Reliable Error · notes on psychology

Attention is a bottleneck

The common metaphor is a spotlight that can be widened with effort. A better one is a narrow doorway: the constraint is structural, and the only real control you have is over what queues up outside it.

Almost everything reaching the senses is discarded. That is not a defect to be trained away — it is the condition under which perception works at all. The interesting question is not why so much is lost, but why the loss is so rarely noticed, and the answer is that the mechanism that would have to notice is the same one doing the discarding.

Selection happens early, and it is not gentle

The classic demonstration is the dichotic listening task: different speech is played into each ear and the listener is asked to repeat, or "shadow", one channel. People do it competently, and afterwards can report almost nothing about the unattended channel — not the topic, often not even the language. Some things do break through, notably one's own name, which suggests the unattended stream is not discarded before any meaning is extracted. But the volume of loss is striking.

The consequence worth internalising: you cannot audit your own attention. The material that did not get through left no trace to be inspected, so introspection reports a complete picture regardless of how much was dropped.

Inattentional blindness, and what it does not show

The best-known illustration asks viewers to count basketball passes in a short video while a person in a gorilla suit walks through the scene. Roughly half of viewers do not report the gorilla. The finding is robust and has been reproduced in many forms, including with expert observers on domain-relevant material.

It is also routinely overstated. Two clarifications matter:

The useful inversion. If concentration reliably produces blind spots, then any task where missing a rare event is expensive should not depend on one concentrating observer. It needs a second pass, a different observer, or an alarm that does not rely on someone noticing. Checklists earn their keep here.

Switching is not doing two things

What is usually called multitasking is, for anything requiring deliberate thought, rapid alternation. The alternation is not free, and the cost has a consistent shape: a delay on each switch, plus a period of degraded performance while the previous task's rules are still partly active.

Laboratory switch costs are typically a few hundred milliseconds per switch, which sounds negligible. The reason it is not negligible is that the costs are paid per switch and switches are frequent — and the residual interference lasts considerably longer than the switch itself.

baseline + switch cost task A, then task B A / B / A / B alternating 0 total time
The same total quantity of work, reorganised. Identical tasks and identical volume; the only difference is how often the performer changes which rules are in force. The gap is the accumulated switch cost — paid entirely in overhead, with nothing to show for it.

Two practical corollaries follow, and they are unglamorous. Batch similar work so that rule sets change rarely. And treat an interruption as costing considerably more than the length of the interruption itself, because the expensive part is the reconstruction afterwards.

Where the effort actually belongs

Because selection happens upstream of awareness, effort spent trying to attend better is mostly wasted. Effort spent reducing what competes for selection is not. The asymmetry is the whole practical content of this page:

Instead ofDo thisWhy it works
Trying to concentrate harder Remove the competing cue Selection is not under voluntary control; the contents of the room are.
Relying on noticing a rare event A second pass, or an alarm Focused observers miss unexpected things reliably, not occasionally.
Interleaving unlike tasks Batch by rule set Cost is per switch; fewer switches, less overhead.
Judging your focus by how it felt Judge it by output Introspection cannot see what selection discarded.

None of this is a productivity system, and it does not scale into one. It is four consequences of a single structural fact: the doorway is narrow, and it was never going to widen.


Next: memory is reconstruction →