AI - unlocking new potential in material we already have
It has now been around two years since I first started using AI as part of my teaching, and the technology still feels very much in an exploratory phase. In many ways, it reminds me of the early days of the internet — a time when nobody quite knew what it would be good for. Since those days of Hotmail, when sending a digital postcard to a friend felt revolutionary, digital technology has quietly become woven into everyday life. I suspect the same will happen with AI once the current hype has settled.
For researchers, however, I think the possibilities are already here — especially through the development of Handwritten Text Recognition (HTR), a technology that could fundamentally change how we approach historical research. HTR does exactly what its name suggests: it turns handwritten text into machine-readable text. The first steps in this direction were taken as early as the 1960s, when computers learned to read printed text with clearly separated characters. Yet, it is only with the advent of AI and neural networks that we have seen the technology become truly powerful.
While the primary uses of HTR may not lie in historical archives — it is also valuable for digitising handwritten notes or supporting the visually impaired — its potential impact on historical research is profound. When applied to older manuscripts, HTR has the power to reshape how we think about writing history itself.
One of the main challenges for historians working with pre-19th-century material is not just the language (Old Swedish from the 14th century is quite different from today’s Swedish) but the form of the material. Reading old handwriting is difficult and time-consuming, often requiring specialised training. Even more problematic is the fact that much archival material remains unindexed and largely unsearchable. As a result, vast amounts of text lie dormant — unread and underused — simply because we don’t know where to look.
This is where HTR could become a genuine game-changer.
The Swedish National Archives have recently released large collections of older material. The big difference compared to previous data dumps is that this material is now searchable, and search results are presented both as machine-interpreted text and as images of the original handwriting. For the first time, we not only have access to the material but can actually find what we are looking for — something that would have been hard to imagine just a decade ago.
Another fascinating aspect is how the National Archives trained their models. The training data came from their own collections and was transcribed by volunteer genealogists, meaning that this approach partly avoids the ethical challenges often associated with data annotation, where low-paid workers in the Global South are exploited.
All in all, this development may very well be a turning point — not only for digital historians but for anyone interested in how technology can open up the past in entirely new ways.
Comments
Post a Comment