A computer scientist specializing in artificial intelligence (AI) applications for textual analysis, Prof. Jonathan Schler is finding new ways to harness this powerful tool to shine new light on historical and literary quandaries, counter disinformation and improve human well-being. We met up with Prof. Schler to hear about what is brewing in his lab these days.
My main focus lies in authorship analysis, which means using computational tools to understand various aspects about the author of a given book or document. The focus is usually on who wrote the document and less about what is written in it. Traditionally, there are three main areas in authorship analysis: authorship attribution, authorship verification and authorship profiling.
In authorship attribution, we try to identify who wrote a given text. In contrast, authorship verification deals with confirming whether or not a specific individual wrote a given text. These issues are relevant to fields like criminology and plagiarism, but I am more interested in investigating authorship of historical literary texts.
My team is using AI to try and identify the unknown writers of Jewish texts based on textual attributes. One such study is focused on quotations and references, which are prevalent in Jewish religious writings. We can use them to trace an unknown author of a book. Our unique approach, which we are about to publish, involves analyzing the quotes and citations in a given book. We trace all these quotes, and assemble the library of books referenced, which is the library that was in front of the author. Then, we use this library to see if there are books with unknown authorship that may have used the same reference set. If any books are identified, they could serve as as a strong directive in determining the authorship of previously unattributed texts.
A recently completed project focused on Burchard of Mount Sion, a German Christian priest who wrote a travel diary about his visit to the Middle East (including Israel) in the 13th Century. There are two early versions of this diary, but they are not exactly the same. My team, with the aid of Prof. Yoni Rubin, a historian at Ben Gurion University of the Negev and a content expert, has attempted to put together the original version from these two notebooks. We believe we have assembled a version that is 90% accurate, but we would, of course, need Burchard of Mount Sion to confirm this. This project showcases how AI can be used to retrace original texts.
We are also looking at the writings of Josephus (Yosef ben Matityahu, or Titus Josephus Flavius), one of the primary sources of knowledge about historical events during Titus’s reign, including the Jewish-Roman War and the destruction of the second Jewish temple. Historians have long wondered how much of these texts did Josephus write as a first-hand witness and which parts were, in fact, relayed to him by others or copied from other documents. We are developing, together with with the e-Lijah Lab at Haifa University, an AI language model that will shed new light on this issue.
In authorship profiling, we use AI textual analysis tools like natural language processing (NLP) to gain knowledge about the author of a text—what he wrote, his writing style. This allows us to characterize an unknown author’s attributes, like gender, age, native language, or psychological profile. NLP can also be used to search for changes in writing styles that may indicate, for example, whether a person has become depressed due to a life-changing event.
It is indeed. I will give you a few examples from my work. During the COVID pandemic, when students didn't come to class, we were interested to ascertain whether we could devise an AI model that identifies from student’s casual written communications, let’s say with our technical support team, whether they were in emotional distress. Impressively, we were able to reach an 85% level of accuracy, which is on par with typical psychological assessments.
In another COVID-related study, we examined the effects of that period on individuals with mental health issues. Interestingly, we found a positive effect on people who suffer from social anxiety. Their levels of anxiety and tension decreased, because everyone was basically in quarantine. We were able to identify such trends due to changes in the study participants’ writing style. Tools of this sort are now being developed to monitor emotional distress online. These could be complimented by ChatGPT-style apps designed to engage individuals in emotional distress and assist them reduce their stress levels.
We also developed an AI model that helps identifying bots on social networks. We have tested this model on Facebook during the previous series of Israeli elections and the information we gained enabled us to define “bot red warning” signs.
In our newest profiling project, we are analysing social media posts to ascertain the effect of the October 7 terror attack and ensuing war on people’s state of mind, like depression.
The field is evolving rapidly. I think that we will see a growing number of specialized AI-based applications, which will eventually lead to the automation of processes. The tasks AI apps will do will become more complicated, which will mean that they will shorten the time for process completion, and they will become a major factor in our lives, as can already be seen with customer service bots. Machines, computers and now AI have increased automation, but they have not left the human out of the loop. AI may change the way we interact but it will not replace humans. It is another step in the technological evolution, and we will need to learn how to deal with it. Just like we have become highly aware of the health consequences of what we eat, we will also need to learn how to choose wisely which information we consume, how we consume it, and where we consume it.
