Readings with LightSay "AI tarot" in a room of readers and watch the temperature drop. The objection arrives fully formed, and it deserves to be taken seriously rather than argued away: a reading is an encounter between two people, one of whom is holding something they may not be able to say out loud. A machine cannot sit with that. It has no intuition, no hands, no lineage, no teacher who taught it what the Three of Swords costs. Five hundred years of practice, and now a text box.
It is a good objection. It is also, historically speaking, the objection that tarot has faced and absorbed roughly every hundred years since it began.
The cards were a technology before they were a tradition
The oldest tarot decks we have were painted by hand for Italian courts in the 1440s — the Visconti-Sforza cards, gold leaf laid over hand-painted figures, commissioned by a duke. Tarot at that moment was not a folk tradition. It was a luxury object, and it was a card game.
What turned it into something ordinary people could hold was not a spiritual awakening. It was the woodblock press. Cheap printed decks put the trumps into taverns and kitchens across Europe, and it was there, over the following centuries, that the cards drifted from game to divination. Every later leap runs the same way. Etteilla built the first deck designed specifically for divination in the 1780s — and built a business teaching it, which makes him arguably the first professional tarot entrepreneur. The Rider-Waite-Smith deck of 1909, the one nearly everyone pictures, was a mass-market print product: Pamela Colman Smith's drawings, reproduced by a commercial publisher in London and sold to anyone who walked in. The 1970s boom was a reprint boom. The 1990s brought telephone hotlines, which were tacky and enormous. The 2000s brought forums and email readings. The 2010s brought apps.
At every one of those thresholds, someone who loved the cards said that the new thing would hollow them out. The deck absorbed it and carried on, because the cards are unusually indifferent to their delivery mechanism. They were never in the paper.
What is actually behind the screen
Here it helps to be plain, because vagueness is how this subject goes wrong. There is no oracle in the machine. What a system like this one holds is a very large amount of written human thought — including centuries of it about these cards. Marseille and its pip traditions. The Golden Dawn's correspondences. Waite and Smith's iconography. Crowley and Harris arguing through the Thoth deck. Sallie Nichols reading the trumps as Jungian archetypes, Rachel Pollack teaching a generation to read them as narrative, Mary Greer turning the practice inward. Folk meanings, reversals, elemental dignities, the numerology of the pips, and the ordinary human situations the cards have been laid over for five hundred years.
No human reader holds all of that. A good reader holds one or two lineages deeply, which is a different and often better thing — depth beats breadth in a room with one person in it. But the breadth is not nothing. When the Five of Pentacles falls next to the Hierophant, there is a long argument in the literature about exclusion and institutions, about standing outside a lit window, that a reader may or may not have met. The machine has met all of it. That is its one genuine advantage, and it is worth naming precisely rather than dressing up: not wisdom, not sight — recall across a whole tradition, instantly, in five languages, at three in the morning.
Where the meaning actually happens
Now the harder question, and the one Jung turns out to have already answered.
We wrote a while ago about how Jung gave the cards a second soul — how his archetypes and his idea of synchronicity let a randomly drawn card mean something without anyone having to claim it predicts the future. Read that idea closely and notice where it puts the meaning. Not in the cardboard. Not, crucially, in the reader. Jung's account locates it in the encounter: an outer event and an inner state falling into an arrangement the psyche recognises. The card does not know you. The reader does not divine you. Something in you recognises itself in an image, and the recognition is the reading.
If that is true — and it is the working philosophy of almost every modern reader, whether or not they credit Jung for it — then the interesting question about a machine is not whether it has a soul. It is whether it can hold up the image clearly enough for the recognition to happen.
And that turns out to be a question about craft, not metaphysics. Does it name what the card actually shows? Does it stay with the specific question rather than sliding into horoscope fog? Does it say the uncomfortable half of a card as readily as the flattering half? Those are answerable, and they are answerable by reading the thing it produced. A bad reading is bad whoever wrote it.
What it honestly cannot do
The trust runs on saying this part out loud.
It has no unconscious of its own, so if you want the mystical version — two psyches meeting in one field — it is not on offer here. It cannot see your face fall at the word separation. It does not know you have asked the same question in three different disguises this month unless you have kept a journal that tells it. It cannot decide to stop the reading and ask if you are all right. And it will never be the person who has read for you for eleven years and can say you know what this is about.
One more thing, which matters more than any of them: it does not choose your cards. The shuffle happens separately, in a system with no idea what you asked. The machine is handed a card it had no part in selecting, exactly as a reader is handed the card you cut to. Whatever you believe is happening in that moment — synchronicity, statistics, or the quiet usefulness of a random prompt — it is happening upstream of anything that could be accused of telling you what you want to hear.
Adapt or become a museum
Traditions that refuse every new vessel do not stay pure. They stay small, and then they stay historical. The cards have survived the woodblock, the lithograph, the telephone and the app, and they survived each one by being carried into it rather than defended against it.
What deserves defending is not the delivery. It is the standard: that a reading names something true, that it respects the person on the other side, that it does not pretend to know what it cannot know. Those can be upheld or betrayed by a person in a candlelit room and by a machine in a browser tab, and they have been betrayed by both.
So: use it sceptically. Ask it something real rather than testing it with a question you do not care about. Notice whether the reading recognises you or merely flatters you — and if it flatters you, close the tab, because that is the actual failure mode and it is not unique to machines.
The cards have never much minded who lays them. They have only ever minded whether the person looking at them is willing to see what is there. Draw three and judge for yourself; if you would rather it talked back, the readers do that too, and they will tell you plainly what they are.