Dispute and Truth – Lesson 14

Table of Contents

• Okimtot and a general principle versus a particular case
• Positivism versus casuistry and the justification for proceeding through cases
• Giving a bill of divorce as an example of the impossibility of exhausting a rule
• Rules as a model: grammar, intuition, and the way to build a “sense of smell”
• AI, neural networks, and the weakness of rule-based thinking
• The limits of generalization: Wittgenstein, sequences, experiment, and measurement
• Rules in the Talmud as limited tools: “one does not derive from general rules” and halakhic examples
• One benefits while the other does not lose: moving from case to rule and back to distinctions
• Positivism, Nuremberg, a “black flag,” and transgression for its own sake
• Autonomous technology, regulation, and a phobia of rules
• Da’at Torah and the scope of the authority of Torah intuition

Summary

General Overview

The text argues that the Talmud and the Mishnah use particular cases in order to express general principles, and that okimtot are meant to justify how the chosen case fits the principle without qualifying the principle itself. It presents the tension between positivist thinking, which seeks sweeping rules, and casuistic thinking, which builds rulings out of cases and precedents, and it argues that systems of Jewish law and civil law cannot be fully exhausted by rules. It explains that rules are models and approximations constructed in order to organize intuition and activate analogy, and that you have to “use them and throw them away” rather than become enslaved to them. It reinforces this through examples from Gittin, Bava Kamma, Kiddushin, disputes between Abaye and Rava, verbal analogy, and then connects this to the world of AI as neural networks that do not operate according to explicit rules.

Okimtot and a General Principle versus a Particular Case

The Talmud starts from the assumption that when the Mishnah or an early Amora brings a case, the case is meant to state a general principle and not a point-specific law. The okimtot explain why the chosen case really does illustrate the general principle when difficulties arise that depend on the details of the example. The example of acquiring a bill of divorce through the woman’s slave teaches that the slave is considered like her courtyard, and the requirements such as “a courtyard guarded with her knowledge” and “a courtyard that is not moving” require, within the laws of acquisition, okimtot such as a slave who is bound and asleep. The principle that the slave is the courtyard of his mistress remains true even without the conditions of being bound and asleep; the okimta relates only to the ability to use that particular situation as an example of acquisition.

Positivism versus Casuistry and the Justification for Proceeding through Cases

A positivist mode of thinking assumes sweeping legal rules and describes the judge as a deductive machine deriving the particular from the general, as is described regarding German law. The casuistic mode of British common law relies on precedents, analogy, and induction, and rules on the basis of similarity to previous cases rather than from general statutes. The casuistic method is presented as sensible because the attempt to exhaust a legal system as a set of laws will always fail, since law and Jewish law are broader and more flexible than a collection of axioms. The text compares this also to limitations in mathematics through Gödel’s theorems in order to emphasize that even formal systems are not complete in a way that allows absolute deduction for every situation.

Giving a Bill of Divorce as an Example of the Impossibility of Exhausting a Rule

The verse “And he shall write her a bill of severance and place it in her hand” is presented as an especially overloaded verse in terms of exposition, and every word in it generates a “sea of topics” in tractate Gittin. The act of “placing it in her hand” receives in the Talmud dozens and dozens of examples of valid and invalid forms of giving, and from them one is supposed to understand what “giving” means in divorce. Ketzot tries to define giving as transfer of ownership but runs into the validity of writing it on things from which benefit is prohibited, and on the other hand runs into the fact that physical giving does not exhaust the cases because one can give through a courtyard and by transferring ownership of the courtyard. Kehillot Yaakov goes through many cases and distills a definition that keeps swelling as more cases are covered, until it becomes almost as complex as the number of cases themselves. The conclusion is that there is no “general law” according to which the Sages worked; rather, the Sages ruled out of a kind of “sense of smell,” and the rules formulated by learners are a model built in order to get close to the whole set of cases and apply it to new ones.

Rules as a Model: Grammar, Intuition, and the Way to Build a “Sense of Smell”

Rules of grammar are described as an approximation built after the language already exists, not as the basis from which the language was created, and therefore every rule has an exception. A native speaker operates intuitively, while the later learner needs rules in order to organize the language, but cannot identify in them “the real thing.” In Jewish law and in law generally, rules are presented as a way to build a framework of thought that enables analogies, but not as a faithful reconstruction of the legislator’s intention or of the way the Sages worked. The rules are created, updated, and thrown away again and again in the face of conflicting cases, and this process organizes intuition so that it becomes possible to rule correctly in future cases even when the rules themselves do not hold up.

AI, Neural Networks, and the Weakness of Rule-Based Thinking

Classical programming is presented as giving a computer rules that it applies to inputs, whereas neural networks learn from lots of cases and correct answers without any explicit rule. The question “what was the rule according to which you worked?” is described as meaningless with respect to a neural network, because the decision is produced out of an internal structure built from examples. The text argues that this ability demonstrates the weakness of positivism and the power of case-based thinking, and that this is how the Talmud works when it prefers to present cases that organize within the learner an intuitive “neural network.” Building rules by learners is described as a technical way of organizing that network, even if in the end “you can throw the rule in the trash” and use a sense of smell.

The Limits of Generalization: Wittgenstein, Sequences, Experiment, and Measurement

The text presents a didactic claim that mathematics is learned better through examples and only afterward through the formulation of a general theorem, because without examples there is no understanding. Wittgenstein, in Philosophical Investigations, is presented as arguing that there is no way to teach rules except through examples and “and so on,” but that same “and so on” is open to an infinite number of possible continuations. Sequences like “three, five, seven” allow a continuation of primes or odd numbers, and one can even tailor a rule that gives “minus pi” as the continuation, so no rule is forced by the examples. Even in physical measurements, a finite number of points allows infinitely many continuous curves, and the choice of a “straight line” is presented as a human decision of simplicity in the style of “Occam’s razor” and not as a logical necessity. The text concludes that rules are a useful fiction that allows movement from example to example, whereas “the truth is in the examples, not in the rules.”

Rules in the Talmud as Limited Tools: “One Does Not Derive from General Rules” and Halakhic Examples

The Mishnah in Bava Kamma that brings “four primary categories of damages” and concludes with “their common side” is presented as a case where the Talmud asks, “what does the common side come to include,” and the text emphasizes that the question is directed against relying on the rule and not against the examples. The rule “it is your property and its guarding is upon you” has trouble containing a custodian and a robber, and the solutions that try to “push” them into the rule are presented as evidence that the rule is not exhaustive. In Kiddushin, the rule is brought that “all positive commandments dependent on time, women are exempt from,” and the Talmud adds, “one does not derive from general rules even where ‘except’ is stated,” and the text emphasizes that this undermines even carefully formulated rules with detailed exceptions. The rule “the law follows Rava except for YAL KGM” is presented as only a general direction, and Maimonides, who rules like Abaye in additional places, serves as proof that one should not become enslaved to the rule but use it when there is no other basis for decision. The conclusion of this section is that the Talmud treats rules with “limited liability,” and the rules are a guiding framework but not a substitute for judgment.

One Benefits While the Other Does Not Lose: Moving from Case to Rule and Back to Distinctions

In the topic of “one who lives in another person’s courtyard without his knowledge” in Bava Kamma, the Talmud begins with a concrete case and moves it into the general framework of “one benefits while the other does not lose.” After formulating the rule, the continuation among the medieval authorities (Rishonim) and later authorities (Acharonim) produces many distinctions that make it hard to apply the rule directly, with distinctions such as a courtyard that is for rent, a person who is apt to rent, “he saves himself an expense,” and “driving away a lion.” The resulting picture is that even a rule formulated by the Talmud itself is not a rigid computational tool but a stage that helps thinking and is then pushed aside in the face of the complexity of cases. The text presents this as an illustration that the Talmud usually remains on the level of cases, and when it formulates a rule it “works like the Acharonim,” and still does not cling absolutely to the rule.

Positivism, Nuremberg, a “Black Flag,” and Transgression for Its Own Sake

The Nuremberg trials are presented as a crisis for positivism, because the claim “I obeyed the law” is a natural defense argument in a system that sanctifies rules. The criticism of the defendants is described as the application of natural law and the determination that there are laws that must not be obeyed, in the Israeli image of a “manifestly illegal order” bearing a “black flag.” The topic in Nazir of “a transgression for its own sake” and Yael, the wife of Hever the Kenite, is presented as a situation in which there is no halakhic justification within the rules, and yet the act is required by the situation, and the attempt to turn this into a “rule of override” is described as a mistake. The text emphasizes that deep moral or halakhic judgment sometimes requires breaking the rules, and sometimes there is even a claim against a person for not breaking them, as was described in relation to Nuremberg. The conclusion is that even when new rules are written after a breach, they too will be breached, and therefore one must make use of rules without turning them into “the thing itself.”

