JERUSALEM (VINnews) — An unusual manuscript entitled Daas Mechona (Machine Knowledge) has been circulating online in recent days, attracting interest in the Torah world. The work, which runs to hundreds of pages, is devoted entirely to the increasingly unavoidable encounter between the world of Jewish law and artificial intelligence.

The cover states that the book deals with “an examination of the issues arising from the development of artificial intelligence in Jewish law and Jewish thought.” It is attributed to “Feivel Finef” of Boro Park. The author’s identity cannot be independently verified from the file itself, although contact information appears in the manuscript and the name is presented as that of the author.

The work does not limit itself to the question of whether it is permissible to ask a chatbot a halachic question. It covers a wide range of subjects, including halachic rulings, rabbinical courts, reliability and establishing facts, Shabbat and holidays, monetary law, damages, copyright, kosher law, STaM (the writing of Torah scrolls, tefillin and mezuzot), Torah study, and even questions of faith and Jewish thought.

The author, however, makes one point particularly clear near the beginning of the book: the work does not constitute practical halachic rulings, and no one should rely on it for practical guidance without consulting a qualified halachic authority.

This caution connects directly to one of the central problems discussed in the book: an AI model’s answer can sound fluent, authoritative and precise even when it is wrong.

The book explains that a language model does not necessarily retrieve an answer directly from a book or from a properly verified Torah database. Rather, it generates the answer anew, piece by piece. As a result, even a citation that appears trustworthy and a source reference that sounds precise may turn out to be fabricated. The work even demonstrates how a system can refer a user to a particular section of the Shulchan Aruch or the Shach that has nothing at all to do with the subject being discussed.

And this is where the practical problem lies. A person might ask, for example, whether a driver who received an expensive package and had it stolen while it was in his possession is obligated to pay for it. The question may sound straightforward, but from a halachic perspective, the answer could depend on the driver’s legal status, the conditions under which he accepted the item, the nature of his responsibility for safeguarding it, the circumstances of the theft, and other details.

An AI system can formulate a lengthy, scholarly-sounding answer within seconds. It may even provide sources. But confident wording, by itself, does not demonstrate that the conclusion is correct.

“Machine Knowledge” therefore distinguishes between relying on a machine as a halachic authority and using it as a tool to assist a halachic authority. The work suggests that such systems can be used to locate sources, organize material, analyze and compare information, but emphasizes that their output should be regarded as raw material requiring examination and verification by a Torah scholar.

In other words, the more convincing artificial intelligence becomes, the more important it is to know where to stop. It can serve as a digital study partner, a sophisticated search engine and a powerful research assistant. But when a question moves beyond learning and exploration to an actual practical halachic ruling, the responsibility still rests with the halachic authority.