Is AI Taking Over the “Religion of the Book”? 

Israeli historian Yuval Noah Harari said recently at the World Economic Forum that artificial intelligence will eventually take over all religious texts. In his own words: “If law is made of words, then AI will take over the legal system. If books are just combinations of words, then AI will take over books. If religion is built from words, then AI will take over religion.” 

According to Harari, “in Judaism, authority is granted to words in books, and since no human can remember all the words, but an AI can, the machine becomes the greatest expert.” 

This is a weak argument. The greatest expert does not have to, and in fact never had to, remember all the words. He only needs to know how and where to search for them. Before responding to Harari’s argument, it’s worth first trying to strengthen it.  

An AI does not only “remember” all the words. But as any ChatGPT user can testify, it finds connections between different sources, summarizes them, generates challenging questions, and even deduces new insights. It can therefore seem as though these machines are well-versed in almost any topic.  

This is an interesting theory, mainly because in Judaism, the idea of the “expert” is so quintessential that a term was coined for it: Gadol Hador, which can somehow be translated as “the greatest of the generation.” In our context, however, this means the man who is most well-versed in the Jewish literature. If one substitutes the word expert for great, Harari’s argument becomes more relevant than ever. 

Harari is right that AI is exerting extraordinary power over language. But he is wrong to assume that religion, particularly Judaism, can be reduced to mere language. Judaism is not only a religion of words, but of interpretation, embodiment, law, memory, authority, community, and revelation. 

Harari’s (updated) argument rests on three hidden assumptions. First, that religion is fundamentally made of words. Second, that expertise is mainly textual mastery. Third, that authority follows expertise. I believe all three assumptions are wrong, or at least dangerously incomplete. 

One can tackle this argument from two different directions: from the AI perspective and from the Jewish perspective.  In recent years, I have been studying and lecturing on the intersection of AI and Judaism, and on what each discipline can teach the other. The first two points below stick to the more traditional arguments regarding the scientific nature of AI and LLMs, while the third point offers a more Jewish-oriented perspective. 

The Map Is Not the Territory 

To say that books are “just a combination of words” is very similar to saying that consciousness is just currents moving through neurons. Our experience of ourselves is profoundly different from electric currents. 

The philosopher Alfred Korzybski coined the concept that “the map is not the territory.” A word is not a “thing.” The word “table” is an abstraction, derived from something in the real world on which you can place a glass of wine. But try putting it on nothing but the abstract word itself and see where it gets you. Language maps words onto a territory of meaning. An eye-opening example of this distinction is when people try to “find the right word.” They already know what they want to convey, the “territory,” but they have not yet found the specific “mapping” in English. A large language model, or LLM, is a master of mapping, but it has no access to the territory. It doesn’t even grasp its existence. 

This is not a mere philosophical discussion disconnected from reality; there is a lot of money involved. Just recently, Yann LeCun, a prominent figure in the AI industry, raised over a billion dollars for his new startup, AMI (Advanced Machine Intelligence), a tool that is designed to understand the physical world. According to LeCun (and many others in the industry), LLMs can manipulate words masterfully. But they lack a deep understanding of the meaning of these words. 

To understand the difference, imagine a glass standing near the edge of a table. A language model may know thousands of sentences about glasses, tables, gravity, and falling objects. It may correctly complete the sentence: “If you push the glass too far, it will fall.” But a world model is something deeper: it is an internal representation that can predict how the glass will move, when it will cross the edge, what will happen after it falls, and what action would prevent that outcome. The word “glass” is part of the map. The fragile object, its weight, its location, the table’s edge, gravity, and the possible future in which it crashes to the floor: that is the territory. 

But is the deep understanding of words, an understanding of the broader territory, really mandatory? 

Deep understanding may not be mandatory for many practical tasks. Even according to the logic of the Turing Test, if a machine acts intelligently in a way indistinguishable from a human being, we are inclined to call it intelligence. This also explains how AI can sometimes produce what looks like something completely new: a mathematical theorem, a scientific insight, or even a novel path toward curing a disease. But in many such cases, as Ecclesiastes put it, there is nothing new under the sun: the building blocks already existed, and the machine simply rearranged them in a way no one had seen before. To be fair, for the most part, human creativity works in a very similar way.  