Autonomous Technology, Regulation, and a Phobia of Rules

The text argues that fear of autonomous cars stems from the mistaken perception that they operate according to rigid rules, whereas in reality they are based on AI that does not work that way. It argues that machines can already today make decisions better than humans in many fields, and that approval of them is delayed because of a phobia of technology and an unreasonable standard of “zero mistakes” instead of “fewer mistakes than humans.” It compares this to the linguistic whitewashing in the IDF between an unmanned aerial vehicle and a remotely operated aerial vehicle, in order to avoid accepting autonomous decision-making, and describes this as an unjustified fear that prevents saving human lives. It adds that the claim that “people rely too much on the machine” is not a problem in his view so long as the machine is better than the human alternative.

Da’at Torah and the Scope of the Authority of Torah Intuition

The text presents the assumption of “Da’at Torah” as the claim that someone who has internalized enough topics and cases acquires a “head” that enables him to decide even in areas that are not his formal field of expertise. It accepts “Da’at Torah” with respect to halakhic questions and attributes to a Torah scholar an advantage there, but rejects applying it to fields that are not Jewish law, such as investment decisions. It describes a distinction between “proofs,” which can be argued about, and a situation in which “the mind says, it seems to me,” where the ruling rests on intuition built from accumulated halakhic experience. It concludes that when a Torah scholar puts his intuition on the scales in a halakhic context, it is likely that the one who disagrees with him is the one who is mistaken.

Okay, we’ve more or less finished the issue of these interpretive qualifications, and what lay at the basis of the picture I drew there was that the Talmud is basically working from the assumption that when the Mishnah brings a case, or an early Amora, it doesn’t matter, brings a case, the assumption is that he is really coming to tell us a general principle, not a law about one specific case. And the case is just an expression of the general principle. Then, when all sorts of questions arise for us that are not connected to the general principle but to the specific case we chose, through which we chose to illustrate it, then we make interpretive qualifications. Meaning, those qualifications basically come to explain why this case illustrates the general principle, but the principle exists regardless of the qualifications. Right? The example of the slave through whom I transfer the bill of divorce to the woman by placing it in the hand of her slave—that comes to teach us the general principle that the slave belongs to, that the slave is considered the courtyard of his mistress. True, in order for him to acquire it on her behalf it’s not enough that he be her courtyard; he also has to be a courtyard that is guarded with her awareness and a courtyard that is not moving, so he also has to be bound and asleep and I don’t know what, lame. But that whole story is needed only so that we can teach, so that we can say that the Jewish law is that he acquires on behalf of his mistress. But the fact itself—that he is her courtyard—that is true even if he is not bound and not asleep, any slave. And therefore, when we talk about a slave in general, the novelty we are trying to teach is a novelty that is true of every slave, not specifically a slave who is bound and asleep. Rather, we implemented that novelty, or expressed that novelty, through the example of acquisition, of courtyard acquisition. But in the laws of acquisition, the courtyard also has to be non-moving and guarded with her awareness, so that’s why we made it a bound and sleeping slave. But the qualification does not come to limit the general rule we are trying to teach, namely that the slave is the courtyard of his mistress. The qualification only comes to say that this example of acquisition demonstrates the general idea that the slave is a courtyard only in the case where the slave is bound and asleep. The qualification is there for the sake of the example, but it does not touch the general principle. And that is basically the idea; I showed this in several, in the three passages that accompanied us in this discussion. What I still want to do before I move on is talk about something I actually already touched on in the first semester, and that is: why, really, why doesn’t the Talmud state the general principle directly, but always chooses—or the Mishnah chooses—to express it through a specific case, a specific illustrative case? I said that this is connected to two forms of legal thinking. There is a positivist form of thinking, which basically assumes that there are sweeping legal rules, and the particular cases that come before the court are specific instances of the general laws. And what I need to do, or what the judge’s role is, is basically to derive the instruction for the specific case before him from the general laws that appear in the lawbook. In that conception, the lawbook really ought to contain a set of general principles, and not deal with cases at all. Cases are for the judge—the judge will see which of the rules the case before him fits, and derive from the rules the law for that particular case. Right? The judge is basically a deductive machine, meaning a machine of deductions. In law, in German law for example, that was the accepted conception; today everything has been moderated somewhat, but in German law that was the accepted conception, the Bundeslaw. In common law, British law, the conception is the opposite. There the conception is casuistic, meaning we basically rely not on general laws but on specific cases, and those specific cases serve as precedents. We build the ruling for a new case before us by analogy to previous cases. There are no general laws—it works by analogy to previous cases. If it resembles that case, then the law will be such-and-such; if it resembles this case, then the law will be such-and-such, and so on. Therefore, if in the positivist world the basic logical tool is deduction, moving from the general to the particular, then in the casuistic world the basic tool is analogy and induction. Analogy and induction: I take one case and make an analogy to another case. I don’t derive from a rule to a specific case the way you do in the positivist world; rather, from one specific case I make an analogy to another specific case, or make a generalization to other cases.

Now, on the face of it, this seems very strange. Why really—why not just give me the general law and I’ll derive the specific case from it? Why go about it this way? Why tell me a specific case, from which I’ll then have to derive the general law and then apply it to another specific case? Just give the general law directly and that’s it. Why illustrate it through a specific case? The casuistic method is, seemingly, an illogical method. Why use such a thing? So here I want to explain what the logic behind the casuistic method actually is. What stands behind it is basically the conception that says you cannot really present the legal system as a collection of general laws. The attempt to present the legal system as a collection of general laws will fail; it will always fail; there is no chance. Because law really is some collection of ways of looking at things that I cannot exhaust in a set of rules. There is something—something more amorphous, broader, more flexible, however you want to call it, than a set of rules. You will not succeed, you simply will not succeed in finding a set of rules that really manages to exhaust what is right in the eyes of the legislator, or what is right to do with respect to all the individual cases. Right, right, right—even in the mathematical world, we know there is a limitation on this perspective, even though mathematics is seemingly one giant instance of positivism. Meaning, give me the set of rules and from that I derive deductively the individual cases or the implications, theorems of one kind or another. But even there, as you know, there are Gödel’s theorems and certain limitations on the completeness of a system and the consistency of a system. So even in mathematics we know that in fact you cannot do this fully. All the more so in less exact worlds like the world of law or the world of Jewish law and so on. It is very hard—impossible—to present it as a set of axioms or general laws from which I can derive the laws for all specific cases.

Maybe a nice example of this: I once wrote about it when I was here doing my doctorate, in Bar-Ilan’s weekly Torah-sheet, on the weekly portion of Ki Tetze. I talked there about the laws of divorce, the commandment of divorce. And I said that when we look in the Talmud, the Torah says: “And he shall write for her a scroll of severance and place it in her hand.” That is the most expounded verse in the Torah. Every word in that verse is mountains of laws in tractate Gittin. “And he shall write,” “for her,” “a scroll,” “of severance,” “and place,” “in her hand.” Every one of those words is an ocean of passages. Every word. I don’t think there is another verse so loaded with laws that are derived and expounded from it. Let’s focus for a moment on “and place,” “and place it in her hand,” if you like. That defines the process of the act of divorce—not the writing, not the preparation of the bill of divorce, but the act of divorce. Taking the bill of divorce that “he shall write for her a scroll of severance”—that defines the preparation of the bill. What is a bill of divorce? What is the thing that has to be given? But once it is ready, you have to take that thing and place it in the woman’s hand, and that is the divorce. And what is “and place it in her hand”? How do you define the act of giving? How do you define the act of giving? In the Talmud there are many dozens—if not more than that—of examples of valid givings and invalid givings. He gives her the bill of divorce in her hand and then pulls it back with his hand. He puts the bill of divorce in his courtyard and transfers it to her. He puts the bill in her courtyard. He puts the bill with the slave who belongs to her, what we saw. It was written on things from which benefit is forbidden. Millions of forms of writing and giving a bill of divorce that are valid or invalid, and from all that we are supposed to understand what it means.

