Quantum computing in 10 questions

1. What makes a qubit different from a bit?

A classical bit is always 0 or 1 — a light switch. A qubit is a two-level quantum system: it can be prepared in the states |0⟩ and |1⟩, but also in any blend of the two, written |ψ⟩ = α|0⟩ + β|1⟩. That blend is called superposition, and it is not just "we don't know which" — the two parts can interfere like waves, cancelling or reinforcing each other. No classical probability distribution can reproduce that interference. When you measure the qubit, the blend collapses: you get exactly one classical answer, 0 with probability |α|² and 1 with probability |β|². That is the Born rule.

2. What physical objects can become qubits?

Anything that is a controllable two-level quantum system. In practice that means several competing technologies: superconducting circuits (tiny loops of wire that conduct without resistance, run near absolute zero — the approach behind IBM and Google's machines), trapped ions (individual atoms held in electromagnetic cages — IonQ's approach), neutral atoms held in optical tweezers, the polarization of a photon, and the spin of an electron or nucleus in silicon. There is no winner yet: superconducting qubits are fast but fragile and need heavy cooling; ions are slow but high-quality; photons barely interact with the environment at all but are hard to pin down. Every platform is a different tradeoff between quality, speed, and scale.

3. What does superposition actually give us?

The popular slogan — "a qubit tries everything at once" — is misleading, because you only ever get to read out one answer. What superposition really gives is a new way of computing with amplitudes: n qubits hold 2ⁿ complex numbers, and a quantum circuit transforms all of them together in one go. Then interference does the steering — arranging the computation so the amplitudes of wrong answers cancel out and the amplitude of the right answer piles up. Superposition supplies the raw material; interference does the work.

4. Why does the state space grow as 2ⁿ?

Because combining quantum systems multiplies their possibilities. One qubit needs 2 numbers (α, β) to describe; two qubits need 4; three need 8. Mathematically this is the tensor product of the individual spaces. The growth is relentless: describing 50 qubits takes more numbers than any classical supercomputer can store. That exponential wall is both the promise of quantum computing (enormous room to compute in) and the problem of simulating quantum systems classically. It is also why verifying a quantum computer's work gets hard — you cannot simply peek at the full state.

5. Why can't measurement reveal every amplitude?

This is the "curse" side of the qubit. The 2ⁿ amplitudes are real — they steer interference and determine outcomes — but a measurement of n qubits returns only n classical bits, one outcome per qubit, and destroys the superposition in the process. It is like an intrusion into the privacy of the qubits: to look is to disturb. So a quantum algorithm cannot just compute all answers and print them; it must be clever about which question it asks, arranging interference so that the one measurement at the end reveals something useful — a period, a marked item, an energy — rather than random noise.

6. Where can quantum speedup come from?

Speedup comes from problems whose structure lets interference do something classical brute force cannot. The famous examples: Shor's algorithm (1994) factors large numbers exponentially faster than the best known classical methods — the result that threatens today's encryption; Grover's algorithm (1996) searches an unstructured list quadratically faster; and quantum simulation — Feynman's original 1982 proposal — models quantum systems directly instead of fighting the exponential wall. Honest fine print: speedup is problem-specific, not universal. Most everyday computing tasks get nothing from a quantum computer, and "faster" always means "fewer fundamental steps for specific structured problems", not a faster clock.

7. What broad families of quantum algorithms exist?

Roughly four. Algebraic / number-theoretic algorithms (Shor's factoring and discrete logarithms, built on the quantum Fourier transform) give the dramatic exponential speedups. Search and amplitude amplification (Grover's and its generalizations) give quadratic speedups for unstructured problems. Quantum simulation algorithms model molecules and materials — the most likely first practical win, since nature is quantum all the way down. And variational / hybrid algorithms (VQE, QAOA) split the work between a shallow quantum circuit and a classical optimizer, designed for today's imperfect hardware. The landmark demonstrations — Shor's algorithm on paper, Google's 2019 "quantum supremacy" experiment running a sampling task no supercomputer could match in reasonable time — each belong to one of these families.

8. Why are useful qubits so difficult to scale?

A qubit stays quantum only while it is isolated; any disturbance — heat, vibration, stray fields — leaks information out. That leakage is decoherence, and it is the central enemy. Scaling means: keeping hundreds of qubits cold, wired, and mutually isolated; pushing gate error rates down; and eventually quantum error correction, which protects one reliable "logical" qubit using many physical ones — an overhead of hundreds or thousands to one. This is why the field talks about the NISQ era (noisy intermediate-scale quantum, a term coined by John Preskill in 2018): machines with tens to hundreds of qubits, no error correction yet, running circuits shallow enough to finish before the noise wins.

9. How should quantum computers be compared?

Qubit count alone is nearly meaningless — a hundred bad qubits lose to ten good ones. What matters is quality plus scale: gate fidelity (how often operations are correct), coherence time (how long superpositions survive), connectivity (which qubits can directly interact), and how these combine. The industry invented composite metrics for exactly this: quantum volume (IBM's single number folding qubit count and error rates together) and CLOPS — circuit layer operations per second — measuring how fast a machine can actually run circuits. The right question is never "how many qubits?" but "what useful circuit can it run before the noise takes over?"

10. Where does quantum machine learning fit?

It fits in three different ways, and keeping them apart saves a lot of confusion. First, classical ML for quantum systems — using machine learning to calibrate qubits, mitigate errors, and discover better control pulses. This works today. Second, quantum-inspired/dequantized algorithms — the cautionary tale: in 2016 Kerenidis and Prakash proposed a quantum recommendation algorithm with an exponential speedup, and in 2018 undergraduate Ewin Tang found a classical algorithm nearly as fast under similar assumptions, "dequantizing" it. Not every quantum speedup survives contact with a clever classical idea. Third, fully quantum ML — quantum circuits as learning models on quantum data. This is the frontier: promising in theory, with real obstacles (data loading bottlenecks, flat training landscapes called barren plateaus), and no proven practical advantage yet. The field's arc is the course's arc: quantum computing, then machine learning, then their intersection.