Let AI analogize with something I already know
It doesn't know what you already know, so it pulls a source domain from the public question bank. Leave that slot open in the question, fill it with something you already get — then the analogy can actually land.
The AI picked a poker analogy that sounded folksy. Alice doesn’t play cards, so she took “sunk cost” to mean “the more you’ve spent, the more you should keep going.”
Alice bought an online course. Three lessons in, she saw the content didn’t match the pitch, but she couldn’t let go of the ¥899 she’d already paid. A friend said that money might be a “sunk cost.” She’d never heard the term, so she asked the AI for an analogy.
The AI said: “Like chips already in the pot. From here, you only ask whether the next bet is worth it.” The sentence is smooth. Alice never plays cards. She doesn’t know what a pot is, or when you’re supposed to call, so she can only guess from the words.
The next day she landed on the wrong conclusion: since she’d already spent ¥899, she should put in even more time. An unfamiliar concept, wrapped in another unfamiliar experience. The explanation looked done. The understanding drifted further.
The problem isn’t only whether the analogy is accurate. The AI doesn’t know whether Alice knows online returns, cooking, or project schedules, so it picks a common stand-in for her. Common isn’t the same as familiar. Smooth to hear isn’t the same as checkable.
Ask it differently: hand it an experience you already have. Alice often tidies the fridge, so she can ask the AI to explain with food that’s already gone bad: the grocery money is gone; what you cook tonight should only depend on whether the food is still edible, and how long a new meal would take.
- Name a source domain you know: write down something you’ve done yourself, where you can judge the outcome. Don’t just say “use a life example.”
- Spell out the mapping: ask the AI to say, for each move in the source domain, what it corresponds to in the target concept.
- Mark the analogy’s limits: ask which details you must not copy over, so a local likeness doesn’t get treated as sameness everywhere.
An analogy can catch you only if the thing it borrows is already in your head. Below, the same concept gets three stand-ins. See which one you can check yourself.
What actually changes the result in the drill is taking the source-domain choice back, and making the AI spell out the mapping and the limits. Copy the two sentences below as a block. Swap only what’s in the brackets.
- First, name a source domain you actually know
I’m trying to understand [sunk cost]. Analogize with [clearing spoiled food from the fridge], which I do every week. Only explain why past spend shouldn’t steer the choice in front of you.
- Then require a one-by-one mapping, and draw the limits
Spell it out one by one: the money already paid, the course that doesn’t deliver, and whether to keep spending time today — mapped to grocery money, spoiled food, and what to eat tonight. Then add a separate “analogy limits” list of details you must not copy over.
“Familiar” has to land on experience. Using an investing app doesn’t mean you understand an investment loss. Cleaning the fridge every week means you’ve seen food go bad, and you’ve decided to toss it and buy again.
Whether an analogy is useful depends on whether you can check every step. Grocery money already spent maps to the course fee already paid. Eating spoiled food won’t bring the value back, just as sitting through a failed course won’t make it good. Whether you cook something else tonight depends only on the payoff and the cost right now.
“Analogize with something from life” still hands the choice to the AI. “Explain it with cooking” is more specific, but it doesn’t say which part of cooking you actually know. Write “use my weekly habit of clearing spoiled food,” and the source domain is specific enough to check.
The fridge analogy has limits too: a course might still be refundable, transferable, or usable in parts. Spoiled food usually isn’t. An analogy only carries the structure you need right now. It doesn’t cover every detail. Asking the AI to point out where it fails keeps you from hearing “stop putting more in” as “drop everything at once.”
That’s a matter of how you use it. Where the stand-in sits in a complete question is in Let AI explain it for my situation.