Now the later authorities, as is their way, try to refine out of all these cases a definition, a general law—yes, the dreams of the positivist. Let’s try to extract from this collection of cases a general law, a general definition of what the giving of a bill of divorce is. So the Ketzot, for example, tries to think maybe it is the transfer of ownership of the bill of divorce. And it cannot be transfer of ownership, because if it was written on things from which benefit is forbidden, it is valid. And regarding things from which benefit is forbidden, there are opinions at least among the medieval authorities (Rishonim) that there is no acquisition in them. So the giving of a bill of divorce is not transfer of ownership. So what is the act of giving? Maybe it is a physical giving—taking the bill and putting it in the woman’s hand? That also doesn’t work. Because you can place it in your courtyard, even—certainly in her courtyard, but you can even put it in your own courtyard and transfer the courtyard to her. So again we are back to transfer of ownership. So is it transfer of ownership or not? Is it an act of giving or not? None of these definitions exhaust all the cases. It’s impossible.

Now the one who goes even further is the Kehillot Yaakov on Gittin, and there he has two or even three sections in which he goes through many dozens of cases and tries to extract a definition from them. And you can see—it’s absolutely enlightening to watch this, because the definition keeps swelling as you cover more cases. We begin with a simple definition, then we bring a case and it doesn’t fit. Ah, so we have to shape the definition differently, add another layer to the definition—but that also doesn’t work with the third case. So we add another layer to the definition. In the end, what comes out is a definition whose number of components is more or less the number of cases that appear in the Talmud. So what did you gain? You’re trying to extract a definition that will be more concise, where one law will exhaust all the cases. But once the definition is no less complex than the collection of all the cases, then why make it at all? What have you gained? So in the end there is no real general law here; you are simply trying to present all the cases in some supposedly general formulation, but it is nothing but the collection of cases that appear in the Talmud. So what is the meaning of this whole process? Why don’t they just give me directly—why did the Sages make things so viciously complicated for us? Why don’t they just tell us the general law directly instead of giving us cases and we try to extract the general law and fail and get tangled up? Just give me the law. How did you decide that this case is valid and that case is invalid and the other case is valid? You too decided according to some criterion. Before you, meaning before the Sages, all these cases were not there. How did you decide? You had some standard, some criterion of what counts as a valid giving and what doesn’t. So give me that! Why bring me all these strange cases?

The answer is that the Sages did not have one. They did not have such a law. So how did they decide which giving is valid and which is not? By smell. This giving doesn’t seem right to me; this giving does; that one does; this one doesn’t. That’s it. They gave you dozens of cases—this one no, this one yes, this one yes, this one no. Okay, now what am I supposed to do with that? What I am supposed to do with that is try to understand from these cases what to do with other cases that do not appear in the Talmud. How do I do that? I basically have to make an analogy. How do I do that? What I really do is take these cases and try to build from them general laws in order to apply them to the coming cases and determine whether this is a valid giving or an invalid giving. Except there really are no such general laws. I am not reconstructing what the Sages did. The Sages did not have general laws. The general laws are laws that I create. They are not really a reconstruction of what the Sages did. So what is the meaning of those laws? The meaning of those laws is that this is the best approximation I have to a theory that will fit the cases in the Sages. I am trying to build a theory that will fit the cases in the Sages. A theory, in the nature of theories, is built out of general principles, but those general principles are ultimately just a model I build by means of which I can think more efficiently or better about the Talmudic cases. That’s all. It is not really a reconstruction of the rules according to which the Sages worked, simply because there were no such rules. They did not work according to rules. So why shouldn’t I continue with my own nose? They worked with instinct, I’ll work with instinct too. Because I am trying to adhere—the assumption is that the Talmud has authority; it determines what a valid giving is and what an invalid giving is. So if I want to stick to their instinct, I can’t just go with my own instinct. But how do you adhere to someone else’s instinct? There is no way. What you can do is try to define rules that will operate similarly to his instinct, and go with those. What? That’s what the Talmud did. Yes, that is what it did. Yes, there is a gain in building these rules. Obviously, obviously—I’m coming to explain what it did. I think what it did is an example of what all of us do, except there it is very blatant, so there you can actually see what the idea behind the move is.

I think I spoke about this in other contexts—I don’t know in this series, maybe yes—about Moshe Koppel’s language model, with the grammatical rules of a language. Did I talk about this? Briefly. What he basically says is: think about the grammar rules of a language. The language didn’t develop according to those grammar rules, right? A language develops naturally among speakers somehow, and after a long time a language emerges. Okay? It’s not that they had a set of rules and built sentences according to those rules and that’s how they learned to speak. So what are the rules, the grammatical rules? After the language already exists, scholars come—linguists and others—and try to understand or distill from natural speech a set of rules. And after they distill that set of rules, for them that is the model by means of which one can describe speech in that language. Now we know that every such rule has exceptions. Why? Because the rules are not really the language. They are a model that helps us approximate the form of speech as best we can. Because we want—instead of learning the whole collection of possible sentences—I want some finite set of rules with which I can use and speak the language. Otherwise I would need to learn infinitely many sentences. I can’t learn it that way. So I need to use rules. But all the while I have to remember that those rules are only an approximation, and every rule has an exception. And therefore the discussion—the rules, the rules of language—are on the one hand not exact; they have exceptions. On the other hand, I really can’t learn a language without them. I need some framework within which I work. So I use rules, but all the time I’m aware that this is not really the real thing and there are sometimes exceptions. And when I spoke about this, I also mentioned that the native speaker—the one who speaks the language as a mother tongue—doesn’t go through the rules. He simply learns it naturally. The speaker who arrives later in life goes through an ulpan. In the ulpan they teach him through rules.

It’s the same in Jewish law and the same in a legal system. The rules are not really what lay at the basis of the legislator’s thinking. The legislator wanted to establish certain laws or certain statutes according to what he understood. But in order to try to apply those things and make analogies and apply them to other cases, we need somehow to understand, or to build, a model that is a collection of rules that defines what the legislator meant. Then we take the rules and apply them to new cases. But it’s not that those rules are really a reconstruction of what the legislator did. Not at all. It is simply our way of getting close to what the legislator wants and applying it to new cases. So if I return to the Talmud in Gittin, to those passages about giving a bill of divorce: the Sages had some instinct for what counts as a valid giving and what counts as an invalid giving. They conveyed it to me through many dozens of cases. They couldn’t give me rules because there weren’t any; they didn’t have rules. But I don’t know what to do with lots of cases. What am I supposed to do with all those cases in relation to a new case? How do I compare, or how do I infer conclusions from those cases to a new case? I try to extract rules from those cases, and with those rules I decide what happens in the new case. But I always, always have to remember that the rules are some kind of approximation, and instinct must not be ignored; it too plays a part in the matter.

By the way, that’s a contemporary comment, a modern comment, yes, about artificial intelligence, about AI. Artificial intelligence, unlike classical programming, works exactly in a non-positivist way. Classical programming basically says: give the computer the rules, and it will apply them to the cases that come before it. Explain to it how to calculate addition, and then ask it eight plus four and it will give you twelve, because it applies the rule programmed into it to that specific case and gives you the answer. So classical programming gives the computer the rules, and the computer performs the calculation on particular cases that come before it. But a neural network, or AI, basically works differently. It gives the computer many, many cases, and the computer says what the result will be in the next case. There are no rules. The computer uses those cases to organize its own internal network through the cases. It receives a case and is told what the correct answer is. So through the cases and the correct answer—say in supervised learning, for those who know—it builds its internal network. And after it has built that network, give it a new case and it will give the correct answer to the new case. If you were to ask the computer, “Tell me, what is the rule by which you worked?” that’s a meaningless question. It does not work by a rule. It simply takes the cases, and from those cases it produces an answer for the next case. It doesn’t pass through rules.

We, as human beings—or as conscious human beings—our consciousness generally works with rules. You take the cases, you want to understand what from these cases I can apply to the next case, so I try to understand what general ideas stand behind the cases. Then, once I have the general ideas, I apply that to the next case. But the computer in a neural network doesn’t work like that. It doesn’t think at all about the general law. It takes the specific cases, builds a network from them, and produces an answer for the next case. And it turns out that this is probably something much more powerful than going through rules. And that is a marvelous proof of the weakness of positivism. People think that working through rules is a much stronger and more precise method. You could not be more mistaken. Clinging to rules leads to many, many errors and many, many computational limitations and many, many weaknesses. Because real thinking does not work with rules. Rules are an approximation. They’re an approximation for the weak. Rules are for the weak. Whoever uses rules is someone who has no other way; he can’t do it otherwise, so he uses rules. But if you work in a casuistic way, inferring from cases to the next case—which is exactly what a neural network does—you arrive at much stronger abilities. And I think that what the Talmud does, precisely because it does not believe in rules, is that it prefers to give me cases. And from those cases my neural network will arrange itself so that I infer a conclusion about the next cases that come before me. And that is stronger than giving me the rules. That is basically the claim.

