How does AI help a PhD student in history?
In this blog series, Flemish PhD students from various fields of study share how they use artificial intelligence for their research and other tasks.
Lith Lefranc (UAntwerp) kicks off the series, and provides insight into the world of historical research. She learned how to program for her PhD, and is constantly exploring new AI tools. For example, she uses machine learning techniques to automatically search handwritten historical texts.
ID kit
What’s your name?
Lith Lefranc
What field of study are you a PhD student in?
History
What year of your PhD are you in?
Just started my fourth year
Which institution are you attached to?
UAntwerp: the ‘Centre for Urban History’ and the ‘Antwerp Centre for Digital Humanities and Literary Criticism’
Searching through historical police reports with AI
What is your research about?
I study social inequality in Antwerp nightlife in the late nineteenth and early twentieth centuries. A lot changed in the city during that period: modern street lighting was introduced, public transport increased, and nightlife flourished. I analyze police reports to find out who was out on the streets at night, and how factors such as gender, age, ethnicity, and social class influenced their experiences.
How do you use AI in your work?
I used machine learning to search through a vast amount of historical police reports – around 100,000 pages! I started with methods that automatically recognize handwritten text. Then I deployed AI models to extract relevant elements from the texts, such as dates, street names, and personal names.
What is the biggest advantage you gained through AI?
Time savings and upscaling. During the timespan of a PhD, it is unfeasible to manually process 100,000 pages of historical sources. I need about 15 minutes per page for its transcription, which (in a 40-hour week) would amount to 12 years of full-time work. The computer only needed one month! This was of course preceded by an intensive period in which I created the training data and trained the models, but still, the time savings were enormous.
“ Manually transcribing 100,000 pages of historical sources would take me about 12 years. The computer only needed one month! ”
An intensive programming course, and learning from each other
How did you educate yourself to use AI in your research?
In the first three weeks of my PhD, I took an intensive course on the Python programming language - I was allowed to join a course of the UAntwerp Master in Digital Text Analysis. I complemented that course with specific modules from codeacademy. I also contacted the core facility TEXTUA: they provided support for the computer models I used to extract dates and names from my sources. For other AI applications, I am incredibly lucky that my co-supervisor Mike Kestemont is amazing at everything related to computational humanities. He set up his own lab, consisting of several PhD students who all use computational methods. We learn a lot from each other!
What is your favourite way to learn something new?
I prefer to learn on the job. And in terms of timing? I tackle specific problems as soon as they pop up in my research and I need to find a solution!
How often do you like to upskill?
That depends on the tool. Sometimes a one-off session suffices, sometimes a more intensive full-week training is the better option.
Whose tips for training do you follow?
I haven't had any digital pre-training, so my supervisor sometimes slightly overestimates my prior knowledge and underestimates my need for training. He guides me very well to the right tools, but after that, I often rely on myself to figure out how they work (usually via blogs, videos, or trial-and-error).
AI wishlist
What task would you like to be able to outsource to AI?
I find creating training data and reviewing test data agonizingly boring. Unfortunately, current AI applications are not yet reliable enough to outsource this completely.
What do you want to learn more about?
In the current stage of my research, I mainly want to learn about digital tools for statistical data analysis.
This or that: AI edition
Do you prefer an AI tool that...
- summarizes 100 papers, or writes the first draft of your paper?
Summarizes 100 papers - finds you the perfect source, or creates your conference presentation?
Finds the perfect source - cleans up your messy dataset, or creates clear visualizations of your data?
cleans up my messy dataset - gives feedback on student papers, or corrects exams?
corrects exams - comes up with great research ideas, or writes perfect project proposals?
writes perfect project proposals
The future of history and AI
Has your discipline changed much because of AI?
Within my department, AI is not used that much. At conferences in digital humanities, historians are also in the minority. I feel that history is still lagging a bit behind compared to other humanities disciplines, such as linguistics and literature.
Do you think AI will positively influence your job or scientific field?
Yes and no. From my modest experience with AI, I believe it can mainly contribute in terms of time savings and upscaling, but that’s all for now. For creating training data, making analyses and interpreting them, human researchers will remain indispensable.
What advice would you give to researchers who are hesitant to use AI?
Don’t be put off! The learning curve is very steep at first, but this is outweighed by the benefits in terms of scale and efficiency. Always combine research based on AI-generated datasets with more classic, ‘manual’ datasets (this could be your training or testing data, for example). This allows you to familiarize yourself with your data, and you always have a back-up in case the results of the AI tools are disappointing.
“ Don’t be put off. The learning curve for AI is very steep at first, but this is outweighed by the benefits in terms of scale and efficiency! ”
Read and learn more on this topic
Generative AI for Professionals in Higher Education
learning trajectory - VAIA & VLIR
AI to Understand and Connect People
online course - KU Leuven Humanities & Social Sciences, AI4Culture, VAIA
AI training for PhD students. What is needed?
AI is changing the work of PhD students: from the impact on research methods to the writing processes. As a result, PhD students need AI training courses. Is the current offer sufficient? Which courses exist already, and which are missing? And can students find their way towards them?
How is AI Transforming the Learning Process in Academia?
Artificial Intelligence (AI) is revolutionising education, from primary schools to PhD programmes. But is it really AI that’s upsetting the PhD education system? Or is the technology mainly exposing deeper, underlying issues? Find out more in this blog by Vicent Botella-Soler.
Lith Lefranc
Lith Lefranc works as a PhD researcher at UAntwerp, under the supervision of Ilja Van Damme and Mike Kestemont. She studies the impact of urban modernisation processes on the social stratification of Antwerp nightlife from a data-driven perspective. Using local police reports (1870-1940), she tries to understand the social dynamics that shaped the nightscape in the past and hopes to create new visions of nocturnal street use.
Related courses
SAIAR Summer Studio: From latent to physical space
Zomerschool - Kortrijk - Howest Hogeschool, SAIARlab
Fundamental concepts of multi-omics data integration
Opleiding - Leuven - Genomics Core Leuven
HPC Café
Seminarie - Online, Karlsruhe - KIT, HammerHAI, Training geselecteerd in de Belgische AI Factory Antenna (AIFA)