药典、阴谋与尘封往事:AI技术破解千年古密码
阴谋、药方与隐秘历史：AI正在揭开的中世纪秘密
导语：全球各地的图书馆与档案馆内，沉睡着无数被复杂密码封存的历史文献。如今，人工智能打破百年来的解密困局，揭开中世纪隐匿的医疗秘方、权力斗争与历史真相，让尘封的细节重现人间。
Across libraries and archives worldwide lie countless historic manuscripts obscured by indecipherable ciphers. Now, artificial intelligence is empowering historians to crack these mysterious coded texts and uncover long-lost secrets of the past.
全球图书馆与档案馆中，保存着大量被难解密码加密的历史文献。如今，人工智能协助历史学家破解这些神秘文字，揭开被岁月掩埋的历史真相。
梵蒂冈尘封四百年的密码文书
The 400-year-old unsolved mystery in the Vatican Archives
Deep in the archives of the Vatican Library rests a mysterious handwritten manuscript scrawled with bizarre symbols, untouched and unread for more than 400 years. Inscribed faintly on its cover are hints of its content: secret remedies for all manner of human ailments.
在梵蒂冈图书馆深处，保存着一本布满奇异符号的手写古书，四百年来始终无人能读懂。封面残存的文字显示：此书记载着各类疾病的神秘疗法。
In the medieval era, such folk healing practices were strictly concealed. At the time, unorthodox medical treatments were easily deemed suspicious, and practitioners often faced accusations of witchcraft, forcing such knowledge to be kept secret.
在中世纪，这类民间医疗手段属于绝对禁忌。偏离主流的疗法极易招致怀疑，医者甚至会被诬陷为巫师，因此这些知识只能秘密保存。
Known as the Borg Cipher, this 408-page manuscript has long baffled scholars. It is encoded with 34 rare and obscure symbols, interspersed with a few Roman letters, and features an Arabic preface. For centuries, no decryption key existed to decode its content. Worse still, age-related damage to the pages has made this ancient code even more inscrutable.
这份长达408页的手稿被后世称为《博尔吉亚密码》，长期以来让学者们一筹莫展。全书以34种罕见符号作为主要加密方式，零星夹杂罗马字母，首页以阿拉伯文书写。数百年来，世人始终找不到解密钥匙，加之书页因年代久远而破损老化，破解难度不断攀升。
Thanks to machine learning technology, a core branch of artificial intelligence, researchers have finally unravelled the manuscript’s secrets. The decoded text reveals thousands of strange medieval remedies, ranging from drinking multiple glasses of fine red wine to cure illness, to fermenting nutmeg in dough to treat dysentery.
依托人工智能核心技术——机器学习，研究人员终于揭开了这份密码的秘密。译文揭示了数千种奇特的中世纪疗法：饮用数杯优质红酒调理身体、将肉豆蔻混入面团发酵治疗痢疾等，皆是古人的独特治病偏方。
“Deciphering ancient codes is like detective work. Every symbol, every repeated pattern and every partial solution brings us closer to individual stories and a lost historical world,” said Beáta Megyesi, professor of computational linguistics at Stockholm University and a core member of the decryption team. She stressed that even with AI assistance, cracking the cipher remained an extremely painstaking process.
瑞典斯德哥尔摩大学计算语言学教授、本次解密团队核心成员贝娅塔·梅杰西指出：“破译古代密码如同侦探查案，每个符号、每处规律、每次局部突破，都让我们更接近古人的故事与失落的历史世界。”她强调，即便有AI帮助，密码破解过程仍然极其繁琐艰难。
Currently, Professor Megyesi and her team are leading global efforts to apply AI technology to ancient cipher decryption, aiming to efficiently unlock countless undecoded coded records that have been sealed by history.
目前，梅杰西教授正带领团队在全球范围内推进AI古密码破译研究，致力于高效解锁海量被历史掩埋的加密记录，让无数失传的历史信息重见天日。
被密码封存的真实历史
The untold history hidden in coded documents
According to academic estimates, approximately 1% of archived documents worldwide are fully or partially encrypted. The earliest known ciphers date back to Ancient Greece and Ancient Rome, carrying the private memories and secret information of countless eras.
据学术界估算，全球档案馆与图书馆中，约1%的文献存在全部或部分加密情况。人类最早的密码可追溯至古希腊、古罗马时期，无数时代的隐秘信息都封存在这些加密文字中。
These coded manuscripts cover diplomatic intelligence, secret society rituals, exclusive medical knowledge and ordinary people’s hidden daily stories — all of which are missing from official historical records. In many cases, decoding these texts can reshape our understanding of famous figures and entire historical periods.
这些加密文献内容广泛：外交情报、秘密社团仪式、独家医学知识，以及古人不愿公开的日常琐事。这些信息恰好填补了正史的空白，在许多情况下，解读这些文字能重塑我们对历史人物和整个时代的认知。
A typical example is the coded letters of Mary, Queen of Scots. Discovered during modern decipherment, these letters, written during her long imprisonment in England, confirmed her involvement in throne-reclaiming plots and exposed her strained relationship with her son James VI of Scotland (later James I of England).
