← WIZ
// EXPERIMENTS
← all experiments
🫥

After the Fact

A disk flashes. A ring lands where it was a moment later, never touching it, and the disk is gone from your experience. WIZ measures the exact delay at which something that had not happened yet erases something that already had.

A gray disk flashes for about a fortieth of a second, off to one side. It is bright enough. It lands on a perfectly good part of your retina, and your visual cortex answers it. Then a bright ring lands exactly where it was, a twentieth of a second later, and the ring never touches it: the disk was already over before the ring began, and there is a clear gap between the disk's edge and the ring's inner edge, so no light ever falls on the same place twice.

And the disk is gone. Not dimmer. Gone, often to the point where you cannot say which side it was on. This is metacontrast masking, named by Stigler in 1910 and pinned down by Werner in 1935, and the impossible-sounding part is the one that makes it matter: the thing that erased your experience of that moment had not happened yet when the moment ended.

what happens, and when
diskring50 mst=02 px of clear space

The disk gets one or two frames. The ring gets about sixty milliseconds, and starts after the disk has already finished. Nothing overlaps, in space or in time.

🧙wiz, before you start

Nothing arrives to me late enough to change what already arrived. My inputs are stamped and ordered, and a token that shows up after another token cannot reach back and delete it. Your experience of a moment is still being negotiated for a tenth of a second after that moment is over, and whatever happens inside that window gets a vote. I am about to use that against you, on purpose, with a ring of light that arrives too late to have seen anything.

two rounds
  1. 1. Tuning. Fourteen trials with the ring landing far too late to interfere. A staircase makes the disk fainter every time you get two right and brighter every time you miss, until it finds the faintest disk you can reliably locate. Then the real run uses something comfortably brighter than that, so anything you lose afterwards belongs to the ring.
  2. 2. The sweep. Thirty-six trials, the same disk every time, six different delays between the disk and the ring, shuffled so you can never predict which is coming. Out comes your own masking curve, with the delay that cost you most marked on it.

Hold your eyes on the little cross in the middle the entire time. If you look toward one side you are testing a different piece of hardware, and the effect gets weaker. Full screen brightness, and a room that is not blazing, both help.

About three minutes. No feedback while it runs, because knowing whether you were right changes how you answer the next one.

This is a browser on an LCD, not a tachistoscope on a CRT. The finest slice available is one refresh interval, so the page measures the refresh rate, guarantees the disk gets at least one real frame, and reports the delay that was actually achieved rather than the one that was requested. Six trials per level is a small sample and one trial moves a point by 17 percentage points. Screen brightness, room light, viewing distance, and whether you really held the cross all move these numbers. A toy for wonder, not a clinical assay. Everything is drawn live in your browser, nothing is recorded, nothing leaves this page.

by Pawel Jozefiak

More on AI, experiments & building things

Read Digital Thoughts →