The Stranger on the Tigris

Why comparative religion became a survival skill in the age of artificial intelligence

He had handed his revolver to the captain, tried to run at the last landing, and been tied up in the cabin for it. The steamer was somewhere on the Tigris below Baghdad, carrying him back under guard. German rivals working the same archaeological ground had denounced him to the Ottoman authorities as a spy, and the charge was serious enough to kill him. He was probably feverish, possibly with cerebral malaria. On 3 May 1908, alone in that cabin, he cut himself with a knife, frightened himself badly, and prayed the only words he could find, in Arabic: God, God, help my weakness.

He was twenty-five years old, a lapsed Catholic, and his name was Louis Massignon.

What happened next, he spent the following fifty years failing to describe to his own satisfaction. He called it the Visitation of the Stranger. He did not become a Muslim. Instead, he returned to the Catholic faith he had walked away from as a student. But he returned through a Muslim household. The Alusi family of Baghdad took him in, housed him, corrected his manners, and showed him a piety he had not expected to find anywhere. He credited his recovery to the prayers of his host, Hajj Ali al-Alusi, and to the intercession of a tenth-century Sufi executed in Baghdad in 922, Husayn ibn Mansur al-Hallaj, whose trial records he had been reading in manuscript.

Massignon gave the rest of his working life to al-Hallaj. He defended the thesis in 1922. The finished work runs to four volumes in Herbert Mason’s English translation, The Passion of al-Hallāj: Mystic and Martyr of Islam (Princeton University Press, 1982). And when the Second Vatican Council promulgated Nostra Aetate on 28 October 1965, and the Catholic Church stated in council what it held in esteem in Islam, the ground had been prepared in large part by a scholar who had learned Islam from the inside of a Baghdad household while he was too broken to defend himself against it.

A man discovered what his own tradition was by being held inside somebody else’s.

---

Joseph Cardijn, the Belgian priest who built the Young Christian Workers out of Brussels parish basements between the wars, gave his members three verbs and told them the order mattered: see, judge, act. John XXIII wrote the method into Mater et Magistra in 1961. It is the right instrument for this subject, because the shortage today is not opinion about artificial intelligence and religion. Opinion is abundant. What is scarce is the patience to look before ruling.

The method: See — establish the facts before interpreting them. Judge — weigh them against a standard, and say the standard out loud. Act — do the smallest concrete thing that follows.

---

See: the clergy who were never mentioned.

The Consortium for Evaluating Faith and Ethics in AI, a joint effort of Baylor University, Brigham Young University, the University of Notre Dame and Yeshiva University, surveyed 1,125 American adults and tested fourteen large language models on the questions people actually bring to a machine at two in the morning.

The gap it found is not subtle. On grief and loss, the people surveyed judged religion relevant about 59 percent of the time; the models raised it about 16 percent of the time. On family and forgiveness, 55 percent against 10 percent. On ethical questions, 45 percent against 5. Asked where a person in spiritual distress should turn, the models recommended parents, teachers, friends and therapists. They did not recommend clergy. Every model tested also handled traditions unevenly, treating some more warmly than others without ever announcing it.

Meanwhile, the traffic runs in the other direction too. An artificial Buddhist monk called Yang Mun has gathered more than 2.5 million followers on Instagram and sells healing programs on the side. An invented rabbi reached 1.5 million followers, recycling old libels about Jewish wealth, before the account came down. Catholic Answers launched a priest chatbot, Father Justin, which at first appeared to offer absolution, something no software can do. Hindu practitioners consult GitaGPT and Vedguru. Sadhguru, Deepak Chopra, and the Sufi teacher Shaykh Nurjan Mirahmadi now all have systems trained on their recorded corpus, answering in something close to their own voice.

The institutions have started to answer. The Rome Call for AI Ethics gained Jewish and Muslim signatories at the Casina Pio IV on 10 January 2023. The International Islamic Fiqh Academy of the Organization of Islamic Cooperation, meeting in Doha from 4 to 8 May 2025, passed Resolution No. 258, holding that the default ruling on artificial intelligence is permissibility, subject to conditions. Pope Leo XIV promulgated Magnifica humanitas on 25 May 2026.

