A flash

When does the light actually turn on?

Ask for 5 ms and you get a whole frame. We don’t see what we asked for.

Start with one letter. Mayzner’s letters appeared on a Fairchild Type 737A cathode-ray tube driven by a PDP-7 computer. Its P24 phosphor had, in his words, a decay time “on the order of a few microseconds”, so the glow died almost as soon as the beam stopped painting. By 1968 a crystal clock let the program set how long each letter was painted, in steps of 50 microseconds.

Your screen works differently. A browser can only change the picture at regular intervals, one frame at a time, so this page can only show a letter for a whole number of frames. It has timed the frames in this window: measuring…

That is the timing of the picture the page hands over. How the panel’s light rises and falls inside each frame is a separate matter, and this page cannot see it.

Drag the slider to ask for a duration, then press Flash. The strip under the panel compares what you asked for with what the page can show.

Ask for 5 ms and you still get a whole frame. Every step that follows works in whole frames and tells you how many it used.

Five letters in order

Again.

Same five letters. Same five places. Same timing.

Now light all five letters of chair one after another, left to right. Each letter turns off as the next one turns on, so only one letter is ever lit.

Start at the slow end and press Run. The letters step across the panel. Then drag the slider toward the fast end and run it again.

At the slow end the order is obvious. Near the fast end the steps blur into a sweep, and at the fastest setting the word seems to arrive all at once. Mayzner found the same thing: when the whole word took less than about a tenth of a second, observers saw five letters appear together.

At every speed, all five letters are seen.

Change the order

Look at the mark. Say what you read.

Keep the letters, their places and the timing. Change only the order in which they light: H first, then I, then C, A and R. Every letter is still lit for the same time, in its own place.

The slider starts near Mayzner’s setting of 40 ms per letter, which is 200 ms for the whole word. Rest your eyes on the small mark above the row and press Run.

Mayzner’s observers reported c a r, with empty spaces where the h and the i belong. His own summary was that when the whole word takes about 200 ms, the visual system erases what arrived in the first 100 ms.

Switch the order back to C H A I R at the same speed and the two letters return. Nothing about the light changed except its order.

Find the window

Mayzner measured the window

Under about 100 ms, all at once. Over about 300 ms, a sequence. In between, letters go missing.

Blanking only happens inside a band of speeds. Mayzner found that when the whole word took less than about 100 ms, the five letters looked simultaneous. When it took more than about 300 ms, they looked like a sequence. Either way all five were seen. In between, with a peak near 200 ms, the first letters went missing.

What counts is the time for the whole word. With ten letters he halved the time per letter and the effect held.

The slider now holds every speed this screen can produce in that range, always in the order H I C A R. Run each one and say what you read. The shaded band on the strip marks where Mayzner found blanking.

On this screen, a few of the settings fall inside the band. Mayzner’s computer could move the timing in steps of 50 microseconds, so he could place a setting anywhere in it.

The machines

One frame of this screen holds 333 ticks of his clock.

Every step so far was timed in frames, because a frame is the shortest thing this page can show. One frame of this screen holds 333 ticks of his clock. This is why my version only half-worked in 2013.

Mayzner’s display had no fixed frame rate. The beam painted a letter when the program told it to, and a crystal clock ticking 20,000 times a second let the program hold it there for any multiple of 50 microseconds. The phosphor went dark within a few microseconds of the beam moving on.

A film projector is a frame machine too, and a slower one. Each frame lasts 41.7 ms, about as long as the 40 ms Mayzner gave each letter when he left no gap between them.

Drag the slider to ask four machines for the same flash.

In 1968 the scarce part was the computer. By 2013 it was the display. The computer could count microseconds and the screen could only show frames.

Two experiments are left for the hardware, because a frame is too coarse for them. With ten letters, Mayzner lit each one for 10 ms with 10 ms between, and one order made the word somersault appear to split in two. And Pollack’s test, from the last step, was timed in tenths of a millisecond.

The rebuild goes back to what Mayzner had, which is lights that turn on and off when told. Each letter is its own LED light box, switched by a transistor from a microcontroller’s hardware timer. The last job is to point a photodiode at each box and measure when the light is on, because the pin that asks for light is not the light. Pointed at this screen, the same photodiode would show what the green block leaves out.

The 1968 machine, as Mayzner described it that year:

Computer

DEC PDP-7 with 8K words of 18-bit core memory

Displays

DEC 340 master and 343 slave, plus two Fairchild 737A consoles with P24 phosphor

Timing

20 kHz crystal clock, in steps of 50 microseconds

Cost

about $200,000, plus about $5,000 a year for maintenance

Software

two display programs, each filling the 8K and taking about three man-months to write

Add a third group

The part I could not picture: add a third group, and the first one comes back.

This step uses a different display. In 1970 Mayzner filled a small grid with one repeated letter and lit the cells in groups, each group on for 20 ms and then 80 ms dark, so a tenth of a second from the start of one group to the start of the next.

Start with two groups. Two letters in the middle row light first, then the three letters around them. The second group erases the first, and the row reads as three letters with two gaps.

Then add a third group: six letters above and below the three, lit last. Those six act on the three the way the three acted on the pair. With the three suppressed, the first pair comes back.

Look midway between the two marks, run each version a few times, and say whether you saw the first pair.

Adding more light brings something back.

How can something that happens later change whether we see something that happened earlier?

All ten of Mayzner’s observers changed in this direction, though not equally. One of them barely lost the first pair to begin with.

He called the effect disinhibition, after a finding by Hartline and Ratliff in the eye of the horseshoe crab, where lighting a third receptor releases the first from the inhibition of the second.

Erased is not gone

So far this could be a trick of one kind of display. Irwin Pollack showed that it is stranger than that. He had watched Mayzner’s demonstration in 1966, and in 1972 he rebuilt it at the University of Michigan with rows of short lines in place of letters.

Each trial showed four rows, one after another. In one row a line was tilted. The observer had to say which row, and to guess if unsure. The tilted line could sit where it was lit in order and seen, or where the lines lit after it erased it.

If the letters disappear, is the information gone too?

Did you see the line? — No.

Which line was different? — That one.

They could tell the tilt of lines they reported they could not see.

The observers picked the right row as accurately for the erased lines as for the ones they saw, and under some conditions more accurately. They could tell the tilt of lines they reported they could not see.

So the light was shown, the observer says nothing was there, and something about it got through anyway.

Information about an event can survive when the experience of it does not.

This page cannot run his test. He set his timing in tenths of a millisecond, and a frame is too coarse for that. It is one of the experiments saved for the LED box.