摘要

历史学家常面临一个古老难题：如何解读那些作者不明、年代模糊甚至文字残缺的古代文献？如今，人工智能正成为破解这些谜团的新钥匙。以谷歌DeepMind开发的Aeneas模型为例，它通过分析数十万份拉丁铭文，像一位“数字考古学家”般预测文本的年代、地理起源，甚至填补缺失的文字——其原理类似于让神经网络学习罗马帝国的语言规律与文化脉络，从而识别出人类专家难以察觉的跨时空联系。这不仅将复原历史碎片的效率提升数倍（如帮助学者在15分钟内完成原本需数天的研究），更可能通过铭文中奴隶的墓志铭或皇帝的诏令，揭示罗马帝国中人员、信仰与思想的流动网络。尽管AI在填补长段缺失文本时仍有局限，但它的推理能力已接近专家水平。未来，类似的AI工具有望扩展到更多古代文明与语言，重新连接人类遗失的记忆。然而，我们仍需思考：当机器开始参与历史叙事，它是否会重塑我们对过去的理解？

Historians frequently confront an age-old challenge: how to decipher ancient documents with unknown authors, uncertain dates, or fragmented text? Artificial intelligence is now emerging as a powerful tool to unravel these enigmas. For instance, Google DeepMind's Aeneas model analyzes hundreds of thousands of Latin inscriptions, functioning as a 'digital archaeologist' to predict text dates, geographic origins, and even reconstruct missing portions—its methodology involves training neural networks on the linguistic patterns and cultural frameworks of the Roman Empire, enabling the detection of cross-temporal correlations that often elude human experts. This approach not only boosts the efficiency of reconstructing historical fragments by multiple times (e.g., assisting scholars in completing tasks that once required days in just 15 minutes) but also holds potential to uncover networks of human movement, beliefs, and ideologies in the Roman Empire through inscriptions such as slave epitaphs or imperial decrees. While AI remains constrained in restoring extensive missing text, its inferential capabilities are nearing expert proficiency. Looking ahead, analogous AI tools could extend to additional ancient civilizations and languages, bridging gaps in humanity's forgotten heritage. Yet, a critical question persists: as machines contribute to historical storytelling, might they alter our perception of the past?