In a certain sense there is here a foreshadowing—yes, with a tav, a foreshadowing—of an insight that today it is much easier for us to understand: the weakness of thinking by rules, the weakness of positivism, and the power of thinking by cases. Simply to use cases to build my instinct, and with that instinct infer conclusions for the next case. And what the Sages wanted to do, I think, in these divorce cases, was that I should rub up against many dozens of cases; this would help me build a healthy instinct, or in more modern language, organize my neural network. Then, when a new case comes before me, I will arrive at the best conclusions regarding it—and not through rules.

So what is the Kehillot Yaakov doing with rules? And not just the Kehillot Yaakov—all of us. This is simply our way of building the neural network. We do not really work with the rules. Rather, we take a few cases from the Talmud and build a rule. Then we discover that this rule doesn’t work—here is another case that contradicts the rule. We update the rule. Then there is yet another case where it doesn’t work. We update the rule again. In the end, in the end, you can throw the whole rule in the trash; it’s worth nothing. But our neural network has by then arranged itself properly. Now activate your instinct, and if a new case of giving a bill of divorce comes before you, your decision will probably be correct. The way to organize the neural network is to try to build rules, but those are only negative descriptions. Basically, through them we understand only what is not the case, not what is. Meaning, this rule doesn’t work, so I correct it; and the corrected rule also doesn’t work, so I correct it too. And after I have corrected everything, in the end I can throw all those rules in the trash. My neural network is now built. Now I can act intuitively. Give me a case and I’ll tell you whether in that case this counts as giving a bill of divorce.

There are things where they do give a rule. Right. That’s what I’m coming to now. The truth is that, like what the Kehillot Yaakov does—not says, does—we generally cannot work without rules. Even if we wanted to take the examples and swallow all the examples, mix them up well in our heads, and then produce a conclusion regarding a new case—how do you do that? What does it mean to mix it up in your head? Take the examples. Just recite the examples, another example and another example, and then the hope is that I’ll give the right answer? That won’t work. So how do you do this mixing, this internalization of the examples into our way of thinking? Through rules. We try to extract rules from the examples, then a case contradicts them, then another rule. This game with the rules—throwing away one rule and adopting a new one, and then throwing that away too and adopting a new one, and so on—that is basically the game that helps us build our neural network properly. Our way of rubbing up against the examples is to try to understand what rules lie at their base. Even though no rules lie at their base. We try to understand what rules lie at their base, and reject them, and then propose a better rule, and reject that too, and so on. And that is basically our way of shaping our intuition, internalizing all the examples into our thinking, and building a kind of thinking that will give the right answer in the next specific case. And that is how one should relate to rules. Rules, even when the Talmud gives rules, are rules that are only a stage in the building of intuition. Whoever clings to the rules, even the rules the Talmud gives, is making a big mistake. Because even the Talmud did not really intend that we use those rules. Those rules are only a guiding rule, but in the end, in the end, I need to throw them away and use my intuition. The rules only help me build it.

In some sense, is this really more a conception of the person than of the object? Right. Those rules too came out of actions, and they also go back to being applied to actions. The rules do not stand on their own, and they are also not precise; they are also not correct. Like the rules of language—the rules of language have many exceptions—so too the rules of the Talmud have many exceptions. The rules are only a stage along the way, in order to build our intuition. Once we have built it, we can throw away the rules and use it. Even—even with the most rigid rules, like verbal analogy, this works. There is Nachmanides and all his students—the early Spanish sages already pointed out that verbal analogy, even though it says a person may not derive one on his own unless he received it from his teacher, and Rashi always says it is a law given to Moses at Sinai—that is not true, that it is a law given to Moses at Sinai. It doesn’t work like that. The proof is that there are disputes about verbal analogy. How could that be, if a person does not derive it unless he received it from his teacher? And the Talmud itself discusses verbal analogies—yes, no, this is a tradition, we received from Sinai to discuss it. Therefore Nachmanides says that when it says a person does not derive a verbal analogy on his own, the intent is not that the verbal analogy was given to Moses at Sinai and then transmitted by tradition. Rather, the verbal analogy is some kind of hint; I’m not shooting in the dark. For example, a tradition was passed down to me that there is a verbal analogy between these two words, but it doesn’t say what I do with it. Or alternatively, one word was passed down to me and it says that one makes a verbal analogy, but it doesn’t say where to. And it doesn’t say—rather, it was said. It was not said to where. Or the law was passed down to me but they didn’t tell me it came out of a verbal analogy, or they told me it came out of a verbal analogy but didn’t tell me from where. In short, verbal analogy requires some hint in order to perform it, but in the end, in the end, you perform it. The hint only helps, points you in the right direction, but you do it.

There is a book by—what’s his name?—on verbal analogy. He has books on several of the interpretive methods, Michael Chernick. Michael Chernick. He put out a book on verbal analogy, on several interpretive methods, verbal analogy, and among other things he argues that the early verbal analogies were verbal analogies of the sort made between two words whose only appearances in the Torah are those two. Only words that appear twice in the Torah. And if they appear twice in the Torah, apparently the Torah intended that we make a verbal analogy between those two appearances. But with the progress of the generations we make verbal analogies even between words that appear more than twice. And then of course it is much less unambiguous—when do you decide whether to make a verbal analogy or not to make one, maybe between these two, between those two, when yes, when no. There is instinct here. It’s impossible—this is no longer mathematics. And so on. I gave several examples of this when I spoke about it; I think I surely also spoke about it in the first semester, I just don’t remember. Right. What does that mean, in the Written Torah? Yes. So. When it says from a rule and from an example. Both. Sometimes a rule, sometimes an example. The Ten Commandments or prohibitions—there are prohibitions that themselves are only examples, say the prohibition against eating pork. That is not a rule; it is a specific prohibition: it is forbidden to eat pork. It does not express a rule; it is the prohibition. But there are situations—for example, mixed kinds. Mixed kinds is a rule, but those mixed kinds are given through examples. “You shall not plow with an ox and a donkey together.” “Ox and donkey” is only a case. Or mixed seeds in a vineyard, for example. Grain—and there are many types of mixed kinds, and it comes through examples. So there is both. Meat and milk—is that an example or a rule? What? Meat and milk—is that an example or a rule? It’s an example, an example that yields a rule regarding meat and milk, not specifically a kid in its mother’s milk, and so on.

Yes. I’ll give you a few examples of this idea. You know, in the teaching of mathematics, for instance, there is an interesting question—what is more didactically correct to teach, how is it more correct didactically to teach mathematics? To teach the theorem and then bring examples, or to first bring examples and from the examples, after you understand the idea, formulate the general theorem? Meaning, clearly on the logical level the theorem comes before the examples. The theorem is the general case and the examples are its specific applications. But didactically it is usually better to work in reverse. Take the examples; after people understand the idea from the examples, now formulate the general theorem. Because if you formulate the general theorem, people won’t really understand what it says. Until we—meaning our minds on the one hand always need general laws, but on the other hand they do not understand the general laws unless you apply them to examples. Therefore, for example, in the university, at least in difficult courses, there is a lecture and a tutorial. The lecture gives the general principles, and the tutorial gives you examples so that you understand what the general principles are saying. And I think each of us knows this—that without the tutorial, you don’t really understand the lecture. Without having seen the examples, you don’t understand the general idea.

Wittgenstein went one step further. The well-known philosopher of the twentieth century, in the first half of the twentieth century, said that there is no way at all to teach rules. How would I teach you, for example, how to count? Decimal counting. If you think about it, how do they teach children? One, two, three, four, ten, eleven, twelve, twenty, thirty, ninety-nine, one hundred, one hundred ninety-nine, two hundred, one thousand, and so on, right? Then you say to him, “and so on.” Now imagine you taught him up to ten thousand. You say, okay, now continue. So he says: minus pi. What does he say? The next number after ten thousand—minus pi. What, you don’t understand? After all, I taught you one, two—after ten thousand comes ten thousand one, ten thousand two. You continue with him up to one hundred thousand. You say, what is the next number? e squared. You say, okay, the guy is psychotic, he needs hospitalization. No, he does not need hospitalization. He is right—no less right than you. You didn’t really give him the rule. You gave him an example, examples, and said “and so on.” But that “and so on” assumes that once you took a few such examples, you’ll already know how to continue. We always teach rules by induction. There is no way really to teach rules. Even mathematicians will write n. n is basically the numbers one, two, three, four. But when you try to explain to a person what n is, you say one, two, three, four, and so on. There is no way to teach the rule except through examples.

