Self-Taught Coder Says AI Cracked a 3,500-Year-Old Script
3 min readFor more than a century, Linear A has defeated everyone who tried to read it. Academic linguists, statisticians, and archaeologists have all come up short. Now a self-taught engineer working alone with a laptop and an AI assistant says he has done what the institutions could not.
A Script That Resisted Scholars for 125 Years
Linear A is a syllabic writing system used by the Minoan civilization of Bronze Age Crete between roughly 1800 and 1450 BCE. It survives on about 1,400 short clay tablets and ritual objects. Despite 125 years of study since Arthur Evans first unearthed it at Knossos, no one has been able to read it. Its sister script, Linear B, was cracked in 1952, but Linear A kept its secrets: there is no bilingual text to act as a key, the corpus is tiny, and the underlying language has no confirmed relatives.
One Person, Claude Code, and a Pattern
Tom Di Mino, a self-taught AI engineer and amateur linguist from New York’s Hudson Valley, began digging into the problem in January. He used Claude Code to build a computational engine that cataloged symbols, mapped grammatical patterns, and stress-tested thousands of transliteration hypotheses far faster than any manual effort could.
On May 22 he says he found the key. Di Mino argues that Linear A is not a lost Minoan tongue but an extinct Semitic language, an early relative of biblical Hebrew. He points to a recurring verb built on the root nawaya, meaning to dwell or build, which maps onto the N-W-Y consonant pattern found in Akkadian, Phoenician, and early Hebrew. In his reading, many of the inscriptions are prayers to a goddess.
A Claim, Not Yet a Verdict
It pays to be careful here. Di Mino’s methods and translations are under peer review at Rutgers University and the University of Cambridge, and they have not been confirmed. Most scholars still treat Minoan as a language isolate, and decades of research suggest no tool, human or AI, can prove a Linear A decipherment without a bilingual text or a much larger archive. As one detailed survey of the field notes, AI can generate plausible readings endlessly, but plausibility is not proof.
Why It Matters
Caveats aside, the story captures something real about this moment in AI. A single motivated person, armed with cheap compute and a capable model, can now attempt work that once demanded a funded research center and years of labor. Whether or not Di Mino’s reading survives scrutiny, the method he used will be pointed at other undeciphered scripts soon enough.