superscript¹

possible and impossible

a song cut up while reading, put on its feet by a machine that cuts as it reads.

¹ cut up live, while reading an essay aloud.

possible and impossible

The words of this song were cut up live while an essay was being read aloud¹. First you hear that reading: the artist's own voice, as it was recorded. Cut a line while it is read, or in the breath after it, and it stops in mid-word. A cut line is impossible; an uncut one is possible.

Then the song is put on its feet. The lines stand in a queue and a machine reads them as Emil Post described in 1921: read the first word, cut three from the front, and write at the end possible possible impossible possible after a possible, or impossible impossible after an impossible. When fewer than three remain there is nothing to read, and it stops.

Each line stands with two blank scraps, because the machine reads only one strip in three: so every line is read once, and every cut matters. Lines from the reading are read in the artist's voice. A cut one is silence: that take is one voice, effectively mono, and outside its centre there is almost nothing. Every line the machine writes inherits the line it was written after and is one generation on, and the generations alternate between the song's two produced versions, a pair of countries. A strip is read about one pass of the queue after it was written, so the versions take turns by the pass, in sections. Lines drift: in a long run, one of them usually inherits the whole tape. The second version leaves out the song's last five lines: it never sings that computers made it possible or impossible, and it never asks. A line one version lacks is sung by the other. The first version came apart: a computer separated it into its tracks, heard here as voices, keys, guitar, bass and rhythm. A strip of it is played by the band its superscript chooses, written in binary: keys are the ones, guitar the twos, bass the fours, rhythm the eights. A possible strip is sung over that band; an impossible one is the band without its voices, or silence where that band does not play. The second version was never taken apart, so an impossible line of it is a ghost: the mix with its centre removed, where the singer survives as reverberation, because a finished mix cannot be perfectly unmixed. Under the strip being read, the page says what you hear. The two strips cut at each reading are heard as cuts in the artist's voice, and the machine writes each new word as a syllable of that voice: pos, im.

Every reading you hear is written by the p3r kernel, which answers every tick within a fixed budget of work: it cannot run forever, and so it cannot compute everything. The tape lives outside it. Whether the reading ever stops, the machine can only find out by running. It can run far ahead in silence, up to five hundred million readings: it may find that the reading stops, or that it returns to a word it has held before and so repeats forever. Both are proofs. Or it may find nothing in the time it has, or the word may outgrow the memory this page allows itself.

No procedure can decide, for every machine of this kind and every word, whether the reading ever stops: Minsky proved it in 1961, and Cocke and Minsky showed in 1964 that cutting two at a time already computes anything a computer can. Post asked whether, under his own rule, every word eventually stops or repeats. As far as is known, nobody has answered.

The machine can be cut but not uncut. It can tell you what it found, and when. It cannot tell you whether it should. That footnote is yours.

Space waits or continues · C or Enter cuts · arrows choose a strip · A runs ahead · P changes the pace · E ends it · M mutes · Escape leaves. Still reduces movement.

The artist's reading, the two produced versions of the song made with Suno from that reading, and the stems Suno separated from the first, are separate from the MIT software. Audio provenance ↗

superscript1

superscript¹