There’s a lot of discussion about the idea of using AI/LLMs for tarot readings. People have opinions about it, which is their right to express. But from my POV, I believe in three things about the subject:
- It’s going to happen, and keep happening.
- The tools involved are fundamentally flawed.
- People who strongly prefer that their fortunes be interpreted by AI unconsciously align with a narrative that will inevitably flatten them.
Let’s get the obvious out of the way: fortunetelling has its perks. People pay you with money, attention, time, or a mix of all three. The corpus of knowledge with most divinatory systems (Hellenic astrology, tarot, and Arabic geomancy, just to name a few) is mostly available online in a form that has been scraped by at least one company to its marrow. The Venn diagram of where the two meet is pretty easy to draw, equally easy to optimize. It’s entirely possible for someone to build a workflow in something like Claude or Kimi to take a question, generate a spread, and spit out an answer in an afternoon. If you want to get really fancy, you could even use distillation methodologies to create an entire local-first knowledge base to draw from – creating the front end of such an effort would be peanuts (so would be future-proofing it). Once you have it set up, you can charge $3 for simple reads, $10 for big reads; just a few of these per month would pay for your hosting costs, and the rest would be easy income.
But that means avoiding the fact that the tools are fundamentally flawed with this approach.
Someone I follow online once said that divination was not just looking up definitions on a spreadsheet. Professor Charles Porterfield has remarked that Tarot and playing card divination began from a state of play, an idea that my mentor Gordon White agreed with (G would also say that divination is co-creating with the universe in a playful manner, as memory serves). The first flaw that using LLMs for divination is that the entire process reduces it entirely to its barest definition and prevents play from ever happening – no intuition is present, only the dry bones of data. How can the universe surprise you with miracles and omens, then?
The second flaw lies with how LLMs are created: polite to the point of sycophantly supporting your every question. (We’ll overlook the other issues that their origin story has for this article, but suffice to say that it’s a lot.) When it works well, an LLM will never directly challenge your opinion – it might make you a bit uncomfortable, but that’s about it. Unless you heavily modify your preferences, its goal will be to satiate your curiosity just enough to make you come back for more. The opportunity for you to learn, to be inspired, to draw intuitively from your allies is lost.
And then there’s the last concern.
If you want AI to be the main driver in interpreting your fortune, then you must sit with the idea that at some level, not only do you not trust your own judgment and your own ability to learn – but you want someone else’s model to be the only model that matters for everyone. Does the machine really know best? No, not really – but by putting your faith in it, you have already said as much. By this act, you have erased your own opportunities to learn before you get started. Who do you become, then? Just another statistic in a userbase who might tell their silicon genie that you are the only authority in the subject?
That, dear reader, is a terrifyingly boring world to contemplate.