Diving Into the Unknown — Rapidly Learning Domain Knowledge with AI

 

Imagine you have been assigned a project in a field you know absolutely nothing about. Medicine, law, logistics, or — tarot. Your first instinct is probably "I need to study up." And the first step of that study is usually googling. But googling has structural limitations that few people talk about.

This is the story of a developer who knew almost nothing about tarot and acquired project-ready domain knowledge through AI conversation alone.

What I Actually Knew About Tarot

Let me be honest. When the project started, here is what I knew about tarot: you flip cards to tell fortunes. Each card has a different picture and meaning. The "Death" card is not necessarily bad — I picked that up from a TV show. That was the extent of my knowledge.

I did not know how many cards were in a deck. I did not know what types existed. I did not know how a reading worked. And most critically, I had no idea where to get card images.

The traditional approach would have been to search "tarot card basics," read through Wikipedia, browse a few tarot-focused sites, and compile what I found. In my experience, this kind of research takes at least half a day and leaves behind a messy pile of disorganized information.

The Structural Limitations of Googling

Let me briefly lay out the limitations of googling as a learning method. These are things I have encountered countless times over 28 years in the industry.

First, search assumes you know what you do not know. To search "tarot card structure," you need to at least know that tarot cards have a structure worth searching for. In an unfamiliar domain, you often do not know what to search for in the first place. I call this the "not knowing the question" state.

Second, you cannot control the depth of information. Search "tarot cards" and you get a mix of beginner guides and expert-level interpretation theory. Filtering to just the level your project needs is a task in itself.

Third, you have to build the connections between pieces of information yourself. One tab explains card structure, another explains reading methods, a third covers image copyright. Weaving these into a single coherent context is entirely on you.

AI conversation solves all three of these problems at once.

"How Many Card Images Do I Need?"

After telling the AI about my project idea, my very first question was "how many card images do I need?" A simple question, but it cracked open the entire domain knowledge structure.

Claude immediately laid out the core framework. A tarot deck consists of 78 cards. The 22 Major Arcana run from The Fool (number 0) to The World (number 21). The 56 Minor Arcana are divided into four suits — Wands, Cups, Swords, and Pentacles — with 14 cards each. Those 14 consist of 10 numbered cards (Ace through 10) and 4 court cards (Page, Knight, Queen, King).

Finding this information through Google would have required reading at least two or three pages. In an AI conversation, a single response laid out the entire structure. Better yet, the answer was framed within the context of "you are building a tarot web app," so it skipped the unnecessary historical background and mysticism. It was exactly the level of information a developer needs.

The Learning Flow That Conversation Creates

This is where the real strength of conversational AI learning showed itself. Once I learned the number 78, the next question followed naturally: "So where do I get the images?"

The answer to that question led straight to a critical issue: tarot image copyright. The most famous tarot deck is the Rider-Waite-Smith (RWS) deck. The originals were created in 1909 and are in the public domain in most countries. However, the recolored version published by US Games Systems in 1971 still carries copyright.

Why does this matter? Most tarot card images circulating on the web are from the 1971 recolored version. Using them carelessly means copyright trouble. I needed to source only the 1909 originals — and I learned this at exactly the right time.

How long would it have taken to arrive at this through googling? Search "free tarot card images," browse a few sites, then wonder "wait, is this the original or the recolored version?" Search "RWS tarot copyright," find a blog post about the legal status, read through it. Ten tabs minimum. In an AI conversation, all of this happened within a single natural flow.

Escaping the "Not Knowing the Question" State

The hardest moment in domain learning is when you do not even know what to ask. When you know nothing about tarot, you have no idea what to search for.

AI conversation solves this elegantly. When I said "I want to build a tarot web page," the AI told me what I needed to know by working backwards. Questions came in the form of "you will need to decide these things," and answering those questions naturally revealed where my knowledge gaps were.

For example, when asked "what reading methods will you support?" I first needed to know what reading methods even existed. I simply asked "what are the reading methods?" and got a clear answer. One-card draws a single card for a daily reading, three-card uses three cards for past/present/future, and the Celtic Cross uses ten cards for a comprehensive analysis and is the most traditional layout.

This created a loop: "not knowing the question" leads to "AI's question reveals my ignorance" leads to "immediately filling that gap." Learning happened within the project conversation, with no separate research phase.

