By Intisar Rabb & Mairaj Syed
The invention of the wheel, the printing press, and the internet itself swept over the world in torrents that shaped human interaction with each other and the world, Islam included along with everything else. When initially introduced, each technology crashed in on the world like a rolling wave that forced the world closer together. More communication and interaction between humans inspired new complex systems and theories: of law, of ethics, of God. Compared to the seemingly sudden eruption of Artificial Intelligence, all three technologies now seem like calm tidal pools. Just like past innovations, AI portends the reconfiguring of old systems and rethinking of old theories. We aim to explore these potential transformations as they pertain to Islam and Islamic law, as both encounter AI. In so doing, we join the many scholars, lawyers, religious leaders, governments, corporations, think tanks, engineers and others actively addressing the prospects and challenges of AI—in this case, for Islam in the 21st century.
We convened six scholars who are thinking and writing on issues of AI and Islamic law to address the most pressing questions we see at the start of the AI revolution, and to contemplate whether and to what extent AI—as it currently stands—will affect the study and practice of Islamic law. That is, given the current state of the technology, and the promises it heralds alongside the ills it threatens, what are the trends and foreseeable trajectories in AI usage among those working with Islamic sources?
So far, the answer is: not that much. Notably, the largest platforms leading the AI wave were not built by or for engineers with deep knowledge of Islam or Arabic. The AI chat and search platforms most familiar to us at home, around university campuses, and in government or corporate conference rooms include a fairly short list: OpenAI’s ChatGPT, Anthropic’s Claude, Microsoft’s Copilot, Google’s Gemini, xAI/Elon Musk’s Grok, Srinivas’s Perplexity, and now China’s DeepSeek. Absent precisely structured prompts, when users ask these AIs questions related to Islamic and comparative law, as we did, they get answers that are either diminishingly underinclusive or imaginatively overinclusive.[1] The answers are underinclusive when they ignore the Islamic world entirely—for example, answering questions about comparative law with reference to US, UK, European, Chinese, and Latin American law: almost every system of law except Islamic law, despite its reach spanning nearly 20% of the world’s population (almost 1.6 billion)[2] and over 1400 years (with 622 as the year of Islam’s advent).[3] The AI platforms are overinclusive in that, when prompted specifically to answer with respect to Islam or Islamic law, they often respond with generalities, hallucinations (generating false information), and inaccuracies that conflict with the specialized needs of a researcher or even the more prosaic questions of an ordinary user. The researcher—for instance: a historian, religious scholar, or a legal professional working at a court or government agency—needs credible, accurate results. Especially if they are to answer questions about law, ethics, or God (and religion more broadly). The ordinary user may well be a Muslim seeking knowledge about their religion. They, too, need credible, accurate results. Especially if they are to ask questions of law, ethics, or God. (Notably, one need not be Muslim or a “believer” to see the necessity for credibility and accuracy. Any historian, lawyer, or other scholar with interests in the Islamic world, global history, or robust theories of law needs access to credible and accurate sources about sharīʿa—perhaps even more so than an “insider” who may be unreflectively practicing one version of Islam—to gain insight into the social, political, economic and other explanations or motivations for a particularly salient religion on major issues of law and society in both the past and the present.)
The stakes could not be higher. Yet the technology as it currently stands falls short. If we think of the Book as the archetypal repository of old-world knowledge and ideas, we find a great gap between the knowledge available in the Book versus the knowledge available online for AI platforms when it comes to Islam and sharīʿa. This is not an especially surprising conclusion given the moral and political visions and priors of the venture capitalists and engineers who have ushered in this new age. Nor is it surprising given the concerns of the mainstream thought-leaders and institutions that developed, digitized, and otherwise determined which ideas to make available online and thus ripe for AI-harvesting. Classical Islamic texts necessary to inform research on Islamic law do not seem to be a corporate priority as a money-making enterprise. Understandably.
So, what to make of these two forms of knowledge, “the Book and AI”? We do not think that the Book and AI is a situation of the Camel and the Wheel—the latter referring to a curious phenomenon wherein, for hundreds of years, Mesopotamian societies rejected the newer and faster technology of the wheel in favor of the camel. In his book by that title famously studying that phenomenon, Columbia University professor-emeritus and Islamic historian Richard Bulliet described how the wheel had been developed between 3000 and 2500 BC and offered a superior technology of transportation than anything before it.[4] But sometime between 500 and 100 BC, inhabitants of the Middle East and North Africa developed saddles that made the camel a better mode of transportation for their needs, adapted for local terrains.[5] We might say that the camel was “context aware.” As is the Book. By contrast, AI technology is not context aware for studies of Islam, Islamic law, or Arabic texts.