The problem is that when I give you a sequence of examples, infinitely many rules can be generated from it. I tell you three, five, seven—what is the next number? For example, it could be eleven, if it’s the prime numbers: three, five, seven, eleven; nine is not prime. But if you think it’s the odd numbers, then it’s three, five, seven. Three, five, seven, nine, eleven. Who is right? Both are right. You can decide that the rule is prime numbers; you can decide that the rule is odd numbers. But I can also tell you three, five, seven, and minus pi. No problem. I’ll build you the rule that for n equals one gives three, for n equals two gives five, for n equals three gives seven, and for n equals four gives minus pi. Simply four equations in four unknowns; no problem at all to build such a rule. Whatever you want, your wish is my command. Tell me what you want the next number in the sequence to be; in any sequence you give me, tell me what you want the next number to be, and I’ll generate for you a rule that yields that as the next number. Or in other words, if you failed the psychometric exam, come to me. I’ll be your lawyer, and I’ll show the examiner that you were right in every answer you wrote there. Every answer you wrote there can be rationalized. There is no correct answer in the psychometric exam. It is a mistake to think there is a right and wrong answer there. The psychometric exam does not test who is smart. The psychometric exam tests who thinks conventionally, who thinks the way they expect him to think. If you think the way they expect you to, you passed the psychometric exam with 800. And someone who fails the psychometric exam is simply someone who is not conventional, not necessarily stupid. He will continue the sequence not in the way you expect him to continue it, but in another way, because his head is built differently. But he is no less right than you; he is simply different from you.

But on the other hand, the university is justified in testing its students by psychometric exam. Why? Because otherwise you cannot teach him. If his head is built differently from yours, you tell him three, five, seven, and so on, and he will tell you minus pi as the next number. You won’t be able to teach him what you want to teach him. Only if his mind is built like your mind. So the psychometric exam is ultimately asking whether the student’s mind is built like the minds of his lecturers. Because if not, there is no point in his coming to university; you cannot teach him. Not because he is stupid, simply because he is built differently. Sorry if I sound a bit postmodern, but that is the bitter truth. If he is a genius, he really is not suited for the university. Right, exactly. And someone whose mind is creative, different from others—you won’t be able to teach him, because his head is simply built differently. What? The SAT? Okay, it doesn’t matter. Other exams can also be attacked in the same way, and in fact they have been attacked. There are many criticisms of the psychometric exam. It is culture-dependent and dependent on many things.

So the claim basically says that we live under the illusion that we operate according to rules. But that is what Wittgenstein basically says—late Wittgenstein, in Philosophical Investigations. He says that in fact there is no such thing—this is the “following a rule” argument, one of Wittgenstein’s famous arguments. It is impossible to follow a rule. There is no such thing as following a rule. Following a rule always means taking a few examples and so on. But that “and so on” assumes that once you took a few examples like that, you will already know how to continue. But you might continue it completely differently from what the person teaching you the rule intended, and you are no less right than he is.

Think about continuous examples, not discrete ones. Suppose I made a few measurements of the relation between force and acceleration. Newton’s second law, after all, says that force equals mass times acceleration. Never mind at the moment—three quantities and that is the relation between them. Now I measure the relation between force and acceleration. It never comes out. Yes, it never comes out—that is known. Everybody just copied from Newton. He was the last one to do the experiment and it worked out for him. Or maybe he copied laboratory reports too. I was stupid enough not to know people copy lab reports, so I failed lab. I think I was the first in history to fail lab—that was me. And that is how I understood that my place was in theory, not in… In any event, you test the relation between force and acceleration. Say you apply force one and get acceleration two. Then you apply force three and get such-and-such acceleration. One gives two, three gives six, and so on. And you get several such measurements. Now you ask—think to yourself—what is the general law? You don’t know Newton’s law; now you want to arrive at it. You make measurements. So you found four results. What is the general law? Every beginning physics student will tell you: a straight line. Right? And I ask: why not this? There is software that does that. Yes, and it also explains all the—yes, but software was written by human beings—that’s the problem. The problem with software is that human beings write it. So who says these measurements mean a straight line? These measurements also fit this, or this if you like. It depends at what resolution you made the measurements. Absolute resolution. Yes, a mathematical point. You’ll get a line, and then what? You’ll get a line—what, here are those measurements, I can fit them in infinitely many ways. You can? Because either there is acceleration… You don’t have infinitely many measurements. You have five measurements, one hundred measurements, one thousand measurements—it doesn’t matter—a discrete number of measurements. A discrete number of measurements, that’s what you have. Right? Now you have to fit a continuous curve through them. There are infinitely many ways to do it. Yes, right? Any finite number of points can be fit by infinitely many continuous curves. Even uncountably many. So now how do we decide which is the right curve? The minimal one. Least squares—that is usually what we do. But with least squares you have already decided in advance that it should be a straight line, and you are looking for the straight line that the least squares gets you to. But who says it’s a straight line? Maybe it’s sine x squared? There is no way to go from the discrete cases to the curve. There is simply no way to do that. The decision that if these are the measurements, then the correct curve is the straight line—that is our decision, because we are built in such a way that a straight line is the most natural, the most obvious, the simplest if you like. That’s all. Ockham’s razor. Choose the simplest theory. But what is the simplest? What seems simplest to me is not necessarily what seems simplest to you. Each person has his own criteria of simplicity. Now with curves, for example, a straight line has only two parameters, so it is simpler than a curve with more parameters. But sine—I don’t know how many parameters, how do you define the number of parameters of sine? And in the lab that’s also how they do it. What? In the lab that’s also how they do it, they move on. Yes, well, that’s obvious. How one gets through labs is another matter. The question is how one really does experiments, not how one reenacts them in the Technion in order to pass the course. Meaning, how Newton worked, not how all those who copied from him worked. He took data from astronomy. Okay, but what did he do? You take the data and try to generate from it the law. But from every set of data you can generate infinitely many laws that are completely different. So how can you know what the right law is? The simplest law. There is some kind of intuition, or whatever, that tells us: this is probably the right generalization. In this case the straight line is the natural thing. If all this falls on a straight line, apparently the straight line is the correct one. Or maybe yes and maybe no. Just as the generalization of three, five, seven seems reasonable to us as nine, or eleven say—those are two natural generalizations. But in fact I can justify any continuation whatsoever. Exactly—you see? It’s exactly the same question. Only here it is discrete and there it is continuous. The same problem. In a series it is discrete, right? What is the next number? Here the question is what curve fits all the points, so it is continuous. But it is the same problem.

In other words, from examples you cannot infer a rule. There is no rule. The rule is only some sort of approximation that seems most reasonable to me. But who knows whether it will work for the next examples or not. But on the other hand, I don’t know what to do without rules. Suppose I measured the relation between force and acceleration, and now they tell me, okay, predict what the acceleration will be if I apply another force that I did not measure. And I have all the measurements. What will I do? How will I infer from those measurements—how will I conclude what happens in the next case? I have no choice but to try to generate a general law from the measurements I have, and from that law try to determine what the acceleration will be for the next force, right? We do not know how to work otherwise. We need to take some law from the examples known to us, and apply it to the next example. That is how we make analogy. Analogy is simply hidden induction. Meaning, you take the examples you have, derive from them some rule that you think is the correct rule, and apply that rule to the next case. That is how one makes analogy. So in short, the excessive trust we place in rules is excessive. Basically we move from one example to another. The rule is a fiction. We use the rule only so that it will help us move from one example to another. But in truth, the truth lies in the examples, not in the rules. And that is the great idea of casuistic thinking. Casuistic thinking understands that rules are not really the right and precise thing. Rules are a temporary aid—use and throw away—like Baron Münchhausen’s ladder. Use the ladder to get out of the pit, and then throw it away. The opposite, he got out of the pit in order to bring a ladder to get out of the pit, yes. But casuistry basically says: the power lies in the cases and not in the rules. The rules are the result of the cases, not the cases the result of the rules. A very important point.

Can one think that this applies also on the side of physical law? No, no. That’s the point. You stretch a spring. When you stretch a spring, Hooke’s law says that the spring—when it springs, Hooke’s law says the force is proportional to the extension. At a certain point it will jump. So you know it’s not amorphous, it’s not something that… I didn’t say it’s amorphous. The law is the best approximation we have to reality, but it is not necessarily exact. It is not necessarily exact. You hope that this is the correct law—maybe yes, maybe no. But I am making a stronger claim. I am saying that even in the legal world, and certainly in the halakhic world, even when we already create a law, we are not supposed to cling to it. You still have to apply common sense and not just go with the law as is. Even after a law has already been created. Because the law is only a stage on the way to analogy. It carries me from these cases to that case. But the law is only the ladder. In the end, in the end, you always have to think: does it make sense?

