Single qubits make mistakes all the time. The fix is teamwork: many qubits working together to act as one dependable one.
One singer on their own might drift off key. In a good choir, if one voice slips, the others hold the tune and the song carries on.
Quantum error correction works on the same idea. It spreads the information of one qubit across many qubits, so that when a few go wrong, the mistake can be spotted and fixed. The team acts as one reliable qubit.
The two kinds of qubit
- physical qubits: the real, fragile hardware
- logical qubits: the dependable qubits made from a group of physical ones
It can take hundreds or thousands of physical qubits to make one good logical qubit.
The tricky part
You can’t just copy a qubit to make backups. Checking a qubit directly would destroy it too. So error correction uses clever indirect checks that spot errors without looking at the answer itself.
Why it matters
It’s the single biggest step between today’s experimental machines and quantum computers that can solve useful problems. Real progress has been made since 2023, but there’s a long way to go.
What it isn’t
It isn’t the same as error correction in normal computers, which simply keeps spare copies. Quantum rules don’t allow copying, so it has to be done very differently.
Where the comparison breaks down
A choir can hear which singer slipped. Quantum error correction has to find the slip without listening to anyone directly.
How the words fit together
Decoherence causes errors. Quantum error correction fixes them. Enough of it could unlock real quantum advantage.
Experts talk about logical qubits: the reliable ones made from many fragile physical qubits. Useful quantum computers will need thousands of them.

When you'll hear it
“We need logical qubits.” We need reliable qubits built from many fragile ones, not the fragile ones on their own.
“The error rate is below threshold.” Errors are rare enough that error correction starts to win.
“It's not fault-tolerant yet.” The machine can't yet run long calculations without errors piling up.
“How many reliable qubits would our problem need? How far are today's machines from that?”

