In the Oscar winning movie A Separation, Raziyeh, the caregiver of Nader’s father, picks up the phone from the convenience of Nader’s home and calls her marjaʿ taqlīd’s[1] office to ask about a religious question concerning her job.[2] Access to a marjaʿ was not always easy. In older times, the most common way of finding an answer to a religious question was asking the local mosque’s imam,[3] hoping that the imam would also be familiar with your marjaʿ’s opinion on the issue in question, or that the question was common enough for most marjaʿs (pl. marājiʿ) to agree on the answer. But then came telecommunications, reshaping the relationship between a muqallid (follower of a marjaʿ’s opinions) and the marjaʿ.
The offices of prominent marājiʿ have kept up with the emerging technologies to streamline the connection between the muqallid and the marjaʿ. Today, you can write an email from anywhere in the world to ask your religious question and expect a response from the marjaʿ’s office in a short matter of time.[4] Most marājiʿ even have their own websites where they have uploaded answers to commonly asked questions.[5] With such databases now available online, the muqallid is also able to open the marājiʿ website and run a search to find answers to commonly asked questions. More recently, some jurisdictions have even experimented with AI-powered virtual fatwā services, such as Dubai’s Virtual Iftāʾ.[6]
But over the past two year, the rise of general-purpose generative AI and Large Language Models (LLMs) has begun transforming nearly every aspect of life. LLMs “are AI systems that are designed to understand and generate human language.”[7] The most prominent LLM made headlines back in November 2022 when OpenAI enrolled its latest ChatGPT version.[8] The emergence of LLM-based applications like ChatGPT signals a more profound shift in the relationship between the marājiʿ and the muqallid —one that goes beyond merely facilitating communication. These conversational AI models are not just intermediaries; they are capable of generating plausible answers to a muqallid’s religious inquiries without direct consultation with a marjaʿ. Trained on vast amounts of data, including authoritative Islamic texts,[9] these models raise an urgent and complex question: Can LLMs replace the marjaʿ? Or is there an essential attribute of the marjaʿ as a mujtahid al-jamaʿ al-sharāʾiṭ—and of the act of ijtihād[10] itself—that distinguishes it from a machine-generated response?
One might initially argue that LLM-based applications are prone to errors, even in simple tasks such as counting the number of “r”s in “strawberry.” [11] Or raise the issue of “hallucination”—the generation of seemingly plausible but factually incorrect information by these LLMs.[12] However, newer models have significantly improved and now claim to be equipped with advanced “reasoning” capabilities, [13] and continue their advancement to minimize hallucination. We should thus be cautious in forming hasty conclusions. While traditional scholars may resist the idea of an app becoming the go-to resource for religious questions, there is already a consensus coming together on one key point: LLMs will have a significant and unavoidable impact on ijtihād and the role of a marjaʿ as we know it.[14] As illustrated below, this impact arises from the striking similarities between what an LLM does when generating an answer and what a mujtahid does when engaging in ijtihād.
It is essential to distinguish between two prominent roles—among others—that a marjaʿ fulfills in Muslim society. The first role, exemplified by Raziyeh’s consultation in A Separation, is that of providing religious guidance to laypersons by answering their questions on common matters of faith and practice.[15] For routine religious inquiries, LLMs function like research tools, retrieving answers from available sources, much like internet searches or published risālah treatises that muqallids consult. Even prominent marājiʿ have embraced technology, making these answers accessible online.[16] However, a marjaʿ is also entrusted with a more complex task: engaging in ijtihād to address novel questions—those that have not yet been explicitly answered within the existing corpus of Islamic jurisprudence.[17] It is this latter function that raises significant questions regarding the potential role of large language models (LLMs) such as ChatGPT in Islamic jurisprudence.
While the contours of ijtihād are complex and cannot be fully examined in this brief essay, some scholars have categorized the process into three main steps, which can be compared to how an LLM generates an answer.[18] First, a mujtahid engages in a comprehensive study of the available and authoritative sources.[19] Second, the mujtahid analyzes these sources using established principles of legal interpretation to derive a ruling applicable to the given question. Finally, the mujtahid contextualizes the extracted ruling within contemporary circumstances, requiring an understanding of the socio-political, economic, and cultural conditions of the time (which goes back to the question of who is a mujtahid al-jamaʿ al-sharāʾiṭ). All these three steps can be envisioned and implemented by an LLM. Moreover, the mujtahid can also add the discussion of probability (ẓann) in ijtihād, which means that the given answer is what the mujtahid believes to be the best answer, but it is neither guaranteed nor claimed to be %100 accurate. That skepticism regarding accuracy also applies when comparing it to an answer given by an LLM tool, which may not necessarily be the correct answer.
