If you believe in hype, then artificial intelligence (AI) is the future. However, new research suggests that this technology may also improve understanding of the past.
Working with classicists and archaeologists from British universities, a team of computer scientists at Google Deepmind described a new machine learning system designed to help experts understand ancient Latin inscriptions.
Named Aeneas (after the mythical heroes of the Roman Foundation's epic poetry), the system is a generative neural network designed to provide the context of Latin inscriptions written between the 7th century BC and the 8th century CE.
As researchers essentially write, Aeneas can get text-context similarities, utilize visual details, and generate speculative text to fill in the inscription gaps.
A handy and accurate tool
All of these are fascinating prospects for scholars who work in inscriptions (known as epics). Interpret and date fragmentary inscriptions is not easy.
How well does it work? The team behind Aeneas has tried out the system on 23 people with epigraphic expertise, ranging from Masters students to professors.
Participants used the output from Aneas as the starting point in experimental simulations of actual research workflows under time constraints.
In 90% of cases, historians found that similarities in the particular inscriptions obtained by Aeneas were useful starting points for further research. The system also improved reliability for critical tasks by 44%.
When restoring partial inscriptions and determining where they came from, Aen and historians working with the work surpassed both human and artificial intelligence alone. Estimating the age of the inscription, Aeneas achieved the results on average within 13 years from a known date provided by the historian.
The system improves performance over some regions and periods than other regions. Naturally, it performs better on materials from periods where the historian has the most evidence, as well as the most accurately outdated inscriptions.
Nevertheless, participants highlighted Aeneas' ability to expand search by identifying important but previously overlooked similarities and overlooked text features. At the same time, it helped to improve the results to avoid excessively narrow or unrelated discoveries.
Fast way to start
So, how useful is that really? Epigraphy is a challenging field with a long and troublesome history. It takes a long time to develop expertise. Scholars tend to specialize in a particular region or period.
Aeneas dramatically accelerates the preliminary analysis process. Sieve through complex chunks of evidence to identify potential similarities or similar texts that researchers may overlook when dealing with fragmentary material.
Another use of Aeneas is to find inscriptions geographically and estimate when they were generated.
Aeneas can also predict missing pieces of fragmentary text, even if the length of the lost is unknown.
This last feature may seem the most exciting, but perhaps the least practical value of the research is. This corresponds to a speculative recovery by human authority, and has the same ability to draw unfair conclusions.
A convenient starting point
On the other hand, you need to ask what Ayen can't. The answer is real research (for all hype, like all the generator AI products ever).
But the team behind Aeneas is completely open about this. They are rightly and proud of the results the system represents, but they are careful to measure their ability to produce “a useful research starting point.”
Additionally, this tool does not remove the basic need to check the data to extract against standard references.
Researchers with appropriate expertise should perform the task of interpreting the results. It is the viable extent of their work that Aeneas changes.
This allows for a wider view of similarities than typical tools (particularly through their ability to utilize visual cues). The ability to quickly search provides scholars with a much faster starting point of research than before.
Also, for epigraphers, it can open up a much wider field of view, allowing them to escape restrictions on a particular geographical region or period.