Understanding the Structure of 78 Cards

As the conversation progressed, the structure of a tarot deck became increasingly clear in my mind.

The 22 Major Arcana are the heart of tarot. They follow a narrative arc called "The Fool's Journey," starting from The Fool (0) and ending at The World (21). Each card represents a fundamental archetype of human experience. The Magician represents creative power, The High Priestess represents intuition, The Tower represents sudden upheaval, and Death represents endings and new beginnings.

The 56 Minor Arcana deal with concrete everyday situations. Each of the four suits governs a different domain: Wands cover passion and action, Cups cover emotions and relationships, Swords cover intellect and conflict, and Pentacles cover material wealth and practicality. Within each suit, the energy starts, grows, and reaches completion from Ace to King.

Understanding this structure took just a few minutes of conversation. But its impact on the entire project was enormous. When designing the data schema, generating interpretation text, and categorizing cards in the UI — every decision was built on this structural understanding.

The Minefield of Copyright

Let me dive deeper into the copyright issue I touched on earlier. This was an area that would have been genuinely difficult to navigate through googling alone.

The copyright status of tarot card images is complicated. The original RWS deck was designed by Arthur Edward Waite and illustrated by Pamela Colman Smith in 1909. Since it is over a hundred years old, it is in the public domain in most countries. Anyone can use it freely.

The issue is the 1971 edition where US Games Systems recolored the cards. This recolored version holds a separate copyright as a derivative work. And the vast majority of images you find when you search "tarot cards" online are either this 1971 version or something derived from it.

The difference between the 1909 original and the 1971 recoloring is subtle. The color tones are different, and some details have been modified. Unless you are an expert, they are hard to distinguish. Discovering this naturally within the conversation flow was a fortunate break. If I had started coding first and sourced images later, I very likely would have unknowingly used copyrighted images.

The solution AI pointed me to was clear: use the 1909 originals, sourced from reliable places like Wikimedia Commons or the Internet Archive. That single piece of guidance saved me hours of headaches during the image sourcing phase.

Upright and Reversed: A Deeper World Than Expected

I also learned during this process that in tarot readings, each card is interpreted in two orientations: upright and reversed. It is not simply "good meaning / bad meaning."

Upright represents the card's energy flowing naturally. Reversed represents that energy being blocked, excessive, or distorted. For instance, The Magician upright means "creativity, willpower, focus." Reversed does not mean "incompetence" — it means "deception, wasted talent, manipulation." The ability is there, but it is being channeled in the wrong direction.

This concept had a direct impact on data design. Each card needing separate upright and reversed interpretations meant that 78 cards actually required 156 interpretation texts. Discovering this early in the project versus discovering it midway would have been a massive difference in workload estimation.

Core Principles of AI-Powered Domain Learning

Looking back on this process, there are several core principles for using AI as a domain learning tool.

First, learn within context. Asking "tell me about tarot" versus "I am building a tarot web app — what do I need to know?" returns vastly different results. The latter gives you information at exactly the depth your project requires, avoiding unnecessary rabbit holes while covering everything that matters practically.

Second, questions beget questions in a chain reaction. Card count leads to image sourcing leads to copyright leads to original versus recolored. Each answer seeds the next question, and this chain rapidly weaves a comprehensive knowledge web.

Third, information arrives pre-structured. AI organizes the 78 cards into "22 Major + 56 Minor (4 suits x 14 cards)" — a clear framework. You skip the labor of manually assembling scattered information from multiple web pages.

What I Learned

First, when diving into a new domain, AI conversation is more efficient than googling because it enables "contextual learning." You get information at exactly the depth your project needs, fast.

Second, the greatest advantage of conversational AI learning is that it resolves the "not knowing the question" problem. Because the AI asks you what you need to know, the shape of your ignorance is revealed quickly.

Third, domain learning does not have to be a separate phase — it can happen simultaneously with planning. Learning about the domain while shaping the project direction in the same conversation is possible, and this makes a decisive difference in side project velocity.

Next Up

I now understand the structure of a tarot deck, and the copyright landscape is clear. Next, it is time to decide specifically what to build. In Part 3, I will share how three key questions from AI locked down the entire project spec in 10 minutes, and how being presented with options makes your own judgment criteria surprisingly clear.