AI system Aeneas deciphers ancient Latin inscriptions for historians
Generative artificial intelligence, often hyped as the technology of the future, is now proving instrumental in deepening our understanding of the past. A groundbreaking new machine learning system, developed by computer scientists at Google DeepMind in collaboration with classicists and archaeologists from universities in the United Kingdom and Greece, is set to revolutionize how experts interpret ancient Latin inscriptions.
Named Aeneas, after the mythical hero central to Rome’s foundational epic, this sophisticated system is a generative neural network designed to provide crucial context for Latin inscriptions dating from the 7th century BCE to the 8th century CE. As detailed by the researchers in Nature, Aeneas excels at retrieving textual and contextual parallels, integrating visual details from inscriptions, and even generating speculative text to fill missing gaps in fragmented historical records. These capabilities offer significant advantages for epigraphers – scholars dedicated to the study of inscriptions – who often face the arduous task of interpreting and dating incomplete artifacts.
To assess Aeneas’s efficacy, the development team engaged 23 epigraphic experts, ranging from master’s students to seasoned professors, in an experimental simulation of real-world research workflows under time constraints. The results were compelling: in 90% of cases, historians found the parallels retrieved by Aeneas to be highly useful starting points for their investigations. Furthermore, the system boosted their confidence in critical tasks by an impressive 44%. When it came to restoring partial inscriptions and determining their geographical origin, historians utilizing Aeneas demonstrably outperformed both human experts working alone and artificial intelligence operating in isolation. For estimating the age of inscriptions, Aeneas achieved remarkable accuracy, with its predictions averaging within 13 years of known historical dates. While its performance varies, proving most effective for periods and regions with the most abundant and accurately dated historical evidence, participants lauded Aeneas’s unique ability to broaden searches, unearthing significant yet previously overlooked parallels and textual features. Concurrently, it helped them refine their findings, avoiding overly narrow or irrelevant conclusions.
Epigraphy is a notoriously challenging field, demanding years of specialized expertise, often confined to specific regions or time periods. Aeneas promises to dramatically accelerate the preliminary analysis phase of research. It can swiftly sift through vast, complex bodies of evidence to identify potential parallels or similar texts that a human researcher might easily miss when confronted with fragmentary material. Beyond textual analysis, Aeneas can also assist in geographically locating an inscription and estimating its period of creation. The system’s capacity to predict missing parts of a fragmented text, even when the length of the lost portion is unknown, might seem its most exciting feature. However, this function, akin to speculative restoration by a human authority, carries an equal risk of leading unwary scholars to potentially unsafe conclusions.
It is crucial to understand what Aeneas, like all generative AI products to date, cannot do: perform the actual research. The development team is transparent about this, emphasizing that the system’s primary value lies in yielding “useful research starting points.” This tool does not negate the fundamental necessity for scholars to rigorously verify the extracted data against standard references, images, or, ideally, the original artifacts themselves. Researchers with appropriate expertise will still be indispensable for interpreting the results. What Aeneas fundamentally alters is the feasible scope of their work, enabling a much broader view of parallels—especially through its ability to integrate visual cues—than previously available tools. Its rapid retrieval capabilities will bring scholars to their research starting points far more quickly, potentially opening up broader horizons for epigraphers and allowing them to transcend the traditional limitations of specific geographical regions or historical periods.