weeklyAI
TertulIA · courses
Understand · lesson 1 of 20next lesson →
E01 · Everything Is Numbers
You trace what a scanner, a chat box or a phone threw away to turn your document into numbers, and check whether the wrong thing was lost.
Ask TertulIA
Ask me about this lesson, or what to learn next.
Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.
The lesson · 31:46
Short cut · 18:13
Trailer · 1:23
The challenge
The challenge asks for your field and not for mine, which is deliberate, because the result has to matter to somebody, and that somebody is you. Pick the file your own work most depends on, the one you'd be in real trouble without on a Monday morning. Then find out what format it's in, what its rate or its resolution was, what its compression assumed about you, and whether a lossless original exists anywhere at all. Nothing here can break: looking at a file's format changes nothing inside it, and a test export is a copy, so the original stays exactly where it was.
Starting points:
- A graduate student: the instrument data behind your thesis; find out what rate and precision it was recorded at, and whether the archive keeps a lossless master or only the convenient export.
- A tenants' organizer: the scanned lease and the photographs your case leans on; find out which of them exists anywhere as an untouched original, before the other side asks.
- A machinist: the drawing that arrives as a scanned PDF; run the digit check on every dimension that carries consequences before you cut metal.
When you're done, tell us what you pursued and tell us what you got.
Field Card · E01 · Everything Is Numbers
The distinction
Everything inside a computer is numbers, and something was thrown away to make them, by somebody, at some rate, to some precision.
In practice
- When a scanned number matters, check it against the original; a substituted digit comes out clean.
- Keep the lossless original; circulate the compressed copy.
- When a letter goes missing, the error is not in the name. Somebody wrote the table, and tables get rewritten.
The question to keep asking
What was thrown away when this became a file?
The fine print
A substituted digit does not look like damage; the page comes out clean, legible and confidently wrong, with no warning on it. The file cannot tell you what it is missing.
Glossary · E01 · Everything Is Numbers
- The trade
- The exchange the whole course is built on: something turned into numbers, and something got left behind so that could happen. Every strange thing one of these machines does comes back to it, which is why the useful question about any file, and about any AI, is what it threw away to answer you.
- Bit
- The smallest possible fact there is, one answer to one question, a switch that is open or a switch that is closed. Put a row of them beside each other and the row as a whole says a number, and every position you add doubles how far the row can count, which is why a small handful of zeros and ones will hold your age, or a color, or the price of your coffee.
- The file
- A sequence of numbers, in a particular order, plus an agreement about how that list gets read, and nothing else in there at all. The icon on your screen, the little sheet of paper with the corner folded over, was drawn decades ago precisely so that you would not have to look at that, and it was a kind decision, but it is still a drawing.
- The table
- The agreement that decides which letter gets which number, written by people, with a counted number of seats in it. The trade calls it a standard, or an encoding, words that sound like nature reporting itself; the American standard written in the nineteen sixties had seven bits and a hundred and twenty-eight seats, and the border around it was drawn by a committee, in a room, in one country, on a particular afternoon.
- Pixel
- Not a tiny picture but three readings about one tiny location on a grid, how much red is here, how much green, how much blue. Zoom into any photograph on your phone far enough and the picture stops being a picture and becomes flat squares, and inside a single square there is no drawing left at all.
- Sampling
- Choosing how often to measure something continuous, which means picking these instants here and not the ones in between them. There is real mathematics behind how often is often enough, from Nyquist and from Shannon, which says roughly that you have to measure more than twice as often as the fastest thing you want to keep.
- Quantization
- Choosing how finely to record each measurement, so that every reading gets snapped to the nearest rung on a ladder of levels somebody settled on beforehand. The word sounds like a measurement and it is a budget, which is the distinction the lesson keeps insisting on: how often to measure is mathematics, and how finely is what somebody could afford.
- Organized forgetting
- The kind of shrinking that decides which numbers you are not going to miss and then throws them away for good. The trade calls it compression, and that one word covers two different things, the honest bookkeeping that finds a shorter way to write the same numbers and can hand the original back exactly, and the irreversible kind that makes streaming and scanning and pocket photography possible, where the discarding is the whole point.
- A bet about your ears
- What a compressed sound file actually holds, since it does not hold the music. The trade calls the mechanism a psychoacoustic model, which sounds like a description of hearing, and what it is is a prediction that a loud sound masks the quieter ones sitting next to it, so the encoder spends almost nothing on what the model says you cannot hear anyway. The same kind of bet gets made about your eye in a photograph, that you notice brightness more than color and broad shapes more than fine grain.
- The stamp
- How the photocopiers of twenty thirteen shrank a scan: marks that looked repeated were stored once as one clean stamp, and that stamp was reused everywhere the pattern seemed to come back. When the matcher decided a scanned six sat close enough to a stamp it already had, and the stamp it already had was an eight, the page came out clean, legible and wrong, and the maker's own notice called it a scanning issue, although nothing had malfunctioned.
- The shavings on the floor
- Under the bench where the world gets turned into numbers there is always a floor of material that had to be cut away so the pieces could travel, and it is not waste, it is a record of decisions: what a formula predicted you would not notice, what a table had no seat for, what a sampling rate was never fast enough to hear. Every file you own is standing over a floor like that one.
The part this lesson earns on the map
- THE FRAME (E01)
- Everything here is numbers, and something was discarded to make them.
Carried in from earlier lessons
This is the first lesson of the course, so nothing is carried in yet, and everything above is what the other nineteen parts will lean on.
References
Boole, G. (1854). An investigation of the laws of thought. Walton and Maberly.
Cerf, V. (1969). ASCII format for network interchange (RFC 20).
El País. (1991, 9 de mayo). [Cobertura de la disputa europea sobre la eñe].
Fraunhofer IIS. (2025). 30 years of MP3.
International Telecommunication Union. (2015). Recommendation ITU-R BT.709: Parameter values for the HDTV standards.
Kriesel, D. (2013, August 2). Xerox scanners and photocopiers randomly alter numbers in scanned documents.
Library of Congress. (s. f.). JPEG image encoding family; JPEG DCT compression encoding; MP3 audio encoding.
Nyquist, H. (1928). Certain topics in telegraph transmission theory. Transactions of the AIEE, 47(2), 617-644.
Real Decreto 564/1993, sobre la inclusión de la letra eñe en los equipos informáticos. Boletín Oficial del Estado.
Shannon, C. E. (1938). A symbolic analysis of relay and switching circuits. Transactions of the AIEE, 57(12), 713-723.
Shannon, C. E. (1949). Communication in the presence of noise. Proceedings of the IRE, 37(1), 10-21.
Sony Corporation. (s. f.). Sony history: The compact disc.
W3C. (2003). PNG specification: Data representation.
Xerox. (2013). Software patches for the scanning issue (KB0230451).
YouTube Help. (s. f.). Recommended upload encoding settings.
Next: E02 · The Pile That Reads You Back. Learn AI at TertulIA