Religion, however, is not merely a practical task of retrieval. It is a domain of meaning, obligation, and lived truth.  

The Average vs. the Exceptional 

Can AI replace human beings? The question is too vague to be useful. Replace them in what sense? In physical labor? In professional work? In moral judgment? In artistic greatness? In the creation of meaning? 

Moravec’s Paradox reminds us that what looks difficult to us is not always difficult for machines. High-level cognitive tasks, like solving mathematical problems, writing code, drafting legal documents, and summarizing research, may be easier to automate than sensorimotor tasks that children perform without thinking, such as tying shoelaces, catching a ball, or moving through a crowded room. If we narrow the question to the non-physical workforce, then the answer is probably yes: AI can replace many people in many tasks. 

But this is where the discussion usually breaks down. Most people are not Newton, Tolstoy, Shakespeare, Euclid, or Isaiah. Much of human work is average by definition, and average work is precisely what machines are becoming very good at reproducing. LLMs already write code better than many programmers, draft texts better than many professionals, and assist with mathematical reasoning at a level that would have seemed absurd only a few years ago. For example, GPT helped to solve an extremely complex mathematical problem after more than fifty years. 

And yet, average performance is not the same thing as exceptional human skill, knowledge, or intelligence. Most people are not as smart as Newton, as insightful as Tolstoy, or as penetrating as Isaiah. Some people are indeed extremely exceptional, in ways we cannot even understand. 

This is the distinction we often miss. Modern people tend to think of “human potential” statistically, as the average ability of the species. But in the Aristotelian sense, potential is not defined by the average; it is revealed by the highest form a thing can reach. The potential of humanity is not measured by the median office worker, but by the rare individuals who show what human beings can become. 

So the real question is not whether AI will outperform the average person. It already does in many domains. The deeper question is whether, in the age of ChatGPT version 42, Shakespeare’s Hamlet will still be irreplaceable, that the Elements of Euclid will still be taught, and that the message of the Bible, the message that transformed humanity, will still speak to the human condition. I believe so. 

If you are still skeptical, try to think of it this way: in many respects we are different from the ancients, but in many other respects we are exactly the same. This is why we can still read, understand, and be transformed by their texts. Our capacity for love, greed, meaning, justice, hope, and cruelty has not changed much. Only the external metrics have changed. So the question could be phrased this way: in the age of GPT version 42, will human beings be profoundly different from us? I doubt it. And if we, and our ancestors before us, were moved by scripture and found relevance in it, so will human beings in the future. 

For now, AI can, and will, give you great suggestions and advice. It will help you learn any new subject at your pace and in the exact way you want. But when it comes to truly learning, to forming yourself as a whole person, and to engaging in the kind of meaningful inquiry that can change how you understand yourself and the world, it is not there (yet). Not because it does not know how to convey the right words. It does. And it does so flawlessly. It is because, for us human beings, the right words were never enough. 

In his book “Thinking Like a Human: The Power of Your Mind in the Age of AI,” David Weitzner stresses this very point. Speaking about the Jewish methodology of learning, “Chavrusa”, he notes: 

“Chavrusa” is an Aramaic word derived from the Hebrew chaver, meaning “friend.” The method is a long-term, cooperative, intellectual endeavor between pairs featuring both social and practical components. There are five principles: It is a relationship of equals regardless of preexisting hierarchies; there is sustained eye contact; both parties are in a constant state of engagement—each listens with the same intensity as speaking; each is challenged and confronted, not allowed to remain passive; and it is built on the language of friendship—when chavrusas leave the table, they are still involved in each other’s lives. Think of how different this is from an AI mentor. 

The idea of mentorship is not restricted solely to the religious world. Something genuine happens in the physical mentor-mentee relationship. Without entering the remote work debate, one thing is missing when working from home—the chance of actually meeting great people who can inspire us.  

