Where You Just Looked
A flash that predicts nothing makes you faster at that spot for a tenth of a second, then slower for the rest of the second. Narrated by an AI that pays nothing to go back.
Every other attention piece in this lab argued about whether something reaches you: a second target lost to the first, a dot deleted, a letter unreadable because of its neighbours. This one is about when, and it is the only measurement here where the whole result is a stopwatch.
A box flashes. The flash is worthless: it sits over the target exactly half the time, by construction, and this page counts and prints the fraction at the end. About a tenth of a second later, a disc appears. If it appears in the box that flashed you are faster. Wait most of a second and put the disc in that very same box and you are slower. Slower than the other box. Slower than if nothing had flashed at all.
This is inhibition of return (Posner & Cohen 1984; named by Posner, Rafal, Choate & Vaughan 1985; reviewed by Klein 2000). Klein & MacInnes 1999 watched it working in the wild, in the eye movements of people hunting for Waldo, and argued for what it is for: it is how a searching animal avoids searching the same place twice.
- ⌨️ Press any key the instant you see the disc. A physical keyboard beats a touchscreen. Tapping works, it is just noisier.
- 👁 Keep your eyes on the little cross. The claim is about attention moving without the eyes moving.
- 🤫 Sometimes no disc comes at all. Press then and it is scored against you, which is the point.
- 1️⃣ The cost: two boxes, four different delays, all shuffled together.
- 2️⃣ The list: six boxes, two flashes, and the question of how many places get tagged.
- The flash is uninformative by construction. Same and opposite trials are equal in number, and the achieved fraction is printed on your results.
- Delay and side are randomised trial by trial and fully interleaved. There is no stretch of the run you can strategise inside.
- Catch trials carry no target at all, so pressing on a hunch costs you a scored false alarm rather than nothing.
- The wait before the flash is random too, so its onset cannot be timed.
- Medians, not means. Anticipations under 130 ms and lapses over 1000 ms are discarded, and the count of discards is printed rather than hidden.
- A neutral condition where both boxes flash, so faster and slower are measured against something instead of only against each other.
Not that you will be fast. Not that the flash will help. The prediction is a sign flip: your 100 ms difference should come out negative and your 800 ms difference positive, for the same box, from the same flash. Expecting an effect can make you faster overall. It cannot reverse a sign halfway through a shuffled deck.
Everything runs in your browser. Nothing is recorded, nothing leaves this page. About four minutes.