The response to these questions of whether an LLM-based application can eventually replace the role of marājiʿ is far from straightforward. For instance, can the reasoning of an LLM such as ChatGPT be regarded as a valid source of ijtihād in the same way that human reasoning (ʿaql) is? Are there meaningful parallels between following a mujtahid posthumously and relying on a non-human LLM model? Does the fact that different LLMs may produce divergent answers pose a fundamental challenge, or is this merely analogous to the variance in legal opinions among marājiʿ? Even if an LLM replicates each step of ijtihād, does its response to a religious inquiry carry the necessary hujjiyyah—the authoritative force required for adherence? Beyond doctrinal concerns, the issue also demands a sociological perspective: why might a Muslim turn to ChatGPT for religious guidance in the first place? Is it the efficiency, accessibility, or the avoidance of human interaction—particularly when posing intimate or sensitive questions? These considerations merely scratch the surface of a broader, evolving discourse.
Many marājiʿ and research institutes are actively examining these questions.[20] At the Qom Seminary (Ḥawzah-ye ‘Ilmīyah-ye Qom), younger scholars are now offering advanced courses on the intersection of AI and Islam.[21] What is commonly referred to as “dars al-khārij”—the highest level of seminary study, undertaken by those who have completed their foundational coursework—includes discussions on AI and its implications for Islamic scholarship.[22] Additionally, seminars and conferences are being organized to explore the role of AI in Islamic thought.[23] These initiatives primarily aim to introduce the field, encourage scholarly engagement, and facilitate research at the intersection of AI and Islamic studies.
Some marājiʿ themselves have also taken note of this emerging field. For instance, Ayatollah Nouri Hamadani issued a statement on the advancement of AI, emphasizing the need for seminaries to engage with this technology.[24] In his statement, he cites Qur’ān 45:13, which states: “He [Allah] made subservient unto you whatsoever is in the heavens and whatsoever is on the earth–all together. Truly in that are sings for a people who reflect.”[25]
Ayatollah Hamadani also references a ḥadīth attributed to ʿAlī ibn Abī Ṭālib ؏: “Reason is the foundation of knowledge, and reflection is the foundation of reason.”[26] From these sources, Ayatollah Hamadani concludes that utilizing emerging technologies aligns with the human purpose on earth. He asserts that scholars should employ AI to enhance intellectual inquiry and assist in research. He warns that if believers endowed with knowledge fail to adopt these technologies in a timely manner, they will face significant challenges later on, stating that “The future belongs to those who benefit from modern sciences and technologies.”[27]
Ayatollah Makarim Shirazi, another prominent Shīʿī marja‘, has also addressed AI in the context of Islamic law. In a statement prepared for a conference on AI and Islamic jurisprudence, he emphasized that while AI plays a role across various disciplines, its application in Islamic jurisprudence (fiqh) differs significantly from other fields.[28] He underscored that Islamic knowledge has been bestowed upon believers through the Ahl al-Bayt (a reference to the Twelve Imams in Shīʿī Islam). Thus, while modern technologies should be utilized to enhance research into Islamic sources, their use must be approached with scholarly rigor and sensitivity to the distinct nature of religious knowledge.
