03 Curiosity 8 min read
The Anesthetic and the Amplifier
The same tool that lets you reach the frontier of any field in a month can also kill the itch that would have taken you there.
There is a version of the future where every human being has a patient, inexhaustible tutor and the result is an explosion of amateur expertise. There is another where nobody wonders about anything for longer than four seconds because wondering has been made unnecessary. Both are plausible. The variable that decides which one you personally get is small, specific, and entirely under your control.
Curiosity is a gap, not a trait
We talk about curiosity as a personality attribute — some people have it, some do not. The better model, from George Loewenstein’s 1994 synthesis, is that curiosity is a state produced by an information gap: the felt distance between what you know and what you want to know.
The critical detail is the shape of the curve. Curiosity is weak when the gap is enormous — you cannot be curious about a field you know nothing about, because you cannot form a question. It is also weak when the gap is zero. It peaks when the gap is narrow and salient: when you almost know, when the answer is tantalisingly close but not yet in hand. That is the condition that produces obsession, and obsession is what produces depth.
Now notice what an instant, fluent, always-available answering machine does to that curve. It takes the narrow, salient, painful gap — the productive one — and closes it before it can do any work. The itch is scratched at the exact moment it becomes interesting.
Guess first
There is a large and unusually consistent literature on this. The generation effect: information you produce yourself is retained far better than information you read. The pretesting effect, which is stranger and more useful: being tested on material before you have learned it improves later retention — even when every one of your answers is wrong. The act of committing to a guess appears to prepare the mind to encode the correction.
The operational consequence is a single rule, and it costs about ninety seconds:
Before you ask the machine, write down what you think the answer is, and how confident you are.
The whole protocol
This does three things at once. It keeps the information gap open long enough to register as a real question. It converts the machine’s answer from content into feedback, which is a much more nutritious form of information. And it calibrates you — over a few months of doing this you develop an accurate sense of which domains you are reliable in and which you are fooling yourself about, which is the foundation of everything on the judgment page.
The amplifier case is real too
It would be dishonest to write this page as a warning only. The amplifier argument is strong and should not be conceded.
The hardest part of entering an unfamiliar field has never been the availability of information; libraries have been free for a century. It has been the absence of someone to ask the embarrassing question — the one that reveals you do not understand something a first-year is assumed to know. That barrier stopped an enormous number of people at the door. It is now gone, permanently, for everyone with a connection. Autodidacts who would previously have needed institutional access can now get to the working edge of a specialism in months.
So the tool is genuinely bidirectional. The person who guesses first and then interrogates gets the amplifier. The person who asks first and reads the answer gets the anaesthetic. Same tool, opposite outcome, and the difference is a ninety-second habit.
Partly right. Curiosity does relocate upward, and mourning the loss of low-level drudgery is usually sentimental. But the calculator analogy hides the load-bearing difference: a calculator answers a question you already formulated. A language model will formulate the question for you, evaluate it, and hand back a conclusion. It operates one level above the calculator — at the level where the relocation was supposed to happen. That is the level worth defending, and the defence has to be deliberate because nothing about the interface encourages it.
Where curiosity sits on the ladder
- 1
Cede Machine
The machine does it. You never see the work.
Asking a model for an answer you never guessed at, and moving on. Efficient, and it leaves no trace in you.
- 2
Approve Machine-led
The machine drafts. You review and sign.
- 3
Collaborate Shared
You and the machine work the problem together.
- 4
Critique Human-led
You do the work first, unaided. Then ask the machine to attack it.
Guess, commit, then ask. The answer becomes feedback instead of content, and the gap stays open long enough to matter.
- 5
Reserve Human
You do it alone, on purpose, knowing help exists.
Before any substantive query, write your predicted answer and a confidence percentage. Keep them in one running file. Review it monthly — the pattern of where you are overconfident is worth more than any single answer you received.
Maintain a list of questions you have not looked up. Let items sit for a week. The ones that keep resurfacing are your actual interests, as distinct from your momentary ones. Follow those hard.
“What do serious people disagree about here, and what is the strongest case on each side?” produces a map. “What is the answer?” produces a destination. Maps sustain curiosity; destinations end it.
Choose one field you learn the slow way — books, practice, other humans, dead ends. Not because it is efficient. Because it is the only way to remember what forming a question from scratch feels like.