I’ll give you an example. The Mishnah in the beginning of Bava Kamma says there are four primary categories of damages: the ox, the pit, the grazer, and the fire. And the Mishnah concludes: the common denominator among them is that they tend to cause damage, and they are your property, and their safeguarding is upon you, and when they cause damage, the damager is obligated to pay compensation from the best of his land. The Talmud in Bava Kamma 6 asks: “The common denominator” comes to include what? Meaning, this is an astonishing question if we were not so used to these distortions—distortions in quotation marks. An astonishing question. They tell you four examples, and after that they bring you the rule. Now, for once, the Mishnah did you a favor and didn’t suffice with examples, but also brought you the rule. I would have expected the Talmud to ask: fine, so why do I need the examples? What does it come to include? I have the rule—why do I need the examples? The Talmud not only does not ask why the examples are needed; it asks why the rule is needed, not the examples. I have examples; why do I need the rule? Does that make sense? I would have expected that in all the mishnayot they would bring the rules, not the examples. The examples—I’ll already apply the rules to the specific cases. So one time the Mishnah already does me a favor: besides the examples, it also brings me the rule. The Talmud asks: “The common denominator” comes to include what? In other words, even when they already bring a rule, it is superfluous. I have examples—why are you bringing me the rule? Because the rule does not work.

Think about what that rule says. Every case in which it is your property and its safeguarding is upon you—if it causes damage, the damager is liable to pay compensation from the best of his land. What about a guardian? A guardian who is guarding an animal—it is not his property. Is he obligated to pay if the animal caused damage? Yes. But the rule is that only if it is your property. Yes, because a guardian is like the owner—that’s what the medieval authorities (Rishonim) say. What does “like the owner” mean? You want to force it into the rule that says you incur liability only when you are the owner. A guardian does not fit the rule. Fine—so a guardian is like an owner. But he isn’t. He isn’t an owner. The rule simply does not work. A robber? What about a robber? A robber is also not the owner. Is he liable to pay for an animal in his possession or not? At the beginning of the chapter “Ha-Kones” there is a discussion about that. Again, what about this rule that says “your property, and their safeguarding is upon you”? No—this rule does not really work. It gives some kind of general framework, gives you some general idea, that’s all.

The Talmud in Kiddushin says that every positive commandment bound to time—women are exempt, except for three or four examples given there. The Talmud asks: what about Hakhel? Women are obligated, even though it is time-bound. So the Talmud says: one does not derive from general rules, even where it says “except.” In other words, don’t confuse me with rules. And there is a rule for that. Exactly. This rule has almost no exceptions—except maybe a few exceptions. The Talmud is basically saying this—look, it’s amazing. If it had said every time-bound positive commandment women are exempt, and then you found an exception, I would say fine, okay, they didn’t go into the exceptions, they just stated the principle. But here they said every time-bound positive commandment women are exempt, except A, B—they listed all the exceptions. And that is the most obsessive formulation possible. Now if there is another exception, don’t make a fuss—so there is another exception. It’s bizarre. But if you had just brought me the rule and I found an exception, I would say okay, they didn’t get into the exceptions; they stated the rule and there are some exceptions they didn’t list. But here they listed the exceptions. Even when they listed the exceptions, don’t rely on it—there may be other exceptions they didn’t list. In other words, put the rules aside. Don’t treat them as decisive.

An example in application: the Talmud says that in disputes between Abaye and Rava, the Jewish law follows Rava except for Y-A-L K-G-M, six cases represented by the mnemonic Ya’al Kegam: unconscious despair of lost property, conspiring witness, and so on, a beam standing on its own—six cases. Maimonides rules like Abaye in additional cases besides those six. For example, regarding “do not form separate factions.” There is a dispute between Abaye and Rava, and Maimonides rules that two study halls in one city count as “do not form separate factions”—that is Abaye’s opinion. Rava does not accept that. Regarding “if one did it, it is ineffective,” the passage at the beginning of Temurah—there are interpreters of Maimonides who argue that he ruled like Abaye there too. There it is disputed, but there are those who argue that Maimonides ruled like Abaye. But the Talmud says the Jewish law follows Rava except for Ya’al Kegam. Outside Ya’al Kegam the Jewish law follows Rava. And everybody starts doing pilpulim: maybe the rule in the Talmud applies only to disputes of Abaye and Rava when they disagree in their own names, not when they disagree over the opinions of other Tannaim. Where they invented that rule, nobody knows. It has no basis, no root and branch, but what can you do—they need to find some sub-rule that will explain why Maimonides still fits the rules. No—he does not fit the rules and he does not need to fit the rules. Because the Talmud also has a rule that one does not derive from general rules, even where it says “except.” And that too is a rule in the Talmud. So if the Talmud says, “except”—the Jewish law follows Rava except for Ya’al Kegam—you also do not derive from that rule. There may be additional places where you can rule like Abaye even though it is not Ya’al Kegam. So what is the meaning of the rule? The rule gives you a sort of principled direction. Usually the law will be like Rava, except in those cases where we already know that the law is like Abaye. But if you have a clear position that Abaye is right against Rava, then rule like Abaye in another case too. If you don’t have a position, go with the rule: outside Ya’al Kegam the law follows Rava. And if you have good reasons to rule like Abaye in additional cases, no problem, go with it.

Do you understand what emerges from this? What emerges is that the attitude the Talmud takes toward rules is a very, very—yes—qualified one. Exactly. Don’t overdo the issue of rules. We cannot do without them. We cannot think without rules. Rules are the infrastructure of our thinking. Whenever we think, we work with rules. But one must be careful not to be enslaved to rules. Rules are a good framework; they give us some direction. In the end, in the end, someone who tries to speak with the rules he learned in the ulpan will not get far. He certainly won’t manage to write a poem. Some people might say that’s good. But he won’t manage to write a poem. He won’t really be able to speak in a reasonable way. People will look at him like some kind of robot. He doesn’t know Hebrew. He is using…

I think I once told this: when I was in Yeruham, I taught the first class of the yeshiva in Yeruham when they reached fourth-year study. When they came back from the army, I arrived at the yeshiva. So I got them and taught them there. The first one who got married there—we were at the wedding, and I was standing there with one of the rabbis and looking at the guys dancing. Each one of the guys danced the way he learns. I pointed to him—this one dances the way he learns. Meaning, this one is thinking all the time: the Yemenite step—the guy was Yemenite—so they did a Yemenite step there, and he was thinking which foot it is, you could see the calculations, meaning when you move the hand, when you move—he was dancing in a very mechanical way, precise according to the rules, but he wasn’t really in the passage. Right? He was calculating, and that’s how he learned. It is exactly like that. I can still remember the people. I don’t remember how they danced; I know how they learned, and based on that I also know how they danced. Because it was really one-to-one. And there were those who danced, you know, for whom it just flowed. Clearly they were not doing calculations. And when they learned as well, their intuition told them what is right and what is not right. They did not make these analytical logical structures that help them organize the matter. The way they thought was the way they learned. It was very beautiful—that the way they thought was the way they danced. And I think it is the same thing here. There are people who go more with the rules and rely less on their natural intuition. There are people who go more with their natural intuition and pay less attention to the rules. Obviously, in any case, you need some combination of the two. We cannot do without the rules, and one must be very careful not to cling to them too much. Now what are the proportions? Each person according to how much trust he has in his intuition, how much analytical ability he has—because you need analytical ability in order to use rules. So in short, here there is a range that depends on different people. But in the end, I think the lesson that emerges from the Talmud is that in the end, in the end, you use rules and then throw them away. Meaning, you formulate rules but you do not cling to them.