苏格兰玛丽女王的加密书信便是典型案例。这些写于她英格兰囚禁期间的密信，经现代解密后发现：她曾多次策划夺权行动，同时也暴露了她与儿子——苏格兰国王詹姆斯六世（后英格兰国王詹姆斯一世）之间紧张疏离的母子关系。
古密码破译的重重挑战
The immense challenges of ancient code-breaking
Ancient ciphers vary greatly in complexity. The Borg Cipher adopts a simple substitution mechanism, where each unique symbol corresponds to a single Roman letter. Yet most medieval codes are far more intricate and difficult to crack.
古密码的破解难度差异极大。《博尔吉亚密码》采用简单替换机制，即一个符号对应一个罗马字母。但绝大多数中世纪密码结构更为复杂，破解难度更高。
Many encrypted texts face three major obstacles: unknown original languages, meaningless decoy symbols inserted to confuse snoops, and multiple symbols representing the same letter. These barriers force cryptologists to rely on endless trial and error, resulting in extremely low decoding efficiency.
多数加密文献面临三大障碍：原文语言早已失传、文中插入无意义干扰符号迷惑窥探者、多个符号对应同一个字母。层层障碍使密码学家只能不断尝试，传统人工解密效率极低。
French cryptologist Cecile Pierrot and her team spent six months deciphering a 500-year-old three-page letter from Charles V, Holy Roman Emperor and King of Spain. The letter used 120 distinct cipher symbols. The decrypted content revealed a surprising truth: one of the most powerful rulers of his era was consumed by fear of assassination by an Italian mercenary serving the French king Francis I.
法国密码学家塞西尔·皮埃罗曾带领团队耗时六个月，才逐步破译神圣罗马帝国皇帝、西班牙国王查理五世的三页密信。这封五百年前的书信使用了120种不同的密码符号。破译结果令人震惊：这位当时欧洲最有权势的君主，竟极度恐惧被法国国王弗朗索瓦一世手下的意大利雇佣兵首领刺杀。
Before any decryption work begins, researchers must manually transcribe blurry, handwritten ancient cipher manuscripts into digital text compatible with decoding software. Faded ink, messy handwriting and aged paper damage make this basic work extremely arduous. Pierrot admits that transcribing just two pages of unfamiliar coded content usually takes her a full working day.
在正式解密前，研究人员必须先将字迹模糊、手写潦草的古老密文手动转录为可被解密软件识别的电子文本。墨迹褪色、字迹潦草、纸张老化，使这项基础工作极其困难。皮埃罗坦言，仅转录两页陌生的加密内容，通常就要耗费她一整天。
AI重塑古密码破译方式
How AI revolutionises ancient cipher decryption
Artificial intelligence is fundamentally changing this inefficient traditional process, greatly accelerating the transcription and decryption of ancient coded texts.
人工智能从根本上改变了这一低效的传统流程，大幅加快了古密码文献的转录与解密速度。
Norwegian linguist Michelle Waldispühl and her team used the AI platform Transkribus to decode a 1637 secret letter. Written by nobleman Sigismund Heusner von Wandersleben to Swedish Lord High Chancellor Axel Oxenstierna during the Thirty Years’ War, the letter contains crucial wartime intelligence.
挪威语言学家米歇尔·瓦尔迪斯普厄尔团队，借助AI平台Transkribus成功破译了一封1637年的战时密信。这封信由贵族冯·万德斯莱本撰写，写给瑞典首席大臣阿克塞尔·奥克森斯蒂尔，记载了三十年战争期间的关键军政情报。
Trained on centuries of multilingual scripts and diverse handwriting styles, the AI platform can automatically identify text blocks and lines from document images, then convert characters into standard digital text. Though minor manual corrections were required, the tool efficiently processed the partially encrypted letter, which combined dot-separated numerical codes and 17th-century German script.
该AI平台经过数百年多语种文字、各类手写字体的海量数据训练，可自动识别文献图像中的文本区块与行列，逐字转化为标准电子文本。这封密信部分由圆点分隔的数字加密、部分为17世纪德文手写体，AI仅需少量人工修正，便高效完成了转录工作。
Existing mainstream AI transcription tools struggle with rare custom symbols, astrological marks and irregular coded numbers. To break this bottleneck, scholars from across the globe have jointly launched the Descrypt project, developing a dedicated AI tool for obscure ancient handwritten coded texts.
现有主流AI转录工具难以识别小众自创符号、占星标记、异形加密数字等特殊密码。为突破这一技术瓶颈，多国学者联合启动Descrypt国际项目，专门研发适配各类生僻手写古密文的AI破译工具。
“We are building highly adaptable AI models trained and tested on a wide range of ancient scripts, alphabets and symbolic systems,” said Professor Megyesi.
梅杰西教授介绍：“我们正在构建适配性更强的AI模型，通过海量古文字、古字母、古符号数据完成训练与测试，适配各类小众加密文本。”
After AI completes transcription, researchers use professional decryption algorithms to analyse symbol frequency — a core decryption logic. For instance, the letter “E” is the most frequent in English, while “Z”, “Q” and “X” are the rarest. Yet complex ancient ciphers break this rule.