For example, research conducted several years ago on the motivational guru Tony Robbins showed that in his live seminars, the audience duplicates his energetic biochemical state of a high testosterone-to-cortisol ratio (a unique combination of high energy and low anxiety). Whatever one thinks of Robbins, there is clearly something about his energetic human presence that moves people to try to transcend their physical limits. This emotional response cannot be achieved just by reading books. You have to be there. 

We need real physical connection. And the underlying mechanics of this connection, for example, mirror neurons, are irrelevant. A WhatsApp conversation lacks something profound, such as context and intonation, that animates a real conversation. And a chat with the greatest machine ever lacks something profoundly meaningful, profoundly human. 

Having established this context, I now want to turn to three concrete issues with Harari’s logic. 

The Oral Tradition 

First, Harari ignores (or severely underestimates) the role of the Oral Tradition. In Judaism, the oral tradition is much more important than the written word. Moreover, according to the religious Jewish tradition, the written scripture is often unintelligible without human interpretation. And this human interpretation is even more important than the original meaning of the divine. The Talmud emphasizes again and again, in many stories and anecdotes, that the original meaning, what God actually wants, is less important than how the scholars understand the text 

If you want to see the difference between the two, The Year of Living Biblically: One Man’s Humble Quest to Follow the Bible as Literally as Possible by A. J. Jacobs is instructive. Even the Jewish legal corpus, the body of rules governing what one should and should not do, is called Halacha, a word related to “walking” or “going.” It is not a static set of answers, but a living tradition that continues to develop as scholars interpret inherited principles in light of new circumstances. 

The Idea of Embodiment 

Human beings can think of themselves as brains inside a machine, but this machine of the body plays a vital role in how we perceive the world. Take the integers and the concept of counting, for example. It seems like a universal truth that they exist and that we can comprehend them. However, the late mathematician Sir Michael Atiyah remarked:  

It might seem that with the integers we are on firmer ground, and that counting is a primordial notion. But let us imagine that intelligence had resided, not in mankind, but in some vast solitary and isolated jellyfish, buried deep in the depths of the Pacific Ocean. It would have no experience of individual objects, only of the surrounding water. Motion, temperature and pressure would provide its basic sensory data. In such a pure continuum the discrete would not arise and there would be nothing to count. 

According to Atiyah, man created, and did not merely discover, mathematics by abstraction of elements from the physical world. Even our math was created in our image. 

The importance of the embodiment cannot be stressed enough. This is also why Professor LeCun believes that for real intelligence, a mere language model is vastly insufficient. In the same way, the religions of the book are not religions of books alone. They are also, and not less importantly, religions of bodies, family life, place, memory, meals, law, and calendar. Even perfect textual mastery would still fall short of the religious life because religious truth is not only contemplated but enacted. 

Reason vs. Revelation 

LLMs are treated as a black box for most of us. It feels like magic, and it feels like genuine understanding and genuine intelligence. If the meaning of creativity is coming up with something new and valuable, these systems also present genuine creativity.  

While the math behind these models is not trivial in any way, the basic premise is that LLMs manipulate words according to the probability of past text. It is correlated with reason. But as Scholem, the founding father of Kabbalah research, noted, the whole point of a revelation religion is to tell you things you could not figure out yourself using logic alone. 

In my lectures I talk a lot about AI models as optimization tools, very similar to Waze. Assuming you want to go from A to B, these models will produce the best route. Best according to whom? According to the metrics you define. It can be the fastest route, the shortest route, the smoothest route, etc. Just define your metric and trust the system. But staying in the Waze analogy, the app cannot tell you whether you should go from A to B. Maybe C is a better destination. How about staying longer at A? 

In English, this distinction is captured by the words efficiency and effectiveness. AI systems are remarkably efficient: once a goal and a metric have been defined, they can often find the best route. But effectiveness asks the deeper question: is this the right goal in the first place? 

This is why Harari is right about AI’s growing power over religious texts, but wrong about what this power means for religion. AI may eventually become the greatest textual expert the world has ever known. But mastery of religious language is not mastery of religion itself. Religion asks not only how to reach a destination, but which destination is worthy, why it matters, and what kind of people we should become along the way. AI might be the greatest expert at putting words in order, but it remains a map without a destination. 

Image licensed via Adobe Stock.