> Prerequisite: Episode 001 (superposition). This post explains why measurement is irreversible—the key to understanding all the engineering challenges ahead. > Next episode: 003, "Telepathic Dice".
One Peek and the World Changes: Why Measurement Is Irreversible
*(Header graphic: "From Coins to Error Correction · Quantum Computing Column", Episode 002 — "One look and the world changes: why measurement is irreversible." Illustration of a superposition collapsing, annotated: 'Collapse — you can only look once.')*
Your fridge has a light inside. Does it stay on after you close the door?
There's a serious solution: put a camera inside and film it. How does the camera "see" whether the light is on? Photons emitted by the lamp enter the lens. A photon travels from the lamp and hits the lens—and note, when a photon hits something, it gives that thing a push. The photon pushing the lens is fine; that push is exactly what we want. But the lamp itself is also being jostled by its own photons. No matter how sensitive you make the camera, the photons used to image the lamp have already physically entered the lamp's world. You cannot watch for free.
In the macroscopic world we don't notice this, because a glass of water has 10²⁵ molecules, and the tiny disturbance from a thermometer being dipped in is negligible. But a qubit is not a glass of water. A qubit's entire fortune is that one fragile "combination of tendencies" (the spinning coin from Episode 001). It is so delicate that any interaction used to observe it is enough to knock it over.
What Actually Happens When You Look
The previous post said: measuring a superposition yields a definite 0 or 1, with probabilities set by the proportions of the tendencies. Now for the harsher second half: after measurement, the superposition is gone.
Measurement is not "reading"—it is "rewriting". You ask the qubit "are you 0 or 1?". Before answering, it's still the spinning coin; after answering, it becomes a coin lying flat on the table—heads or tails—and it continues to exist in that new state. The original "spinning" is permanently overwritten and cannot be written back.
Three words summarize the nature of this rewrite:
- Irreversible. Collapse has no undo button. This is not an instrument-craftsmanship issue—it's not that our measuring devices are too crude; "obtaining a definite result" itself means the death of the old state. In information-theoretic terms: collapse erases the original tendency information, and quantum mechanics leaves no back door for already-erased information.
- Probabilistic. You cannot predict which way a given collapse will go; you can only compute the probabilities of each side. The same machine, the same operation, measured twice gives two different answers—and there is no hidden table with the answers written in advance (as Episode 001 noted, this is settled by experiment, not by preference of interpretation).
- One answer only. A superposition simultaneously holds two tendencies, but measurement redeems only one; the rest are voided. Hoping to measure one superposition and get two pieces of information? Impossible—one measurement spits out one bit.
- Their computing power comes from superposition (three hundred qubits carrying an astronomical number of tendency combinations at once);
- Their output can only be obtained by measurement—and the moment measurement happens, the superposition dies.
- Disanalogy 1: When the camera watches the lamp, the lamp is basically fine—the disturbance exists but is tiny. A qubit is "destroyed by one look"—the disturbance isn't small, it's fatal. The fridge analogy only holds on the point that "photons must participate".
- Disanalogy 2: Camera footage can be replayed and copied; quantum measurement results are one-shot, and quantum states themselves cannot be copied (the no-cloning theorem, Episode 028)—you don't even have the option of "back it up, then look".
- Disanalogy 3: The fridge question has an objective pre-existing answer (the light really is on or off); a qubit has no pre-existing answer before measurement (the dispute from Episode 001). So "installing a camera" in the quantum world is not "passive observation" but "rolling the dice with your own hand".
What If We Look Gently?
A reasonable follow-up: if looking too hard knocks it over, what about looking softly? Squinting, looking through frosted glass, in the gentlest possible way?
The physical world's answer: you can look gently, but gentleness has a price—as the disturbance shrinks, the information you gain shrinks proportionally. Peek at the qubit with an interaction so weak it barely disturbs it, and you get a reading that is almost pure noise; to squeeze definite information out of that noise you must repeat many, many times—and every repetition disturbs it. Put "seeing clearly" and "disturbing little" on a balance, and quantum mechanics caps their sum: however much information you take away, you must pay that much disturbance. No discounts.
This "handle with care" trick has legitimate uses in quantum computing (you'll meet it in the error-mitigation post), but it doesn't change the big picture: any measurement that yields a complete answer is necessarily the measurement that flips the coin.
The Quantum Computer's Dilemma—and Its Solution
Now you can see a fundamental dilemma facing quantum computers:
Isn't that "killing yourself by looking the moment you boot up"?
The solution is elegant enough to deserve its own post (Episodes 036 and 041 cover it in detail), but the idea can be stated now: the entire design of quantum computing is to quietly complete all the operations that should be superposed and interfering before measurement, measuring exactly once, at the very last step, to read out one answer. During computation, the programmer never "looks back" at the quantum state—all information flows inside the superposition, and measurement appears only as the award ceremony at the finish line.
This explains a quirk of quantum algorithms: the answer isn't "computed out" but "arranged to appear with high probability". The fourth post's search algorithm will demonstrate this full weighted-dice technique.
Half the dilemma remains: if you can't look midway, how do you know whether the machine has accumulated errors? No supervision allowed, errors piling up—this is the starting point of Episode 6 on error correction. One spoiler: the answer is "send other qubits to take the beating in its place"—monitor the protagonist indirectly by measuring supporting characters, so the lead never collapses. Humanity built the entire edifice of quantum error correction for this, and the second half of the column is devoted to it.
Tearing Down the Analogy
Where the fridge-camera analogy borrows and breaks:
What to Take Away
1. Measurement is not reading but rewriting: the superposition collapses into a definite result—irreversible, probabilistic, one bit at a time. 2. "Looking" must go through physical interaction (a photon pushes something), and a qubit is so delicate that any interaction is a fatal disturbance; if you want to handle it gently, the information gets discounted—no discounts available. 3. The entire art of quantum algorithms: complete the computation before measuring, and never look in between. As for preventing errors when you can't check in—send other qubits to take the beating (that's a story for later).
Hook for the next stop: two qubits can "grow together", and measuring one instantly makes the other "know" what to become—no matter how far apart they are. That's entanglement. Does it look like a faster-than-light telegraph? Yes—but it can't deliver anything. Episode 003, "Telepathic Dice".
Series: Quantum Computing · From Coins to Error Correction — Measurement, Collapse