What this actually measures
N-back is not a memory test in the usual sense. Nothing has to be stored for long — the whole task lives inside a few seconds. What it loads is the updating: with every letter the queue has to shift, the oldest item drop out, and the comparison happen before the next one arrives.
That is why 2-back feels merely awkward and 4-back feels impossible. The number of items barely changes; the number of updates per second does not change either. What changes is how long each item must survive being interfered with by the letters that follow it.
Where the machine is
- A computer solves any n-back with a ring buffer of n cells: constant memory, constant time, no interference, no fatigue.
- Human accuracy falls off a cliff between 3-back and 4-back, and training moves the cliff much less than the training industry claims.
- Errors here come in two flavours the machine has neither of: missing a real match, and confidently reporting one that never happened.