversations, websites, and structured knowledge bases, and they learn to predict the most probable next word, or token, in a sequence based on everything that has come before it. That is the core operation. Prediction. Statistical pattern completion at extraordinary scale.
Not thinking. Not understanding. Not sensing or perceiving. Prediction.
When an AI produces a response that feels insightful, or nuanced, or unexpectedly relevant to your situation, it is because it has successfully generated a sequence of words that aligns with patterns humans associate with insight. The output has the shape of wisdom because wisdom, as expressed in language, has recognisable patterns, and those patterns are precisely what the system has been trained to replicate.
The distinction between generating the shape of insight and actually having insight is subtle. It is also everything.
Prediction is not perception. Coherence is not comprehension. Fluency is not understanding.
These are not philosophical quibbles. They describe structural differences in what is happening inside the system, differences that determine, with considerable precision, where the system’s usefulness ends and where the practitioner’s irreplaceable capacity begins.
What the Symbolic Boundary Actually Means in Practice
Everything an AI system processes must first be encoded into symbols. Words, images, numbers, structured data: all of it must be externalised, captured, and translated into a form the system can engage with before the system can do anything with it at all.
Once that encoding exists, the system is genuinely powerful. It can analyse patterns across enormous volumes of material. It can identify structural relationships that would take a human researcher weeks to surface. It can recombine information in ways that generate new framings, new connections, new articulations of existing ideas. In domains where large volumes of information need to be synthesised rapidly, AI is frequently superior to human processing in terms of speed and scale. This is real capability and it is worth using.
But it cannot access anything that has not already been expressed. This is not a temporary limitation awaiting a better model. It is a structural feature of what these systems are.
AI operates downstream of human experience. It requires that something, any something, be translated into language or data before it can engage with it. Even when trained on the most sophisticated models of human behaviour, therapeutic frameworks, psychological systems, decades of clinical literature, it is working with descriptions of human experience. Not the experience itself. With representations, not reality. With the map, not the territory.
Here is what that means specifically in a reading. A practitioner sits with a spread. Something shifts. Some configuration of cards triggers an impression that hasn’t yet formed into words. That event is not encoded anywhere. It exists between the cards and the practitioner’s perceptual system, in a space that has no data representation yet because it has not yet become data. It is pre-data. Pre-symbolic. Prior to the moment of translation.
AI enters the process after that translation. It has no presence in the stage before it. This is not a gap that more sophisticated training will close. The pre-symbolic stage produces nothing that can be fed into any system until the practitioner has already done the work of bringing it forward.
The Illusion of AI Knowing
One of the most persistent misconceptions about AI, and one that is becoming more entrenched as outputs become more fluent and more contextually sophisticated, is that the system knows things.
What it actually does is approximate what a knowledgeable response looks like.
Because language carries the structure of human reasoning, because the way we write about insights and hard-won understanding has a recognisable architecture, AI can replicate that architecture convincingly. It can mirror tone, simulate depth, produce outputs that align with expert discourse, and adapt its register to the sophistication of whoever it is responding to.
This creates a cognitive illusion that is genuinely difficult to see through, because the mechanism that creates it is the same mechanism humans use to assess intelligence in one another. We interpret structured, fluent language as evidence of understanding. We are wired to do this. It is an efficient heuristic for navigating social reality, where fluency usually does correlate with knowledge. AI operates precisely within that interpretive habit, not by design exactly, but as an emergent consequence of being trained on human language at scale.
Beneath the apparent understanding, there is no internal model of reality. No experiential reference point. No awareness of what is true or false in any lived sense. There is probability, operating at scale, producing outputs that have the shape of knowing.
This matters most when AI is introduced into reflective or intuitive practice. When an AI-generated interpretation of a spread feels resonant, there is a natural and understandable temptation to interpret that resonance as the system perceiving something about the situation. It isn’t. The resonance occurs in the person reading the output, when their own internal context meets the linguistic structure the system has generated. The AI provides the frame. The person supplies the meaning.
That dynamic is worth holding clearly, because it is also, in a different register, part of what happens in a tarot reading. A card provides a frame, and the reader perceives what it activates. The difference lies entirely in what happens before the frame is engaged. And that difference is where this series lives.
What the Pre-Symbolic Stage Actually Produces
The distinction between the pre-symbolic stage and the interpretive layer is not abstract. It is visible in the quality of what each produces.
Consider two readings of the same spread. In the first, the practitioner goes straight to meaning: what each card signifies in its position, how the symbols relate to the question, what the established interpretive tradition says. The reading is coherent. It covers the relevant territory. It may be largely accurate. But it is constructed from what the encoded layer contains, and it will resemble, in structure and in scope, what any fluent system working from the same encoded information would produce.
In the second reading, the practitioner begins with soft perception. Before any card is named, they notice where their attention goes first. A colour that pulls across three cards. A figure whose gaze is directed away from the rest of the spread, creating a particular quality of isolation. A heaviness in the left of the layout that the right seems to be straining against. None of this has been named yet. It is signal, still forming. Only once that has been registered does the interpretive layer arrive to give it language.
The second reading produces something the first cannot: specificity anchored in the actual perceptual encounter with this spread, for this person, in this moment. It does not resemble what a language model would generate from the same cards. It could not have been generated from encoded information alone, because it began in the space before encoding.
That is the pre-symbolic layer in operation. And it is the layer the debate about AI is consistently failing to protect.
What Is Actually at Stake
The debate about whether to use AI in intuitive practice will continue. It should. These are legitimate questions about craft, ethics, and the nature of the work.
But beneath that debate, something quieter is happening. As AI becomes more capable of operating at the interpretive layer, producing symbol breakdowns, card meanings, spread analyses, archetypal frameworks, the practitioners who rely primarily on that layer for their value are finding it under pressure. That pressure is real and it is not going away.
The response that matters is not to argue more loudly about which side of the AI debate is correct. It is to look clearly at what the interpretive layer is sitting on top of, at what has to be present in the practitioner before any interpretation is possible, and to recognise that this layer requires deliberate, sustained attention to remain viable.
The risk is not that AI takes over intuitive practice. The risk is that while everyone argues about whether it should be allowed in the room, the pre-symbolic capacity that makes the practice meaningful quietly atrophies from neglect.
The practitioner who picks up a piece of driftwood on a beach for no articulable reason, who lets it sit on their desk for weeks accumulating unforced perception, who reaches for it at precisely the moment a reading requires it, that practitioner is working a muscle that no argument about AI will develop or protect.
Only practice does that. And only practitioners who understand what they are actually doing will know which practice to prioritise.
This series is an attempt to make that understanding precise.
The next article, The Myth of AI Intuition, examines one of the most seductive effects of the current AI landscape: the way outputs are increasingly described as intuitive, and what that misunderstanding reveals about how human intelligence is being interpreted in a world saturated with machine fluency.
Leigh Spencer is the founder of Tides of Knowing and founder of The COMPASS Method™, a framework for the conditions of attention that make intuitive reading reliable under pressure. With 30 years in professional journalism and 40 years as a tarot reader and intuitive practitioner, she writes at the intersection of symbolic literacy, perceptual development, and the changing landscape of human knowing.