Why the Smartest AI on Earth Can't Read the Indus Script

Support the channel - https://ko-fi.com/withronny For fifty years we have been building a machine that can read almost anything — trillions of words, two hundred languages, a Roman scroll baked solid by Vesuvius, the shape of two hundred million proteins. Point it at a code and the code breaks. So point it at four little marks scratched into a stamp of soap-stone four thousand years ago. Nothing. The Indus script — around 400 signs, 4,000 surviving inscriptions, none longer than seventeen signs — has humbled the codebreakers of Bletchley, the crackers of Egyptian hieroglyphs, and now the full weight of modern AI. There is one million dollars (about ₹8.57 crore) on the table for anyone who can read it, put up on 5 January 2025 by M.K. Stalin, the Chief Minister of Tamil Nadu. It has not been claimed. This is the story of why — from the civil-servant hero who mapped the whole script on his evenings and weekends, to the Finnish scholar who thinks a fish means a star, to the physicist who ran it through the same next-symbol maths that powers ChatGPT, to the three scholars who say there was never anything to read at all. And the honest, uncomfortable answer at the frontier of knowledge: we still don't even know what kind of thing it is. Part 1 of a two-part series. Next time, we open the 5,000-year-old ledger and follow the money out of the Indus — all the way to Mesopotamia. 0:00 The machine that read almost everything 2:18 A $1,000,000 code nobody can crack 3:08 Look at it: one Indus seal 4:09 Four thousand texts, five signs each 5:28 Seventeen signs — the longest sentence they left us 7:01 The Dholavira signboard 8:05 The faces we named: Priest-King & Dancing Girl 9:46 The hero: Iravatham Mahadevan 11:25 The first wall: no Rosetta Stone 12:31 Parpola & the fish that means "star" 14:15 2009: the entropy test 16:32 Why the AI can't read it 18:03 "It was never writing" — the collapse thesis 20:12 We don't know if it's even writing 20:54 The most sensitive fault line in South Asia 23:15 The real achievement: the cities 24:16 The prize, and a civilisation undefeated 26:05 Next time: the 5,000-year-old ledger Sources & further reading - Iravatham Mahadevan, The Indus Script: Texts, Concordance and Tables (1977); 417 signs catalogued; Padma Shri, 2009. Asko Parpola, Deciphering the Indus Script (1994) — the Dravidian "fish = meen = star" rebus reading. Rajesh P. N. Rao et al., "Entropic Evidence for Linguistic Structure in the Indus Script," Science (2009) — conditional-entropy analysis. (Rao explicitly did "not" claim a decipherment.) Steve Farmer, Richard Sproat & Michael Witzel, "The Collapse of the Indus-Script Thesis: The Myth of a Literate Harappan Civilization," EJVS (2004) — the "not writing" argument; plus Sproat's entropy rebuttal. K. Rajan & R. Sivananthan — reported overlaps between Indus signs and later Tamil-Nadu pottery graffiti. On the 2025 "Sanskrit decipherment" claim (Yajnadevam): "unverified, not peer-reviewed" — see critiques by Nityanand Mishra and others. Key objects: the "Pashupati" seal (M-304) and longest single line (M-355), Mohenjo-daro; the Dholavira signboard, Gujarat; the Priest-King and Dancing Girl. Every on-screen claim is sourced above; where the scholarship is contested (the fish-star reading, the language family, the "is it writing" question) the video says so out loud.