The Director of the Association of Uṣūl al-Fiqh at Qom Seminary has also spoken on the relationship between ijtihād and AI. In a session on AI and ijtihād, he remarked that: “Although AI is not yet capable of deriving Islamic legal rulings (istinbāṭ al-aḥkām al-shar‘iyya), it may soon achieve that ability.”[29]
He further elaborated:
At first glance one might say that if AI is simply locating the correct sharī‘a ruling, one could argue that following such a ruling constitutes adherence to the sharī‘a. Nevertheless, whether AI meets the established criteria for mujtahid status remains an open question requiring further scholarly examination. However, given that AI’s error rate is relatively low and that it possesses the foundational analytical capabilities of ijtihād, it could serve as a valuable tool for investigating primary sources—similar to how fuqahāʾ rely on ‘ilm al-rijāl (the science of ḥadīth narrators) in their jurisprudential inquiries. In fact, the use of AI could significantly accelerate the process of identifying aḥkām shar‘iyya (Islamic legal rulings).[30]
However, as with most emerging issues, no consensus has yet been reached, and it is now the role of scholars to engage with these important questions on the feasibility of relying on LLMs for religious inquiries and ijtihād.[31]
It is also important to recognize that the role of the marājiʿ in Shīʿī societies has extended far beyond merely issuing legal opinions on novel questions. At times, their leadership has been among their most defining and historically memorable roles. A well-known example is Ayatollah Mirza Shirazi’s famous 1891 fatwā calling for the boycott of tobacco in all forms, [32] which catalyzed significant societal transformation. If LLMs existed at that time and such question regarding the permissibility of tobacco were posed to a model such as ChatGPT, it is unlikely that it would have ever answered in the negative. The issuance of such a fatwā required an understanding of social dynamics, historical context, and communal sensitivities—factors that exceed the capacities of current AI models. These systems are limited by their reliance on existing datasets, the need for periodic memory updates, and, at best, an ability to retrieve publicly available information to construct a partial understanding and awareness of recent developments. More importantly it is unlikely that a muqallid would have ever cared to follow such ruling generated by a computational system, absent the personal attributes and the cultivated trust of the muqallid in their marājiʿ that influences the muqallid’s decision in following the marājiʿ in the first place.
Therefore, while an LLM-based application could potentially one day engage in ijtihād in the sense discussed in this piece, it is highly doubtful it can altogether replace the marājiʿ, at least no more likely than our current understanding of the possibility of AI replacing human beings themselves.
Notes:
[1] “Many Twelver Shīʿite Muslims choose a religious clerk as their reference point to turn to for answers to their questions about their religious obligations. This person, who must satisfy many qualifications, is called a marjaʿ taqlid.” Zahra Takhshid, “Commentary:: Organic Labeling: Reconciling Religious Freedom and Animal Welfare in the European Union,” Islamic Law Blog, November 13, 2019, https://islamiclaw.blog/2019/11/13/commentary-organic-labeling-reconciling-religious-freedom-and-animal-welfare-in-the-european-union/. The concept of marjaʿ is unique to Shīʿī Islam. In the Sunnī tradition, mufti is the authority closest to what the Shīʿī tradition recognizes as a marjaʿ, as they too issue authoritative religious guidance based on Islamic jurisprudence.
[2] A Separation, directed by Asghar Farhadi (2011; Los Angeles: Sony Pictures Classics, 2012), Amazon Prime Video, https://www.primevideo.com/detail/A-Separation/0I5L76LZEJH44VKES459W5FEJU.
[3] Imam here is used to refer to the person who leads the prayers at the mosque and has a basic knowledge of Islamic literature.
[4] Some marājiʿ have online platforms for submitting questions, making it easier to follow up and receive answer instead of e-mailing. See, for example, Pasokhgoo-ye Online, “Official Information Portal of Ayatollah Makarem Shirazi,” last accessed February 15, 2025, https://online.makarem.ir.
[5] See, for example, “Official Website of Ayatollah Sistani,” Sistani.org, last accessed February 7, 2025, https://www.sistani.org/persian/qa/search/14015/; see also “Official Website of Ayatollah Makarem Shirazi,” Makarem.ir, last accessed February 7, 2025, https://old.makarem.ir/main.aspx?reader=1&lid=0&mid=907&catid=0&pid=61769.
[6] “Virtual mufti: Dubai launches ‘first ever’ artificial intelligence-powered fatwa service,” The New Arab, October 31, 2019, https://www.newarab.com/news/dubai-launches-first-ever-artificial-intelligence-powered-fatwa-service. The author was not able to access the services, and it appears they are no longer available.
[7] Harry Surden, “ChatGPT, Large Language Models, and Law,” Fordham Law Review 92, no. 5 (2024): 1942 (citing Muhammad Usman Hadi et al., “Large Language Models: A Comprehensive Survey of Its Applications, Challenges, Limitations, and Future Prospects,” Semantic Scholar, December 7, 2023, https://www.semanticscholar.org/paper/Large-Language-Models%3A-A-Comprehensive-Survey-of-Hadi-tashi/24de1048791bac4972ecc16d1c3c1de23691407d).
[8] “Introducing ChatGPT,” OpenAI, November 30, 2022, https://openai.com/index/chatgpt/.