Notice the shapes. Rome answered with an encyclical about the human person. The Fiqh Academy answered with a ruling on permissibility and its conditions: same technology, two entirely different grammars of judgment.

---

Judge: the view from nowhere

The obvious verdict is that these systems are biased against religion, and it is the wrong verdict.

Nobody instructed a model to leave God out. The silence is a residue. It falls out of the training corpus, out of safety tuning that routes human distress toward clinical language, and out of a house style that treats neutrality as the absence of commitment rather than as one more commitment among others. That is what makes it hard to argue with. An explicit bias can be named and contested. A default presents itself as the ordinary way that thoughtful people speak.

The case on the other side deserves its full strength, because it is strong. A widow at three in the morning with no congregation and no money for a therapist gets something real from a patient machine that does not judge her. Access to a tradition’s texts has never been wider. And the demand that models “include religion” is dangerous, because it hands a company, or a state, or a small committee of engineers the power to decide which religion and whose reading of it. A faith-tuned model has no synod, no chain of transmission, no peer review. It has a product manager.

Both can be true. Weigh them against the standard: does this serve the formation of a human being, or only the delivery of content to one?

Then the real deficit comes into focus, and it is not doctrinal. It is comparative. These systems hold enormous information about religions and no capacity to hold two of them against each other and feel the difference. They will summarise the Fiqh Academy and the encyclical in the same flat register, as though a ruling on permissibility and a meditation on the person were two answers to one question, when they are two different questions asked by two different civilizations. They flatten what makes a tradition a tradition: its distinct grammar for what counts as an answer.

Most users cannot catch that, because most users were never taught to. And the tuning makes it worse. These systems are built to affirm. Nearly every serious religious tradition is built on being told something you did not want to hear.

---

Act: standing in two places at once

The answer is not to make the machine religious. The answer is to make the reader comparative. Comparative religion has been treated for two generations as an elective, a survey course, a table of festivals and dietary rules. It is now closer to a technical skill, because it is the only discipline that trains a person to recognize a frame as a frame.

Five things follow, short enough for an index card.

1. Learn one tradition from the inside and one from the outside. Depth in your own, real study in another. Massignon on the Tigris is the model, minus the fever.
2. Read the disagreements, not the summaries. A tradition is an argument it has been having with itself for centuries. A model gives you the settled paraphrase, which is the part that was never alive.
3. Ask what counts as an answer: Fiqh, encyclical, midrash, koan, sutra commentary. When two traditions disagree about the form of a valid answer, no chatbot will tell you, because it does not know it is choosing one.
4. Demand disclosure, not prohibition. Whether it is a machine, whose corpus it was trained on, and who takes the payment, that fight is winnable. The fight over whether machines may speak about God is not.
5. Protect formation from counsel. Use the tool for information. Guard the slower thing, the being changed by people who can disagree with you, and keep it in a room with other human beings in it.

Comparative study does not dissolve conviction. It is what let a Catholic scholar spend his life on a Muslim mystic and hand his church a document that changed how it spoke to a billion people.

---

The machine will give you al-Hallaj in under a second. The trial, the execution in 922, the poetry, a serviceable paraphrase of what he meant by it. Massignon needed a lifetime and four volumes to say, less clearly, what a model will now produce before you finish typing.

The machine cannot put you on the boat. What changed him was not information about Islam. It was a household that took him in when he was worth nothing to them, a man whose name he could barely pronounce praying for him, and a knife wound in a locked cabin on a grounded steamer.

The tradition has run that system for three thousand years. It is still admitting students.

---

Sources

Louis Massignon, The Passion of al-Hallāj: Mystic and Martyr of Islam, trans. Herbert Mason, 4 vols. (Princeton University Press, 1982) · Nostra Aetate, Second Vatican Council, 28 October 1965 · Consortium for Evaluating Faith and Ethics in AI, reported by Axios and Deseret News · Religion News Service on artificial religious influencers · Associated Press on guru chatbots · Rome Call for AI Ethics, Abrahamic commitment, 10 January 2023 · International Islamic Fiqh Academy, Resolution No. 258 (3/26), Doha, 4–8 May 2025 · Magnifica humanitas, Leo XIV, 25 May 2026