AI完成文本转录后，研究人员会依托专业解密算法，通过符号出现频率分析破解密码，这是解密的核心逻辑。例如英文中字母E出现频率最高，Z、Q、X最为罕见。但复杂的古密码会刻意打破这一规律。
In the Thirty Years’ War letter decoded by Waldispühl, the writer used eight different symbols to represent the single letter “E”. Decryption thus required repeated human-machine verification and the researcher’s professional knowledge of ancient German.
在瓦尔迪斯普厄尔破译的三十年战争密信中，作者用了八种不同符号对应同一个字母E。这意味着解密工作必须依托人机反复校验，同时结合研究者的古德语专业知识才能推进。
“The decryption process relies heavily on constant interaction between machine computation and human verification. We believe AI will eventually complete the work independently,” Waldispühl noted.
瓦尔迪斯普厄尔表示：“目前的解密工作高度依赖机器运算与人工校验的反复配合。未来，AI有望实现全自主解密。”
The decrypted letter revealed critical wartime secrets: the writer was forced to strategically retreat after uncovering a conspiracy among Sweden’s Protestant allies, including Lord Franz Heinrich of Saxony.
这封密信的破译结果揭露了关键战时秘密：写信人发现萨克森公爵弗朗茨·海因里希等瑞典新教盟友暗中密谋，无奈之下只能战略性撤军。
AI开启历史解密无限可能
AI’s boundless potential for unsolved historical codes
The Descrypt team is pushing technological boundaries, aiming to skip the manual transcription step entirely. Their new AI system can directly decipher coded content from manuscript photos, integrating transcription and decryption into one single process.
Descrypt项目团队正在突破技术边界，试图彻底省去人工转录步骤。全新的AI系统可直接扫描古籍图片破译密文，实现“图像识别-文本转录-密码解密”一站式完成。
The team has verified the technology’s feasibility through the Copiale Cipher, a 105-page 18th-century manuscript recording the rituals and doctrines of a German secret society. After training on partial decoded content, the AI accurately deciphered unread sections of the manuscript. This technology is especially suitable for ciphers with unknown original languages.
团队以105页的18世纪德国秘密社团《科皮亚莱密码手稿》为测试样本，验证了技术可行性。AI通过学习手稿部分已破译内容，可精准解读剩余未知文本。这项技术对原文语言失传的冷门密码尤为适用，破解优势显著。
To improve AI’s code-breaking accuracy, researchers are combing global archives to build a dedicated ancient cipher database. The team collects massive volumes of historical coded manuscripts of different eras and types, providing sufficient data support for AI model training and optimization.
为提升AI破译精准度，研究团队遍历全球古籍档案馆搭建专属古密码数据库，海量收录不同时代、不同类型的历史加密手稿，为AI模型训练与迭代优化提供充足的数据支撑。
More importantly, the team has developed an AI chatbot-style decryption tool. It integrates decryption algorithms trained on massive cipher-text pairs, historical large language models and annotated handwriting image recognition technology. The tool can also self-optimise by learning from expert revisions.
更重要的是，团队已研发出聊天机器人式的一站式解密AI工具。该工具融合了密码配对训练的解密算法、适配多时代的历史大语言模型，以及标注手写体图像识别技术，还能通过学习专家修正方案持续迭代优化。
In tests on the Borg Cipher, the AI tool decoded and translated a 500-symbol excerpt in just 29 minutes, outputting accurate English translations and detailed, credible decryption logic — effectively avoiding AI’s common “hallucination” problem of fabricated interpretations.
在针对《博尔吉亚密码》的测试中，该AI工具仅用29分钟就完成了500个密码符号的破译、释义与英文翻译，同时输出完整可信的解密逻辑，有效规避了AI解读常见的“凭空杜撰”幻觉问题。
The tool also successfully decoded two other ciphers of different eras, languages and complexity levels, proving its universal adaptability for diverse ancient coded texts.
研究团队还选取了两个不同时代、不同语种、不同难度的古密码进行测试，AI均快速精准完成破译，证实了这款工具对各类古加密文本的通用破译能力。
“AI excels at large-scale processing, rapid computation, pattern mining and task integration,” said Megyesi. It brings unprecedented opportunities for unsolved ancient texts.
梅杰西总结道：“AI的核心优势在于大规模处理、极速运算、规律挖掘与任务整合。”这为无数未解的古文字谜题带来了前所未有的破解机遇。
For millennia, iconic ancient texts including the 4,000-year-old Phaistos Disc and the ancient Greek Linear A script have remained undecipherable to humans. With the help of AI, these frozen historical secrets may finally be awakened.
数千年来，克里特岛4000年历史的费斯托斯圆盘、古希腊线性文字A等世界级古文字难题始终无人能解。依托AI技术，这些被时光封存的历史秘辛终将迎来被解读的一天。
“What excites me most is not just solving a single historical puzzle, but building a universal method to assist researchers in unlocking countless hidden histories,” Megyesi added.
“最让我振奋的，不是破解某一个历史谜题，而是我们正在搭建一套通用方法，助力学界解锁千千万万被密码封存的过往。”梅杰西如是说。