There is a nice example of this. In the passage of “this one benefits and that one does not lose” in Bava Kamma, the Talmud brings there—it begins like this. There was once a panel at the National Library. There was a series of lectures there on the Talmud, and each time they invited someone else. So one time it was me with Meir Buzaglo, a professor from the Hebrew University. We did a panel, and Tomer Persico was there, and who else? I don’t remember, someone else, doesn’t matter. So I suggested that we deal with the passage of “this one benefits and that one does not lose” in Bava Kamma. The passage begins like this: if someone lives in his fellow’s courtyard without his knowledge, does he need to pay him rent or does he not need to pay him rent? Right? Someone enters his fellow’s courtyard and uses it—does he need to pay him? And the Talmud begins to discuss it. It depends: is the courtyard normally for rent or not? Is the person someone who needs to rent or not? Meaning, does he need lodging? Did he save money by living there because otherwise he would have paid for lodging elsewhere? Or does he not need lodging—he has a house, he just entered there in order to be there; he doesn’t really need it. Is it a person who normally rents or not? Meaning, the question is whether he benefited. Then there is the question whether the courtyard is normally for rent—because if not, then if I used the courtyard, what did you lose? In any event you did not plan to rent it out, so you lost nothing. So the question is whether you suffered a loss or not. Then the Talmud reaches the conclusion that the question is basically: in a case where one benefits and the other does not lose, is he liable or exempt? If one benefits and the other loses, he is liable. If one does not benefit and the other does not lose, he is exempt. And the question concerns the case where one benefits and the other does not lose.

Now what is that transition? The transition is that we began with a case—someone entered his fellow’s courtyard; the question is whether he has to pay—and in the end we arrived at a formulation that is the formulation of a rule. What is the law in an entire category of cases where one benefits and the other does not lose—must one pay or not? The case of someone living in his fellow’s courtyard without his knowledge is just a specific instance. So the move there in the passage began with a case and moved to a general formulation in a general way. But afterwards, when you look further in the Talmud and also in the medieval and later authorities, in the halakhic decisors, who begin to apply the rule of “this one benefits and that one does not lose,” you would not believe how many distinctions there are. It is impossible really to apply that rule. “This one benefits and that one does not lose is liable”—okay, but the question is what it means that he does not lose. Not lose in this way, it saves him expense, warding off a lion, all kinds of sub-rules that they of course generate there in order to explain why the rule is not relevant here and not relevant there. It is impossible really to apply that rule. Even after the Talmud has already formulated the question not as a case but as a rule—you now want to say, okay, now we have a rule: whenever you benefit and the other one does not lose, you have to pay. So let’s see—here we have such a case. Oops, it turns out not—in fact you are exempt there. How does that fit with the rule of “this one benefits and that one does not lose”? Ah, okay, here it is something else, because he doesn’t lose in such-and-such a way and that one doesn’t benefit in such-and-such a way, I don’t know—you start drawing distinctions. Reminds you of the Kehillot Yaakov. Same thing. You formulated a rule, and that rule hardly works for any case. But the rule helped you think about the passage. After you have thought about the passage, put the rule aside, throw it aside. And this is a rule that the Talmud itself formulated.

The Talmud itself usually brings only the case: someone living in his fellow’s courtyard without his knowledge—does he need to pay rent or not? The conclusion will be yes or no, doesn’t matter—that will be the conclusion. Now I, as a later learner—the medieval authorities (Rishonim), the later authorities (Acharonim), and we as learners of the Talmud—we try to formulate it in the form of rules. Ah, this is a case of “this one benefits and that one does not lose,” and in the Talmud it is basically written that “this one benefits and that one does not lose” is liable. Okay, so those are the rules we make. Here the Talmud itself made the rule, saved us the trouble. It made a move of the later authorities, the Talmud did. The moment the Talmud starts formulating rules, it is basically working like later authorities. Usually that doesn’t happen; the Talmud speaks about cases. The commentators make the rules. But sometimes the Talmud itself formulates the rules. And even there, we do not really treat those rules as binding, we do not cling to them all the way. We make use of them, but we do not cling to them all the way. Because rules are something that must be treated with respect and suspicion. It is a very dangerous business to rely on rules.

By contrast, the thing that seems to us more dangerous—to rely on precedents, who knows whether it really resembles that or not, analogy is some kind of loose thing, not unambiguous—that is much more reliable and much stronger than using rules. It gets you much better to the intention of the one who taught you or the one who legislated. You will hit the mark better if you use analogies than if you use deductions. It is an illusion to think that positivism is possible.

By the way, one of the big crises that positivism experienced was the Nuremberg trials, after the Holocaust. One of the well-known defenses of the accused, of the Nazis, was that they were following orders. That’s what the law in Germany established, and they were simply obeying the law. What do you want from me? I obeyed the law. Blame the legislator, or whatever. How do you even accuse a legislator, by the way? According to what law do you judge the legislator? International law? There is no such thing. Fine, so then this isn’t a trial. So what do you do with such a defense claim? That defense claim is basically a classic German defense claim. Because German law is positivist. And from the German point of view, that’s their character. There are rules. They called it the science of law. For them law was a science. It was a set of rules that you apply, like mathematics, like the natural sciences, an exact science. Nonsense, of course, but that was the illusion in which they lived. And that is what it gave rise to. What it gave rise to was: okay, there is a law, it has to be obeyed—that is positivism. So the law there said to do that; that’s what we did.

And the judges there, or the law according to which they criticized them or prosecuted them, was basically natural law, which is the antithesis of positivism. Which basically says: true, that was the law, but you should not have obeyed the law. You should not have obeyed the law. A law like that must not be obeyed. Why not? According to what must it not be obeyed? “A black flag”—that’s the figurative expression, but in practice they are telling you: do not obey the law. That is the Israeli figurative expression; at Nuremberg it was there earlier. What does that actually mean? It basically means that what they prosecuted the Nazis for was being positivists. That if the law determined it, then for them that was what had to be done. What do you want? I am subject to the law; I am a citizen. How can you prosecute a citizen for obeying the law? That is absurd. According to what law are you prosecuting him? That is the law to which he is obligated, German law. That was the law. What do you want? It is not a simple conceptual problem. According to what are you prosecuting him? I am prosecuting him according to the principles or the spirit that the law was supposed to express and did not express. You should have disobeyed the law.

If a woman does not fulfill the commandment of Hakhel because it is a time-bound positive commandment and women are exempt—what do you want? She should have understood that despite the fact that women are exempt from time-bound positive commandments, regarding Hakhel she is supposed to fulfill it. It is a command marked by a black flag, telling the woman not to come to Hakhel—even though according to the rule women are exempt from time-bound positive commandments. No, that’s the rule. But the law is not exhausted by the rules. There are situations where it should be clear to you that one does not obey the law. Passover commandment. What? That a woman is obligated. There they already created the sub-rule: whoever is obligated in the prohibition is obligated in the positive commandment. But that too is positivistic. What? So at Nuremberg they said, okay, we need to legislate a system of rules, international law. Right, that is usually what they do. After you make the casuistic or anti-positivist breakthrough, the positivists come and formulate rules. But those rules too will be breached in the same way. Therefore it is fine—we always use rules. I am not against rules; I am against clinging to them. After we make that breakthrough, from that breakthrough we will indeed formulate new rules, exactly as the Kehillot Yaakov did about divorce. And then you attack those too, and slowly understand what is right. You use rules, but do not cling to them. Do not treat rules as if they are the thing itself. The rules are the closest model you have, but they are not the thing itself. That is exactly the point. We always work with rules; you cannot work without rules, but treat them with respect and suspicion.

There is the passage in Nazir about a transgression for the sake of Heaven. What is a transgression for the sake of Heaven? Yael, the wife of Heber the Kenite. That is the example the Talmud brings there. She had sexual relations with Sisera; she was a married woman. Sexual immorality is not set aside for anything. Right? It is one of the three most severe transgressions. So why is she praised for committing sexual immorality in order to kill Sisera? That is a transgression for the sake of Heaven. What is a transgression for the sake of Heaven? The medieval and later authorities get very frightened by this concept, a transgression for the sake of Heaven. It means that you are actually doing a transgression—so it is permitted to do a transgression? That cannot be. So they turn it into some kind of rule of override, like a positive commandment overriding a prohibition. There are override rules where halakhic principles collide, where one prevails over another. So a transgression for the sake of Heaven too seems—at least in many medieval and later authorities, more among the later ones I think, who are more hysterical—like some sort of override rule. And clearly that is not correct. A transgression for the sake of Heaven means: this is a transgression. It has no halakhic justification. But it is what she needed to do. It is what she needed to do even though it is a halakhic transgression—there is no justification. Right? And according to the rules it is forbidden to do this. But when you are in that situation, you needed to understand that here the rules must be broken. And with commands marked by a black flag, not only is it justified if you break the rule—if you do not break the rule, they will prosecute you. That is what happened in the Nuremberg trials. Here they tell you it is justified to break the rule, but at Nuremberg they said something stronger. If you did not break the rule, then we will prosecute you. Not only that if you did break it we forgive you—no. If you did not break it, they will prosecute you. You were obligated to break it, not merely permitted to break it. You were obligated to break the rule because you should have understood that the rule was not right.

