This week, there has been a lot of publicity around the latest development in the efforts to extract texts from the fire-damaged scrolls from Herculaneum. News outlets tend to focus on the role of AI in the process, not distinguishing between various forms of technology that fall under the broad label of artificial intelligence. As has been pointed out by several commentators on social media, however, the role of artificial intelligence consists of distinguishing ink from papyrus, leaving the actual transcription and identification of the content of the scrolls to human agency. In other words, the main part of the work is done by human researchers with expertise and knowledge. In a post by Dr. Naomi Scott of the University of Bristol, she pointed out that the main work is done by human experts, and posted a quotation from Professor Llewellyn Morgan of the University of Oxford published in a news update from the university in 2025.
In the quotation from Llewellyn Morgan, there was one particular passage that stood out to me. Morgan likened the pattern recognition by machines to how eighteenth-century copyists who copied what was written on scrolls, but who could not understand the content. This reminded me of an anecdote from my experience with manuscript fragments that really made me understand how experts often need inexpert eyes to help their work progress.
About nine years ago, I was researching fragments of medieval manuscripts in the special collections at the University Library of Southern Denmark. Most of these fragments were used as bindings, and had been subject to a lot of wear, tear, and years of occasional water damage. Consequently, the texts on these fragments could be very difficult to decipher, as it was often challenging to distinguish a letter from a letter-shaped smudge. In some cases, I also came to realise that my expertise in palaeography and Latin hampered me when the text became too illegible, because I was too focused on the word in its entirety rather than the individual letter, which meant that I often kept searching for one of many possible words that seemed like reasonable option for one specific string of letters.
In order to better identify the letters, I took close-up photographs of the fragments or parts of the fragments, printed these out, and traced or separated the various letters with a pencil. This method was sometimes successful, but it depended entirely on the state of the fragment, and I was again hampered by the expectations which were leading my gaze.
I showed some of the pictures to a friend of mine. At that time, he had no experience with medieval manuscripts, palaeography, or Latin. But he is a very good photographer with a keen eye for details, and I thought that he might see things that I did not. This turned out to be the case. As my friend looked at the photographs, he told me very simply what he saw, and I then could focus on that one thing and determine whether this was a letter or simply a letter-shaped smudge. This was a kind of a refocusing, or a recalibration of the gaze, and it was an immense help. In some cases, what my friend saw was irrelevant, but I could only assess this irrelevance after a closer scrutiny, and when transcribing a worn and faded fragment, it is just as important to know which things to ignore. Through his help, I was able to transcribe some lines that had frustrated me for a long time, and I was then able to identify the text itself, which in turn was very important for the general identification of the fragment.
The exercise was a revelation for me, because it showed very clearly how much help I could get from someone who was not in my academic field or even close to it. In this sense, my friend was doing much of the same job that machines are doing for the Herculaneum scrolls. He was finding patterns, noticing things that I overlooked because I was distracted by the bigger picture. And even though this kind of work would have been done much faster by a machine, the process was also very educational for me and helped me not only to do my work but also to improve my skills as a fragment researcher. As in most things, such research requires a lot of practice, and the kind of slow, incremental progress that my friend and I were doing was what I needed to improve at an early stage in my career as a researcher.
In many cases, machines are needed to deal with certain aspects of source work, at least for speeding up the process. But sometimes speed is neither of the essence nor very helpful, because it is the slow, trudging pace that helps you improve, making you become an expert through constant practice. In short, human agency is not only still very much needed, it is also to be preferred.
Ingen kommentarer:
Legg inn en kommentar