Does this mean that those interested in studying Islam will use the Book over AI as inhabitants of the Middle East once adopted the camel over the wheel? Not necessarily. We do not expect scholars to wholesale or always reject AI in favor of traditional books as those ancient societies once chose camels over wheels. But to avoid that fate will require adapting AI to unlock the Islamic Book—that is, at a minimum, it will require locating and digitizing classical Islamic and Arabic texts, converting them to formats that large language models (LLMs) can readily consume online, and organizing and fine-tuning those models to make the AI consumption and output context-aware and useful. In short, if AI is to offer the world a truly universal new knowledge base that is meaningful and useful and representative right at its foundations, AI is in urgent need of adaptation to Islamic texts and contexts.
Roundtable Essays: Overview
The specific challenges and prospects for bridging the divide between the Book and AI take shape along various axes, as the scholars taking part in this Roundtable on Islamic Law and AI reflect. These scholars show how the relationship between Islamic law and AI is emerging as a critical area of study, as they explore both the practical applications and theoretical implications of this new technology. This collection of essays offers diverse perspectives on how AI might transform—or fail to transform—Islamic legal studies and practice. (NB: Below is an adapted version of summaries of the essays to come in this Roundtable, first compiled by Google’s NotebookLM, and substantially modified for content and clarity by the authors.)
One of us—Mairaj Syed, professor of religious studies at the University of California at Davis—provides a foundational understanding of LLMs, which are the basis for AI platforms. He shows how LLMs are fundamentally different from the traditional relational databases that store information. LLMs consume far more books-as-data and are programmed with the ability to learn patterns rather than simply store and retrieve texts. This capability, he argues, could revolutionize Islamic legal research by automating routine tasks and enabling sophisticated historical analysis. Recognizing the complexity of this endeavor, Syed advocates for collaboration between Islamic studies/Islamic law scholars and AI researchers to develop effective tools.
The other of us—Intisar Rabb, professor of law and history at Harvard—explores the relationship between Islamic sources and AI to ask why the AI platforms underperform when conducting research on Islamic law. Seeking to evaluate and explain the poor performance, she tests seven AI platforms then reviews the key components at play: inputs (lack of accessible Islamic sources), processes (experimental “thinking” modes), and outputs (inaccurate research results). Testing 7 LLMs, she argues that current AI models struggle with Islamic law research due to inadequate Islamic source material in their training data (inputs). Thus, in addition to describing the outputs and processes, she charts the landscape of available Islamic sources. In the end, she argues that addressing the lack of adequate inputs for Islamic research is a critical first step toward improving AI’s accuracy and reliability, and that it is a prerequisite to answering other questions about whether and how researchers or general users can or should ethically use AI.
Building on these general attributes about the structure of LLMs and landscape of Islamic sources technical considerations, Ezieddin Elmahjub, professor of law at Qatar University, examines the practical challenges of creating AI systems for Islamic law. His analysis reveals that the primary obstacle lies not in the AI technology itself, but in the preparation of suitable training data. From manuscript digitization errors to dialectical variations in Arabic, Elmahjub identifies numerous technical hurdles to adapting the Islamic Book to formats necessary for AI and LLM consumption and processing. A surprising but pervasive problem is the lack of software that can accurately convert classical Arabic books to digital text—the technology called optical character recognition (OCR) that makes our scanned pdfs human-readable, highlightable, and otherwise manipulable on the front end, and machine-readable on the backend. More fundamentally, he raises concerns about AI’s inability to grasp contextual nuances and higher-order principles, such as the five purposes said to animate Islamic law (maqāṣid) and that are essential to Islamic legal reasoning. However, Elmahjub also highlights promising developments. Recent advances in the current technology will help reduce Arabic OCR error rates and create cleaner, more structured datasets. And new platforms are at least focused on the needs for large repositories for digitized Islamic books (such as OpenITI) and of fine-tuning LLMs and building context-aware AI tools for Islam-related research (such as usul.ai), even if they have not fully cracked the code.
Moving from theoretical frameworks to practical evaluation, M. Zubair Abbasi, a scholar of Islamic and comparative law, points to a crucial direction for developing effective AI tools for Islamic law: the creation of rigorous benchmarks for evaluation. By developing a series of increasingly complex questions (and answers) about Islamic inheritance law, with focus on Sunnī law, Abbasi creates an AI-testing framework that only scholars with deep domain expertise could construct. His systematic approach demonstrates why Islamic legal scholars must play a central role, alongside developers, in adapting AI tools for the field: their detailed knowledge of Islamic legal principles, methodologies, and edge cases is essential for meaningful AI fine-tuning, prompt engineering, and evaluation. His findings reveal that, while current AI systems can structure responses effectively, they often fail to draw on the right sources or correctly weight their relative importance, and they frequently misalign with established Islamic legal methods of interpretation.
