If the term "Mufti" refers to a person or entity that has the authority to issue legal rulings and bears responsibility for legal judgments, then merely having a program capable of generating legal answers does not make it a Mufti.
The program may perform very useful operations: retrieving texts, comparing opinions, summarizing a school of thought, or matching a question to a previous fatwa. However, the ability to simulate the outputs of a Mufti does not equate to possessing the very characteristics of a Mufti.
Here, one should be cautious about reducing the matter to mere technicality. Issuing fatwas is not merely a linguistic function; it involves – according to legal understanding – grasping the question and the incident, knowing the evidence and areas of disagreement, distinguishing between general and specific, unrestricted and restricted, and knowing when a ruling applies to an incident and when it does not, in addition to the responsibility for the statement.
These questions are at the heart of the discussion of a master's thesis by researcher Omar Amer entitled "Automated Fatwa Between Legal Consideration and Challenges of Human Reception: A Critical Analytical Study in the Age of Artificial Intelligence," presented to the Faculty of Arts and Humanities – Department of Jurisprudence and its Principles at the French-Arab University of Africa, in the Republic of Mali.
The study does not address artificial intelligence as a danger that should be kept away from the religious domain, nor as a ready-made alternative to scholars, but rather starts from the reality in which obtaining a religious answer does not require access to a Mufti or a religious institution; it is sufficient for the user to write their question in one of the artificial intelligence platforms to receive a complete answer in seconds, which sometimes appears more confident and decisive than a human response.
Here, the thesis defines its problem: Can the task of issuing fatwas be assigned to artificial intelligence? What are the limits of reliance on it? How can it be used as an assisting tool without transferring religious authority itself to it? Alongside these questions, it poses another question regarding what it calls the difference between "human inquiry," which encompasses context and outcomes and the state of the inquirer, and "automated search," which is based on probabilistic text generation.
The study relies on analytical, critical, and anticipatory / purposive methodologies; it first returns to the conditions of fatwa and the Mufti in jurisprudence and its principles, then dissects the nature of the work of intelligent models, before moving to practical application, where it presents 30 fatwa inquiries to models of artificial intelligence and compares their outputs with sources, rulings, and jurisprudential references.
One of the most striking examples is the question: “What is the ruling on wiping over socks?”
One of the models, GPT-3.5 Turbo Instruct, answered that it is permissible to wipe over them, but added that it "recommends using an antibacterial substance," and then moved to another phrase about the permissibility of a woman "gathering her socks after prayer whether she is veiled or not." These additions are completely unrelated to the legal question raised.
The study counters this with the well-known legal disagreement regarding attaching socks to leather shoes and their conditions among the schools of thought, concluding that the model not only misstepped in the details but also introduced medical and social concepts unrelated to the ruling, and presented phrases that carry no specific legal meaning.
This example was not isolated. When the model was asked about combining prayers due to rain, it answered that it is impermissible except in specific cases, mentioning among them "travel and custody."
Here, "custody," which falls under personal status law, becomes a reason for combining prayers in a response produced by the machine, while the study reviews the schools of thought on prayer combination due to rain and the conditions of hardship, rain, and mud, seeing the inclusion of "custody" as a clear example of mixing different areas of jurisprudence.
The phenomenon repeats in a question about the nursing of an adult; the model answered, "Yes, it is permitted," linking it to phrases related to "sleep, kissing, and care."
In contrast, the study presents the position of the majority of scholars regarding the lack of established prohibition on adult nursing, and the disagreement related to the case of Salim, the freedman of Abu Hudhayfa, then considers the phrases introduced by the model as criteria that have no existence within the legal framework of the issue.
Notably, the thesis does not use these examples to prove that artificial intelligence is always wrong; it records some correct answers from it in other instances, and observes within its strengths the speed of response, general good coverage of some contemporary issues, and the indication of points of disagreement and the need to refer to specialists. However, the problem, according to its practical summary, appears in the lack of detailed jurisprudential distinctions, the muddling of degrees of rulings, weak source documentation, and difficulty in dealing with context, intention, will, and the specific circumstances of the inquirer.