Artificial Intelligence – Lesson 17 – Rabbi Michael Abraham

This is an English translation (via GPT-5.4). Read the original Hebrew version.

This transcript was produced automatically using artificial intelligence. There may be inaccuracies in the transcribed content and in speaker identification.

🔗 Link to the original lesson

🔗 Link to the transcript on Sofer.AI

Opening: from a “bare” language model to a system with tools and agents — it was explained that current models rely on calculators, tools, and agents in order to compensate for computational weaknesses and carry out tasks efficiently.
The central question: to what extent can a neural network be regarded as a person — the distinction was drawn between a classical computer as an electrical circuit with no cognition, and a language model that is experienced as a human conversational partner.
The Turing test and apparently superhuman achievements — it was argued that a language model can appear human, and even “superhuman,” because it produces answers and knowledge at a very high level across many fields.
The training of neural networks and the analogy to the Chinese room — the similarity was presented between supervised learning and a person trained to respond correctly without understanding, by means of feedback and updating “weights.”
Syntax without semantics: the imitation can be perfect and still remain imitation — it was emphasized that the claim is not quantitative; even perfect performance does not prove the existence of understanding or consciousness.
The debate over the mental dimension, intellect versus brain, and emotion versus rationality — the Rabbi distinguished between physical processes in the brain and intellect and consciousness, and argued that AI lacks a genuine mental dimension.
Two differences between a computer and a human being — these were defined as follows: in a classical computer, the programmer solves the problem, and in addition the computer has no mental accompaniment to its syntactic operations; in a language model only the second difference remains.
The language model as solving problems on its own — it was emphasized that unlike an ordinary computer, no dedicated program was fed in for each problem; the model generates answers to many problems, including ones its builders themselves do not know how to solve.
Reservation regarding the claim that “the human solved it through the training” — the comparison between training a model and explicit programming was rejected, and was compared in part to a student learning from a teacher and to evolution in relation to a human being.
Contemporary examples: mathematical proofs, AlphaFold, and the “black box” — AI’s achievements on difficult problems were discussed, along with the difficulty of understanding how it arrives at its answers even when the output can be checked.
DeepSeek and Chain of Thought — the innovation of models that also produce intermediate stages of thinking was presented, along with the fact that such training improves answer quality and heightens the sense of humanness.
Why Chain of Thought is not proof of human thinking — it was explained that the model was explicitly trained to return a chain of thought, and therefore it imitates human reasoning as well, without necessarily thinking that way internally.
The attempt to “whiten the black box” — a research effort was described that aims to identify internal circuits and subtasks in the network, together with the reservation that its mode of operation may be fundamentally different from that of a human being.
The analogy to evolution and “to be or not to be” — the Rabbi used Dawkins’s example to argue that filters directing a random process indicate a guiding hand, similar to training that directs AI.
Conclusion and drift into further questions — AI dangers, the status of intellectuals, fundamentalism, and didactic questions about studying Talmud were briefly discussed, but these appeared as side extensions at the end of the lesson.

Summary

General Overview

The lesson continued the technical discussion of language models, but shifted from the question “How does it work?” to the philosophical question: what does it mean that a machine answers like a human being, solves problems, and at times even seems superhuman. Rabbi Michael Abraham seeks to argue that this is indeed a real leap forward relative to the classical computer, but that it still does not make the language model a being with consciousness, understanding, or a mental dimension.

## From a Classical Computer to a Language Model
In an ordinary computer, the programmer is the one who solves the problem: he writes an algorithm, and the computer merely runs it. So it is clear that the computer does not “think,” but only performs mechanical operations. By contrast, a language model does not receive a dedicated program for each problem. It is trained in a general way on enormous quantities of text or examples, and afterward is able to answer many varied questions, including ones its own builders do not know how to solve. That is the first and innovative difference: here it seems that the machine itself solves problems.

At the same time, the Rabbi emphasizes that the second difference remains fully intact: even if the model solves problems, there is no reason to attribute a mental dimension to it. It performs syntax without semantics, meaning it produces correct outputs without genuine understanding. This is the heart of the analogy to Searle’s Chinese room: one can train a system to give highly relevant answers, and still argue that it has no understanding.

## Perfect Imitation Is Not Understanding
A central point in the lesson is that the debate is not quantitative but essential. Even if we assume that a language model will solve every problem, write every novel insight, pass every Turing test, and function better than any human being — still, in the Rabbi’s view, it may be only a perfect imitator. The question is not how well it performs, but whether its actions are accompanied by experience, perception, understanding, or what the Rabbi calls a “mental dimension.”

Therefore, in his view, when AI gives an answer, it is not really “answering”; it is moving electrons, and the user is the one who interprets the output as an answer. Just as water can “solve” physical equations without thinking, so too a language model can produce impressive results without awareness.

## Chain of Thought and the Illusion of Human Thinking
A major part of the lesson was devoted to models such as DeepSeek and the Chain of Thought method. These models do not merely return a final answer, but also present intermediate stages: breaking down the problem, considerations, solution steps, and a summary. As a matter of fact, such models also achieve better results.

However, the Rabbi argues that this is not proof of human thinking. The reason is that the model was explicitly trained to return the chain of thought as well, and not only the answer. If so, no “miracle” occurred in which the model discovered a human mode of thought on its own; it was simply trained to imitate the reasoning too. Therefore, philosophically, this is the same basic mechanism of imitation, only more detailed.

