AI Engineer Claims to Have Cracked Linear A
An AI engineer claims to have deciphered portions of the ancient Minoan script Linear A using custom tools built with the Claude large language model, proposing that it encodes an extinct Semitic language related to Biblical Hebrew. Commenters highlight both the potential significance of such a breakthrough and the long history of failed amateur decipherment attempts, stressing that no manuscript or data has yet been made public and that experts at major universities have not formally validated the work. The exchange broadens into questions about how much AI can genuinely contribute to cracking undeciphered scripts given tiny corpora and unknown underlying languages, and what rigorous verification would need to look like.
Overall claim and status
- Thread discusses a claimed decipherment of Linear A as an extinct Semitic language (possibly ancestral/related to Biblical Hebrew).
- Researcher reportedly has a draft manuscript shared with linguists at major universities; no preprint or full data is public yet.
- Many commenters stress this is, at best, an interesting attempt and should not be treated as a solved decipherment until experts review it and further evidence appears.
Role of AI tools
- An AI coding assistant was used mainly to:
- Build Python tools to parse and organize the Linear A corpus (GORILA, SigLA).
- Run large numbers of simulations to test whether a Semitic signal is statistically stronger than chance.
- Multiple commenters emphasize the human did the conceptual work; the AI sped up coding and analysis.
- Others argue the achievement is inseparable from modern tooling; without AI, the work might not have been feasible for a non‑specialist.
Reproducibility and skepticism
- Strong pushback on the lack of a public manuscript, code, or translation tables (despite claims of ~300 translated words).
- Concerns that LLM-based workflows are “softly reproducible” due to stochastic outputs and prompt sensitivity.
- Some liken this to past high‑profile “breakthroughs” that collapsed under scrutiny (e.g., other Linear A claims, room‑temperature superconductors).
- A classics-focused community elsewhere is reported as dismissive of the claim.
Linguistic and technical issues
- Linear A is notoriously difficult:
- Tiny corpus (~7,500 characters across ~1,500 mostly list-like inscriptions).
- Possibly multiple languages and heavy use of abbreviations.
- Unknown underlying language and incomplete understanding of the script.
- Debate over:
- Whether assuming Linear A signs share values with Linear B is justified.
- How much can be inferred from recurring formulas like the “libation formula.”
- Whether mapping fragments of words to Semitic roots is convincing or just cherry-picking amid huge degrees of freedom.
- Why a Semitic language would be written with a CV syllabary including vowels, given typical consonantal Semitic scripts.
Broader implications
- Some see promise that AI-augmented methods could help with other undeciphered scripts (e.g., Indus, Isthmian, Voynich).
- Others argue information-theoretic limits may make fully cracking an unknown language/script pair impossible without more data or bilingual texts.