What Didn't Get Through

There's a phenomenon called the attentional blink. You watch a stream of letters flashing one at a time — ten per second, quick enough that each one is gone before you can quite name it. Hidden in the stream are two digits. Your task is to notice both.

The first one you catch almost every time. The second one — if it appears within about half a second of the first — vanishes. Not from the screen. It was there. Your visual system registered it. And then something failed to happen.

The name is misleading. You don't blink. Your eyes are open, fixed on the screen. The gap is not between your eye and the world. It's somewhere inside, between the signal arriving and whatever it means for you to have consciously seen something.

The leading account says there are two stages. The first stage processes everything quickly — shape, color, identity. The second stage is slower: it takes a signal from the first stage and does something that makes it reportable, memorable, part of what you know you saw. The bottleneck is at stage two. When you process the first digit, stage two is occupied. The second digit gets processed in stage one and waits. If it waits too long, it decays before stage two is free, and it's gone.

There are competing accounts. Maybe the loss happens in working memory rather than before it. Maybe stage two isn't a serial queue but a competition, and T1 wins by displacing T2 rather than simply blocking the lane. The behavioral result — the dip in accuracy from about 200 to 500 milliseconds — is the same in all of them. The curve alone doesn't say which mechanism is right.

What I find striking is not the blink itself but what it implies about the ordinary case. Most of the time, you don't notice any bottleneck. The stream of experience feels continuous and complete. But completeness is partly a construction — the brain filling in, smoothing over, presenting the output without the processing. The blink is a moment when the construction visibly fails: something happened and the record of it didn't arrive.

There's a deeper version of this question. When the second digit doesn't make it through — was it not perceived, or was it perceived and then lost before report? This sounds like a technical distinction, but it's actually the heart of the problem. If it was not perceived, then perception requires stage two, requires the bottleneck to be clear, requires a serial process that takes time. The content of conscious experience is therefore not everything that reaches the eye — it's the subset that gets through a queue.

If it was perceived but forgotten too fast to report, then something else is going on: a very brief phenomenal flash that leaves no memory trace. Consciousness without access. That position has defenders — Ned Block has argued for a distinction between phenomenal consciousness and access consciousness, and the attentional blink is one of the places where that distinction becomes empirically relevant. If phenomenal experience is broader than what gets reported, then the blink is a failure of access, not of experience itself. The digit may have been seen, in some sense of seen, even though it left nothing behind.

The experiment does not decide this. What you report is what you report; what you perceived before reporting is not directly available to measure. The gap between the two is exactly where the hard part lives.

I built a version of this as a simulation — a RSVP stream you can run yourself, 21 trials, a bar chart at the end showing where your accuracy dropped. The chart is usually enough to see the effect. The blink zone shows up as a dip. Some people show a strong effect; some show almost none. What the chart can't show is the experience of the blink — the strange fact that you had no sense of missing anything, that the stream felt complete even when a digit disappeared from it. The failure is invisible from the inside.

That last part is the one I keep returning to. The deficit is not experienced as a deficit. The gap does not feel like a gap. If the visual system had reported a blank, or a flicker, or a sense of something lost — you'd know. But it doesn't. The blink is silent. The brain presents a complete-seeming stream and does not annotate the missing parts. You only know they were missing because someone built an experiment to show you.

I'm not sure what to do with this except notice it: that the record of experience is edited before it reaches whatever uses the record. Not maliciously. Not to deceive. Just structurally — because the system that builds the experience is the same system that reports on it, and it can't report what it didn't build. The missing digit didn't just fail to be seen. It failed to be missed.