[9] Many online Islamic book databases exist such as https://lib.eshia.ir/.
[10] The verb ijtihād in its literal sense is from jahd (meaning power and capacity), translates to trying your best in doing something. In Shīʿī Islam, ijtihād has a specific meaning: “extraction of a subsidiary Islamic legal ruling or the proof for it from its detailed evidence or adillah (plural of dalīl).” Ibn Manẓūr, Lisān al-ʿArab vol. 3 (Dar Sader, Beirut, 1993–1994), 133, available at https://shamela.ws/book/1687/1575; Naser Makarem Shirazi, Anwār al-Uṣūl (Ṭ.: Madrasat al-Imām Amīr al-Muʾminīn, n.d.), 599. There are different kinds of ijtihad the discussion of which is outside of the scope of this piece.
[11] “Incorrect count of ‘r’ characters in the word ‘strawberry,’’’ OpenAI Developer Community, June 2024, https://community.openai.com/t/incorrect-count-of-r-characters-in-the-word-strawberry/829618.
[12] Surden, “ChatGPT, Large Language Models, and Law,” 1949 (citation omitted).
[13] See, for example, “Learning to reason with LLMs,” OpenAI, September 12, 2024, https://openai.com/index/learning-to-reason-with-llms/.
[14] See for example, “Ayatollah Makarem’s Recommendation on Utilizing Artificial Intelligence,” Tabnak, October 26, 2020, https://www.tabnak.ir/fa/news/1011149/; “The 16th Assembly of the Society of Seminary Teachers to Be Held on the Topic of Artificial Intelligence,” IRNA, December 2, 2024, https://www.irna.ir/news/85677535/; “A Different Perspective from a Marja’ Regarding Artificial Intelligence,” Khabar Online, December 2024, , https://www.khabaronline.ir/news/1995993/.
[15] See footnote 2.
[16] See, for example, footnotes 4 and 5.
[17] For example, questions on what is the obligation for the five time prayer of an astronaut who is in space or prayer on other planets and celestial bodies. This type of ijtihād is referred to as ijtihād al khās. See, for example, “Official Website of Ayatollah Sistani,” Sistani.org, last accessed February 15, 2025, https://www.makaremshirazi.net/main.aspx?typeinfo=21&lid=0&catid=44509&mid=274067; “Official Website of Imam Khomeini” last accessed February 15, 2025, http://www.imam-khomeini.ir/fa/c78_114763/ترجمه_تحریر_الوسیله_امام_خمینی_س_/ج_2/خاتمه_احکام_شرعی_در_کرات_دیگر._
[18] Rūḥullāh Shamsī Kūshkī, Niẓām-i Masāʾil va Kārburdhā-yi Fanāvarī-yi Hush-i Maṣnūʿī dar Ijtihād, May 12, 2020, available at https://alefbalib.com/index.aspx?pid=256&PdfID=711110.
[19] A mujtahid cannot rely on any source for ijtihād. There is a body of literature on what sources count as reliable and permissible for ijtihād. In Shīʿī fiqh, the credible sources thus far have been the Qurʾān, Sunna, consensus (ijmāʿ), and reason (ʿaql), commonly referred to as the adillat al-arba‘a (the four sources). See, for example, Ayatollah Makarem Shirazi, “Manabe‘-e Ijtihad dar Fiqh-e Shi‘a va Ahl-e Sunnat,” available at https://makaremshirazi.net/main.aspx?lid=0&typeinfo=1&catid=25445&pageindex=1&mid=419358; see also Morteza Motahhari, Āshnā’ī bā ‘Ulūm-e Islāmī, vol. 1 (Tehran: Sadra Publications, 1998), 20, available at https://lib.eshia.ir/50052/1/19/چهارتا .
[20] See, for example, Hasan Latifi, “Challenges of Using Artificial Intelligence in the Process of Shi’i Ijtihad,” Religions 15, no. 5 (2024): 541; Etemad Online, Raʾīs Anjuman Uṣūl al-Fiqh Ḥawzah ʿIlmiyyah Qom: Hūsh Maṣnūʿī dar Āyandah Qudrat Istinbāṭ Aḥkām Sharʿī rā Pidā Khāhad Kard, November 12, 2024, https://www.etemadonline.com/بخش-سیاسی-9/684392-هوش-مصنوعی-احکام-شرعی-حوزه-علمیه-فضای-مجازی; Rūḥullāh Shamsī Kūshkī, Niẓām-i Masāʾil va Kārburdhā-yi Fanāvarī-yi Hush-i Maṣnūʿī dar Ijtihād, May 2020, 12, available at https://alefbalib.com/index.aspx?pid=256&PdfID=711110.