Same thing with commands marked by a black flag, as in Qibya. The Qibya operation—the same thing. That was Landau there, right? Or Zilberg, who said that it was an order over which a black flag flew, when they shot there in Qibya at people who broke the curfew there. Kafr Qasim or Qibya—I don’t remember. Maybe Kafr Qasim. The same thing with the Japanese. The Japanese too would be judged according to their law, according to the Japanese order. They would say: it was an order. Okay, fine, the Japanese are Yekkes, we know. It is the same mentality, the same mentality. It is a positivist culture. That is the problem: a positivist culture. That is why everything works so perfectly in Japan, by the way. Everything works perfectly because everything is by the rules. Everything on time, everything to the second, everything according to schedule—like Germans. It is the same mentality. But that is also their bug. Because there are problems with creativity in societies of that sort, for example. Okay, that is another discussion.

The rule that there is also a rule that there is an exception to every rule—and one person will say this is a black flag and another will say otherwise. Of course, right, there is no choice. You cannot fit everything into rules. In the end, you have to leave it to a person to make decisions. Don’t fool yourself into thinking that you can dictate to him the correct answer with rules.

By the way, one of the big concerns about autonomous cars—today I’m drifting all over the place—autonomous cars are exactly this. That there is no one to blame. No, not that there is no one to blame—that is the result. First of all there is the fear that they won’t drive correctly because they were not programmed correctly. What about human beings? What—do human beings always drive correctly? No, but with cars the fear is that you live under the illusion—by the way, those who are afraid are mainly people who do not understand this stuff. The problem is that unfortunately those are usually the legislators. Otherwise there should long since have been autonomous cars. And the fear comes from the fact that people conceive of the autonomous car as some kind of robot operating according to rules, and who knows, it might suddenly kill a hundred people because some rule there doesn’t quite work. The assumption is that human beings don’t work according to rules, but maneuver; they understand when it is right and when it is not right, and they will act properly. Those people do not understand that an autonomous car also does not work according to rules. It is AI. It does not work according to rules. And therefore, obviously, long ago—not just today, long ago—an autonomous car would have saved very many human lives had they activated it. Right, there are mistakes. Every time an autonomous car makes a mistake, it makes headlines, everybody goes hysterical, “must not approve.” Human beings kill human beings by the hundreds every day around the world. Hundreds every day. None of that makes headlines; it doesn’t bother anybody. An autonomous car kills one person once every five years—this is a murderous car, yes, like a Sanhedrin that executes once in seventy years. A murderous car—they won’t let it get a driving permit. Why not? Human beings—the alternative—is human beings, and human beings kill far more. But there are phobias. There are phobias; people have phobias, that’s all. That is the only thing stopping this, in my opinion.

The same thing with autonomous weapons. Why does the army today call—you know, today they don’t call drones drones. Hezbollah has drones. We have remotely piloted aircraft. Do you know the difference between an unmanned aircraft and a remotely piloted aircraft? An unmanned aircraft is an unmanned aerial vehicle. Heaven forbid, an unmanned aerial vehicle. A remotely piloted aircraft means: what do you mean? You can’t leave the machine to make decisions. There has to be a human pilot. Like an autonomous car. So the linguistic laundering today says: no, no, of course, there is a human there, it’s just remotely piloted—the human on the ground directing this aircraft, he is the pilot, so don’t say there are no human beings here. There are no human beings here, but there is a pilot on the ground, so everything is fine. It is not an unmanned aircraft, it is a remotely piloted aircraft. And in the IDF today, that is the name. You may not say drone, only remotely piloted aircraft. Only Hezbollah has drones, because Hezbollah are evil, that much we know. But the IDF are righteous, the most moral army in the world, so with us there are no drones, only remotely piloted aircraft. And again, this is the same phobia of autonomous, automatic decision-making by machines. It is a trust in human decision-making that has no basis. Machines today make decisions much better than human beings in many fields. Not in all fields—in many fields. And they are not allowed to act because there is tremendous fear. Every time a machine makes some wrong decision, everybody starts trembling. It’s a phobia of technology, that’s all, nothing more than that.

What? Yes, but even there it’s moving much more slowly—without our knowing—but I’m telling you it is moving much more slowly than it should have, because of the fears and the regulation and everything. Again, there is logic in having regulation and checking things, but the slowness of the regulation is costing us a great many human lives. Because human beings operate less well than machines, and for machines to be approved to replace human beings they have to reach a level where there is no error at all—when really it should be enough that they make fewer errors than human beings in order to approve them. They do not have to be perfect. You know, it is like the famous joke about two people walking in the forest and a tiger is chasing them. Fine? At what speed do I have to run in order to be saved? Faster than my friend. Right? If I run faster than my friend, the tiger will eat him. I do not need to run faster than the tiger. Same thing here. To put the machine into operation, it does not need to be perfect. It needs to be better than the alternative, which is human beings. That’s all. If it is better than human beings, put it in—you are saving human lives. But no, we have phobias. Yet those phobias are not justified. Those phobias might have been justified if the machine were classically programmed, because then it would have been a positivist machine, a machine operating according to rules—and rules really do miss things. Someone who does not work with rules makes better decisions than someone who does work with rules. But once we arrived at AI, this is a machine that does not work with rules, so it has the advantages of a machine—speed and processing power and all that—and also none of the disadvantages of a machine, because it doesn’t work with rules. So it is simply perfect. Yes, this should be put everywhere possible.

What? But the Chinese are not waiting for our regulations; the Chinese do whatever they want. Regulations are only on us; that’s always how it is. Regulations are… What? Yes, but don’t approve taking it from the Chinese—that I understand—but why not develop your own? Ah, you’re saying the Chinese will attack by cyber. Even after all those cyberattacks, it will still be more efficient than human beings. Human beings don’t need cyberattacks in order to make mistakes. Okay, I’ve really drifted a bit here. There is some issue here with some study that shows the psychological problem is trust—the point is that that scientist says that someone who trusts the machine after two or three times, after a month he will already trust the machine one hundred percent, and even if they tell you to keep your hands on the wheel, you take out your phone, and then they see that after a month or two… Excellent—trust the machine, wonderful! No, the problem is that they trust the machine too much. So what’s the problem? Trust the machine too much; it will still do it better than you. Why do I care that there is dependence? The fact that people trust the machine too much means they won’t hold the wheel. Right! So let them not hold it. Again, that is exactly the same phobia—let them not hold it; they don’t need to hold it. The machines already today do this better than I do without my holding it. Long ago—not only today, for years—much better than human beings. Autonomous cars—there have been experiments on this.

Okay, I really drifted a bit here into the matter of… Fine, I’m not going to start the next topic now, so let’s stop here. This interview—what do you think about the concept of “Torah perspective”? Meaning, Maimonides says, I—this is the assumption. That is the assumption of the concept of Torah perspective. The assumption is that after he has internalized enough passages and cases and so on, he has acquired a certain kind of mind also for fields in which he is not necessarily an expert. Now, I don’t buy that, because I don’t think there is any connection at all. Meaning, I do accept the concept of Torah perspective with respect to halakhic questions. There, in my opinion, a Torah scholar has an advantage over someone who is not a Torah scholar but studied the passage well. But on other matters, I don’t accept it. I don’t think it is the same neural network. It is not the same kind of ability. If you want to look at the world from the perspective of what is right from the standpoint of the Torah—that, I did internalize, I embedded it in my head, and I will look at things through that lens. And we said: if you’re speaking about halakhic questions, I agree. If you’re speaking about questions not connected to Jewish law, then what does “from the standpoint of the Torah” even mean? Should one invest in the stock market here or there—that’s what I mean. If you’re speaking about halakhic questions, as I said, yes, a Torah scholar has an advantage. No, but practical daily matters that have the same thing in them. On the Torah aspect, he has an advantage. But there are also other aspects there, and then you have to see how you weigh them.

There are those who attack it from another angle and say no, he has to prove it to me, he should explain to me with proofs from Torah why he… That I don’t buy. There is a well-known saying in yeshivot: when the rosh yeshiva brings a proof for his words, you may be able to find a flaw in the proof or a counter-proof. But if the rosh yeshiva says “it seems to me,” there is no point arguing with him—he is surely right. And that is exactly the same thing. It is his neural network. Forget the—if he only has proofs, he doesn’t actually feel that it is so. He has proofs. Proofs I also know how to bring. But if the entire rosh yeshiva stands here on the scales and says, listen, I’m telling you this is so, this is my intuition—then that is built on all his halakhic experience. It is not something local. So there you are probably mistaken.