The Future of AI in Statistical Analyses of Ritual Practices

With the steady rise of virtual communication, especially in light of the post-COVID-19 pandemic, we can see a constant rise in televised or online broadcasted sermons worldwide. For example, from the period of 2020 to 2024, the third largest megachurch in the USA, Lakewood Church, uploaded 640 recorded sermons and bible studies for its viewers to watch online. With video lengths ranging from 30 minutes to 2 hours, it can be estimated that Lakewood Church produced over 1000 hours of analyzable footage in less than 4 years. As a result, for the first time in history, religious scholars have access to more sermons than even a team of scholars could watch and analyze in four years' time. The question arises- how can we analyze so much material efficiently while utilizing modern technology? In this paper, I argue that Artificial Intelligence (AI) can and should complement traditional methods like ethnography and textual analysis. Implementing AI to analyze large data sets of video/audio material will allow scholars to process large quantities of data efficiently and with precision.

By using Ronald Grimes' schema of the family characteristics of ritual, we can evaluate the degrees of rituality of practices within any given religion (Grimes 2014).  Grimes developed a framework for understanding and recognizing patterns of rituals, such as Prescribed Behavior, Formalized Rituals, Symbolic Communication, Structured Time and Space and others. These characteristics of ritual can then be written as conditions into an AI engine, which would then identify such patterns when processing video and audio material (So, 2020). It is essential to understand that AI is not a nebulous concept but is simply a software program that is based on initial parameters that then uses algorithms to enhance its recognition of such parameters through new data input or a process of trial and error. Luckily, one does not need to start from the beginning when building an AI solution, as a large enough pool of AI tools already exists that can be implemented as plugins into whatever project one is working on. For example, Google Cloud Vision API is a platform that can be trained to "interpret and analyze visual data and derive meaningful information from digital images, videos, and other visual inputs" (Google Could Vision, 2024). A religious scholar would need to use his theoretical knowledge to decide which criteria to train such an AI engine to recognize. Based on Grime's schema, we could identify linguistic, colour schemes, musical tonality, facial expressions, and ratios of different intonations as different parameters to teach the AI to look out for. Such an AI engine can create detailed and precise statistical data with the proper symbiosis of scholarly theory and technological tools. Such an analysis would not be limited in scope due to the time constraint of analyzing each video sermon, but with readily available computing power, it would analyze a multitude of video sermons in a short time. As a potential benefit of such large-scale statistical analyses of online sermons, significant new insights can be made in the field of comparative religions and the study of ritual practices. 

The development of such a specialized AI engine is a very complex and lengthy process, which I will attempt to do with adequate funding in my future PhD work. However, for the purposes of this paper, I will create a prototype AI engine that will recognize a few of the ritual characteristics identified by Grimes from one of the recorded sermons. I will provide statistical data gathered by such AI analyses as an example of what could be achieved in the future with a fully built tool.

Therefore, in this paper, I will detail the potential approaches to classify different characteristics of ritual while also giving a practical report on which AI tools would be needed to be implemented to make this project possible. I will also highlight some of the potential dangers to such statistical analyses, such as AI-generated mistakes known as AI hallucinations (Salvagno, 2023). I will also talk about the limitations associated with studying religions and rituals through a purely video/audio interface. In many cases, much of what constitutes religion still happens behind the camera lens; it is, therefore, important to know that such statistical AI analysis should still be complemented by fieldwork and physical anthropological analyses of the religion. Nonetheless, creating such an AI-powered engine, which would be built with the specific goal of recognizing ritual patterns, can be an outstanding tool for religious studies scholars to use worldwide. It appears that no such initiative is currently being developed within the fields of comparative religion or the study of ritual practices, and therefore, this could be the first attempt to conceptualize a new field of research within ritual studies.