In an essay on practical applications, Zahra Takhshid, professor of law at Denver University Sturm College of Law, focuses specifically on Shīʿī law. Her analysis draws intriguing parallels between AI reasoning and traditional Shīʿī legal interpretation. She offers preliminary comparisons of differences in approaches to Islamic law questions by AI systems as opposed to mujtahids, trained experts qualified to interpret law and render legal opinions—either in an advisory capacity (fatwās) or as legally binding in Iran. In her outline of the types of problems that might arise, she raises provocative questions about the nature of religious authority and the legitimacy of AI-generated fatwās. Her work points to deeper questions that remain to be explored about the relationship between human and artificial reasoning, as well as bases for the type of authority that bears on daily practice and major decisions in religious law.
Finally, shifting to the needs of historians and book curators, Sohaib Baig, librarian for Middle East, South Asian and Islamic Studies at UCLA, demonstrates the potential of quantitative analysis of metadata on Islamic manuscript collections to inform the history of the Islamic Book and, with it, Islamic legal history. Baig profiles a newly available, massive collection of over 2 million Islamic manuscripts (1.5 million of which are in Arabic). This impressive collection sponsored by the Muʾassasat ʿIlm li-Iḥyāʾ al-Turāth wa-l-Khidmāt al-Raqmiyya (The ‘Ilm Foundation for the Revival of Heritage and Digital Services) in Cairo, Egypt is drawn from 3000 libraries across 65 countries and is the result of the efforts of some 400 contributors over the last 20 years. Baig, focusing specifically on Islamic legal studies, highlights the rich information and insights that the metadata—which the database curators collected, digitized, and thus made potentially consumable by AIs—can provide into the history of the Islamic Book. For example, his examination of the geographical distribution of legal texts by school of law shows a divergence between the expected regional spread of the more populous sphere of Ḥanafī school practice and the provenance of the collections in domains of Ḥanafī imperial dominance. He aptly raises questions that this data prompts as ripe for further AI-guided human research: how might scholars explain the metadata that these millions of unpublished book manuscripts represent? What legal, social, and material factors explain the numbers? What historical factors explain the lack of some numbers—including, especially, the mass-destruction of books (as occurred in places such as Central Asia and South Asia and more recently in places such as Iraq and Palestine). The metadata—and even more so, the data, once made available online and to AIs—may challenge conventional assumptions about developments in the field and can feed many appetites for future inquiry by humans and AI.
AI-Powered Summary, Analysis & Podcast
Not to be Luddites, we deploy an AI tool that can meaningfully provide value to the world of Islamic Studies and Islamic Law by summarizing a small set of texts. We fed Google’s NotebookLM AI, the very essays in this Roundtable, and asked it to summarize them and generate a table comparing similarities and differences between them on the themes they covered. We find that AI does fairly well at extracting major themes from individual papers uploaded to its platforms whether in English, or remarkably, Arabic too–with only a few caveats given errors in the details. Below is the outline of themes that NotebookLM determined were relevant from the essays about the prospects and challenges of AI, followed by the table it generated. And, here is the autogenerated podcast (gmail login may be required) that provides a friendly way of listening to the highlights from this roundtable (if listeners are not too miffed by the mispronunciations of Arabic names and terms).
Here is a list of the most salient themes related to Islam and AI that arise from this collection of essays:
- AI’s potential to transform Islamic legal research. AI can expedite legal interpretation, facilitate access to texts, enable comparative analyses, and identify patterns across vast corpora. LLMs can recognize deeper semantic patterns and automate routine tasks. AI can assist in answering frequently asked questions, freeing up experts for complex inquiries.
- Challenges in using AI with Islamic data sources. There are disparities in Arabic versus English data availability, incomplete digitized collections, and complexities in unstructured data processing. AI models may struggle with contextual nuances and higher-order principles. Issues include manuscript digitization errors, dialectical variations in Arabic, and a lack of software for accurate conversion of classical Arabic books to digital. Digitization does not guarantee machine readability, and OCR conversion of Arabic texts has low accuracy.
- Ethical and methodological considerations. AI systems must adhere to methodological rigor and epistemic integrity, aligning with established Islamic law principles. There are risks of AI hallucination and algorithmic bias. AI’s outputs must align with established principles of Islamic law, including tafsīr (exegesis) and uṣūl al-fiqh.
- The importance of collaboration. Collaboration between Islamic law scholars, data scientists, and AI specialists is crucial for building robust, context-aware AI tools. Interdisciplinary collaboration can foster a research community that openly shares data and expertise.
- The role of metadata analysis. Analyzing the metadata of Islamic manuscript collections can help to better understand Islamic legal history (among other fields). Examining the geographical distribution of legal texts by school of law can reveal insights into questions of historical practice and political economy.
