What can be learned from ancient coded texts
The Vatican’s Apostolic Library houses a 408-page handwritten book dating back more than four centuries, much of which until recently was impossible to read. The text uses 34 unusual symbols mixed with individual Latin letters, and the inscription on the title page is in Arabic. The encryption key, known as the Borg cipher, was lost. In addition, some pages have been damaged due to the age of the book.
how Write “BBC” Scientists were able to decipher the contents of the book using . The text, as stated in the inscription at the beginning of the manuscript, contains recipes for treating “physical ailments.” After decoding, it turns out that, for example, to combat dysentery, it is recommended to drink several glasses of good red wine or nutmeg brewed in dough. Such treatments were kept secret because they might raise suspicions of witchcraft.
About 1% of the materials stored in libraries and archives It may be encrypted. Beata Medgyesi, a professor of computational linguistics at Stockholm University, who worked on cracking the Borg codes, believes that the decryption process can be significantly accelerated with the help of artificial intelligence.
At different times, people have encrypted intelligence, rituals of secret societies, medical knowledge, love correspondence – everything they wanted to keep secret. Deciphering such documents, which are not now in historical records, can change our understanding of the era or of specific historical figures. For example, in 2023, crypto Decode it More than 50 letters written by Mary, Queen of Scots during her imprisonment in England, from 1578 to 1584. From these letters, details of Stuart’s participation in plots against Elizabeth I in order to regain the throne are known, as well as details of her tense relationship with her son, James VI of Scotland, the future King James I of England.
Some ciphers are very simple: each symbol corresponds to a Latin letter. Sometimes several different signs can be used to denote the same letter. There are more confusing options. In addition, additional meaningless characters may be intentionally added to the encryption. In some cases, researchers know nothing about the language in which the ciphertext was originally written.
All this makes their work very difficult. For example, Cécile Pierrot and her colleagues spent about six months Decode it Three pages of text. It was a letter from the Holy Roman Emperor and King of Spain, Charles V, to his ambassador to France, Jean de Saint-Maurice. The Emperor wrote about a possible assassination attempt, which King Francis I of France was believed to be preparing. The message was encrypted using 120 characters. Some of them have replaced entire words.
What does the decryption process look like?
First, the handwritten encrypted document must be converted into a digital format. Two pages, according to Pierrot, can take a whole day: often complicated by illegible handwriting and faded pages. Next, the electronic document is loaded into a special program that tries to find the encryption key.
Artificial intelligence is beginning to accelerate this process. Thus, Michelle Valdespol from the University of Oslo and her colleagues used an online platform Duplicator bus To decipher messages from the era of the Thirty Years’ War (1618-1648). Written by the nobleman Sigismund Heusner von Wandersleben in 1637 to the Swedish Chancellor Axel Oxenstern. He mentioned the threat of a conspiracy among Sweden’s allies.
Trained on texts written in several languages written from the 15th to 18th centuries, Transkribus can recognize handwritten text and translate it into a digital format. The tool successfully processed von Vandersleben’s message, which was encrypted using numbers, although minor intervention from scientists was still required.
Existing transcription tools have difficulty when a document uses non-standard or fictitious characters. Therefore, Beata Medgyesi, Michele Valdespol and their colleagues from different countries are working together to develop a new AI-based tool that can convert such texts into machine-readable documents – Decryption.
Once the original text is converted to electronic format, cryptographers can work to decrypt it using special software (which does not yet use artificial intelligence). Simple ciphers can often be deciphered using symbol frequency analysis: they are compared to letters of the alphabet that occur most frequently in a given language. For example, the most common letter in English is E, while Z, Q, and X are the least common. But in the same von Vandersleben letter, eight different symbols were used to represent the letter E. This cipher can only be deciphered very gradually, by trial and error – and with the participation of someone who knows Old German. Valdespol hopes that artificial intelligence will accelerate this work over time.
What do researchers want to achieve?
Medgyesi and her colleagues are trying to completely remove the copying step from the process and instantly select the code key based on images of the pages. The researchers found that this approach can work with simple ciphers, where one letter corresponds to one letter.
The system was tested on the Codex Copiale, a 17th-century German cryptic manuscript of 105 pages telling about a secret society of the Masonic type. The AI, which was trained on samples of handwriting and images of cipher fonts along with their decryption, was able to decode parts of the text that it had never seen before.
The main challenge facing the Descrypt team is the need to collect enough data to train the AI. Large language models like ChatGPT are trained on trillions of words taken from books, articles, and the Internet. Meanwhile, the scope of findings by researchers working on Descrypt is more modest. For example, they managed to collect 400 postcards from the late 19th and early 20th centuries. Judging by the few fragments that have been deciphered, these are love letters in German.
The Medgyesi team created an AI chatbot that transcribes and decodes text in one step. It also documents the process and explains its decisions, which helps rule out the possibility that the AI is hallucinating. The robot was able to translate and decode a 500-character portion of the Borg code in about half an hour. He also provided an English translation. The system also handled two other ciphers, keys that scientists had previously discovered.
The Descrypt team hopes to be able to crack the ciphers that remain unsolved. The tool will likely help in working with ancient texts written in the alphabet that no one can read today. For example, from Crete, created in the second millennium BC.
“I’m excited not only about the opportunity to solve a specific historical mystery, but also about the prospect of inventing methods that can help researchers solve many other problems,” says Medgyesi.