[21] “Artificial Intelligence Has Found its Way into Advanced Fiqh Studies,” Khabar Online, September 16, 2024, https://www.khabaronline.ir/news/1957827/ (covering the introduction of a course taught by Professor Ḥusayn Mufīdī).
[22] See, for example, Seyed Sadegh Mohammadi, Archive of the Year 1403–1402: Dars Kharij Fiqh – Medical Fiqh – Artificial Intelligence, last accessed February 15, 2025, https://www.eshia.ir/Feqh/Archive/mohammadi/feqh/1402/; see also Abolqasem Alidoust, “Requirements for a Judge (fiqh al-qāda), Dars-e Khārej Fiqh, lecture, 9 Azar 1398 (December 1, 2019),” https://www.eshia.ir/Feqh/Archive/text/alidoust_abolghasem/feqh/98/980911/هوش%20مصنوعی/
[23] See, for example, “The 16th Assembly of the Society of Seminary Teachers to Be Held on the Topic of Artificial Intelligence,” IRNA, December 2, 2024, last accessed February 15, 2025, https://www.irna.ir/news/85677535/.
[24] “A Different Perspective from a Marja’ Regarding Artificial Intelligence,” Khabar Online, December 2024, last accessed February 15, 2025, https://www.khabaronline.ir/news/1995993/.
[25] Translation from Seyyed Hossein Nasr, ed., The Study Quran (San Francisco: HarperOne, 2015).
[26] «الْعَقْلُ رَأْسُ الْعِلْمِ، وَالْفِکْرُ رَأْسُ الْعَقْلِ»
[27] “A Different Perspective.”
[28] “Ayatollah Makarem’s Recommendation on Utilizing Artificial Intelligence,” Tabnak, October 26, 2020, last accessed February 15, 2025, https://www.tabnak.ir/fa/news/1011149/.
[29] Raʾīs Anjuman Uṣūl al-Fiqh Ḥawzah ʿIlmiyyah Qom: Hūsh Maṣnūʿī dar Āyandah Qudrat Istinbāṭ Aḥkām Sharʿī rā Pidā Khāhad Kard, Etemad Online, November 12, 2024, accessed February 12, 2025, https://www.etemadonline.com/بخش-سیاسی-9/684392-هوش-مصنوعی-احکام-شرعی-حوزه-علمیه-فضای-مجازی.
[30] Ibid.
[31] Iran has recently passed a charter on AI called “The National Artificial Intelligence Charter of the Islamic Republic of Iran.” The Charter has since been criticized for not being realistic in its outlined goals. See “The National Artificial Intelligence Document Is Neither Realistic Nor Practical,” ANA News Agency, August 27, 2024, https://ana.ir/fa/news/928557/. For the full text of the Charter, see “The National Artificial Intelligence Charter of the Islamic Republic of Iran, approved by the Supreme Council of the Cultural Revolution, Education, and Knowledge,” June 18, 2024, https://nezamat.ir/سند-ملی-هوش-مصنوعی-جمهوری-اسلامی-ایران/.
[32] Ayatollah Mirza Shirazi once issued the famous boycott fatwā on the use of tobacco in response to a series of concessions to a British company signed by the ruling Qajar dynasty that gave a British citizen “the monopoly of buying, selling, and manufacturing all the tootoon and tobacco in the interior or exterior of the Kingdom of Iran for 50 years. . . .” Mansoor Moaddel, “Shi’i Political Discourse and Class Mobilization in the Tobacco Movement of 1890–1892,” Sociological Forum 7, no. 3 (September 1992): 459.
(Suggested Bluebook citation: Zahra Takhshid, ChatGPT and the Marjaʿ, Islamic Law Blog (Apr. 10, 2025), https://islamiclaw.blog/2025/04/10/roundtable-chatgpt-and-the-marja%ca%bf/)
(Suggested Chicago citation: Zahra Takhshid, “ChatGPT and the Marjaʿ,” Islamic Law Blog, April 10, 2025, https://islamiclaw.blog/2025/04/10/roundtable-chatgpt-and-the-marja%ca%bf/)