- Applications of Generative AI (GenAI) to Islamic law. GenAI models can generate structured and comprehensive information, improving search speed and potentially allowing far-ranging reading of relevant sources. However, all references must be verified for accuracy and authenticity, and comprehensiveness is not assured. GenAI can assist in understanding Islamic inheritance law, but users should exercise caution due to unreliable or otherwise incorrect information.
- Limitations of current AI models. AI platforms often underperform when conducting research on Islamic law due to inadequate Islamic source material in their training data. AI agents sometimes provide inaccurate, general, and nonrepresentative details, and may hallucinate sources and statements. The quality and quantity of available Islamic sources are lacking, which affects the output and processes of AI platforms.
Here is a table summarizing the convergences and divergences among the sources, based on the themes identified in the previous response:
| Theme | Convergence | Divergence |
| AI’s Potential | Agreement that AI tools can enhance research speed, access to information, and comparative analysis in Islamic jurisprudence. Recognition of AI’s ability to streamline complex legal questions. | Differing opinions on the extent to which AI can provide new scholarly insights. Varying levels of optimism regarding AI’s capacity to replace or augment traditional scholarly methods. |
| Challenges in Data Structuring | Acknowledgment of the difficulties posed by the limited availability of Arabic data, the heterogeneity of Islamic legal sources, and the complexities of processing unstructured data. Consensus on the need for improved OCR and speech-to-text technologies for Arabic. | Different perspectives on which challenges are most pressing (e.g., language disparities, incomplete collections, unstructured data). Varying proposals for how to address these challenges, ranging from better data curation to the development of new AI models. |
| Ethical & Methodological Concerns | Recognition of the importance of upholding methodological rigor and epistemic integrity in AI applications. Concern about the risks of AI hallucination and algorithmic bias. Emphasis on the need for AI systems to align with established principles of Islamic law and ethics. | Differing views on how to balance the pursuit of efficiency with the need for deep contextual analysis. Varying degrees of emphasis on the potential for AI to distort or undermine traditional Islamic scholarship. |
| Importance of Collaboration | Agreement on the necessity of interdisciplinary collaboration between Islamic scholars, data scientists, and NLP specialists. Recognition that effective AI tools require both technical expertise and a deep understanding of Islamic legal traditions. | Different ideas about how to best foster such collaboration, ranging from collaborative workshops to shared research agendas. Varying levels of emphasis on the need to bridge the gap between traditional academic models and new forms of collaborative infrastructure. |
| Metadata Analysis | Recognition of the value of analyzing metadata from Islamic manuscript collections. Acknowledgement that library catalogs may not provide a direct, unfiltered window into historical manuscript culture. | Differing opinions on the extent to which metadata analysis can provide useful insights for legal scholars. Varying levels of optimism regarding research possibilities moving forward. |
| Application of GenAI | Acknowledgement that GenAI models can generate structured and comprehensive information. Recognition of the need to verify all references for accuracy and authenticity. | Differing opinions on the utility of GenAI models in generating information regarding issues on which there are differences amongst various schools of Sunnī and Shīʿī laws. Varying views on using GenAI models to interpret statutes across diverse jurisdictions, as these tasks require human expertise. |
Notes:
[1] For details, see Intisar Rabb, “The Book and AI [Part 2]: Testing AI Research Agents for Islamic Law,” Islamic Law Blog (forthcoming), an essay in this Roundtable series.
[2] Pew Research Center, “Mapping the Global Muslim Population: A Report on the Size and Distribution of the World’s Muslim Population,” The World Bank, October 2009, https://allafrica.com/download/resource/main/main/idatcs/00011909:cbf45d797f6515d212cec2ec5ef6fb5f.pdf. For the United States, with its Muslim population at 1%, see Pew Research Center, “Religious Landscape Study,” February 25, 2025, https://www.pewresearch.org/religious-landscape-study/.
[3] The Editors of Encyclopedia Britannica, “What are the traditional events of Muhammad’s life?,” Britannica, last updated February 12, 2025, https://www.britannica.com/question/What-family-did-Muhammad-have.
[4] Richard W. Bulliet, The Camel and the Wheel (Cambridge, MA: Harvard University Press, 1975).
[5] Ibid., 87.
(Suggested Bluebook citation: Intisar Rabb & Mairaj Syed, The Book and AI: How Artificial Intelligence is and is not Changing Islamic Law, Islamic Law Blog (Mar. 11, 2025), https://islamiclaw.blog/2025/03/11/roundtable-the-book-and-ai-how-artificial-intelligence-is-and-is-not-changing-islamic-law/)
(Suggested Chicago citation: Intisar Rabb and Mairaj Syed, “The Book and AI: How Artificial Intelligence is and is not Changing Islamic Law,” Islamic Law Blog, March 11, 2025, https://islamiclaw.blog/2025/03/11/roundtable-the-book-and-ai-how-artificial-intelligence-is-and-is-not-changing-islamic-law/)