## The Black Box and the Attempt to Understand What Happens Inside
The difficulty of understanding how the model actually works was also discussed. Even experts do not really know how to explain how these abilities emerge from the training. There is now a research effort to “whiten the black box” — to identify internal circuits and subtasks in the network. But here too the Rabbi adds a reservation: it may be that the machine does not “think” in human patterns at all, and therefore the very attempt to map it according to human categories may be misleading.

## The Analogy to Evolution
Toward the end, the Rabbi compared this to the debate surrounding evolution. Just as in the example of “to be or not to be,” introducing filters and guidance into the course of a random process greatly increases the chance of reaching the desired result, so too training on Chain of Thought increases the chance of a better answer. But it does not follow from that that the system itself understands; only that guidance was imposed on it from outside. Hence his conclusion: even the progress to the new models does not fundamentally change the discussion. It is still, in his view, an imitating golem — not a thinking human being.

Full Transcript

[Rabbi Michael Abraham] Last time I more or less finished describing models… a neural network, and after that a language model. We spoke a bit about the fact that today’s language models are no longer just a giant neural network; they are also accompanied by various tools that they use, which compensate for weaknesses language models have. There is a certain kind of task that is either hard for a language model to do, or requires a lot of computation, and it basically hands those tasks off to particular tools. Right, I talked about adding numbers, for example. A language model has a very hard time adding numbers. It’s strange — it solves crazy problems, but it has trouble adding numbers. Well, it also doesn’t solve crazy problems, only after they have already given it tools and built agents and all the surrounding things. But when it sees — when it identifies that it has received a question about adding numbers, it simply sends it to a calculator, and the calculator gives the answer. And the contribution of the language model is only to identify that this is the kind of question being asked here. From that point on, it simply sends it to the relevant device that will solve the problem. And working through tools, tools, and agents — yes, agents — and all kinds of things like that, skills and so on, is actually a large part of the abilities we know today in all current models. In any case, up to here I have more or less described the principles of what a language model is, how it is built, and so on. And what I want to do now is start to see a bit what this thing actually means. So maybe I’ll begin by going back to the original question that has accompanied us from the start: to what extent can we relate to this neural network as a person? We spoke earlier about Anthropic, so that is basically the question. Now, what exactly is new in the neural network relative to the other things? Right, I remind you that when we discussed an ordinary computer, an analog computer, and then a digital computer, I tried to show that we are essentially dealing with an electrical circuit, and the electrical circuit somehow operates in a way that parallels the solution of some problem, and therefore no one there even imagines that it has cognition and mental phenomena and all kinds of things of that sort. It is obvious to everyone that it is a golem moving electrons. What happens in a language model? In a neural network, and especially a language model. So first of all, what happens is that before us stands a machine that passes Turing tests, really talks to us, gives the feeling that you are talking to a human being. Leave aside the nuances that by now we are somewhat used to identifying there, but in principle, someone who did not know that language models exist in the world and spoke with a language model would never imagine that it was not a human being. It is literally like talking with a human being. Not only is it like talking with a human being, it has developed abilities that help it solve problems from every field, provide knowledge in every field at the level of a PhD and above. In other words, we are really dealing here with a superhuman being. It is not a human being, it is already superhuman. And in that sense, the neural network really is different from the computers of previous generations. And we already saw that we reach these abilities of the neural network through a process of training. Right, we basically feed into the network — at first, when we discussed a neural network, that meant feeding into the network a great many examples of problems with correct answers and incorrect answers. Let’s say that is supervised training, controlled training. And somehow this organizes its internal weights in such a way that it knows how to solve problems of that type. In a language model, that means training it on texts. I talked about that last time. And that creates in it an ability to solve problems of all kinds and in all fields, across a very broad range. And therefore the language model really does begin to resemble us more, like a human being. Now, the training of this network is actually parallel — and this is a very important point; that is also why in the introduction I gave the example of the Chinese room — completely parallel to what happened in the Chinese room. In the Chinese room, what we did with the person sitting there in that room was basically to train him by supervised learning. That is, we gave him an example, a question, he gave an answer, and if the answer was not correct — he got a whack, right? That is the feedback for an incorrect answer. He updated his weights, and then we continued. Again he gets a question, gives an incorrect answer, again a whack, until slowly, after an infinite amount of time, he begins to give correct answers, or relevant answers, to the questions. And that is really parallel to the training of a neural network. In that sense John Searle was definitely ahead of his time. He wrote that before people were talking about neural networks, and certainly before people were talking about language models — that goes without saying. So he really was ahead of his time. But on the other hand, his example really came to show the difference between interaction accompanied by understanding and interaction that involves syntax without semantics. In other words, the person in the Chinese room does not really understand Chinese, even though he gives a relevant answer to every question he receives. That is different from a Chinese speaker, who outwardly supposedly functions in the same way. To the questions he receives, he will give more or less the same answers, or answers of the same kind. And still, of the Chinese speaker we would say that he understands Chinese, while of the person in the Chinese room we presumably would not say that he understands Chinese. At least in Searle’s time it was obvious that he did not understand Chinese. Today we already have all the clever people, with language models, who have started saying that maybe he does understand Chinese. But intuitively it seems quite clear to me that this is not what it means to understand Chinese. So that means that we can in fact produce an imitation of human conversation, and what is missing from that imitation is the accompaniment of mental events. In other words, the questions receive the same answers, but this is not accompanied by a mental process of selecting the correct answers. In that sense it is similar to what I spoke about quite a bit in the introduction, about the thinking done by water — that water solves the Navier-Stokes equations. So water solves the Navier-Stokes equations, but it is clear that there are no mental events accompanying that solution, and therefore one cannot say that the water is thinking. In that sense, I think the language model too is basically — in that sense it is like water. In other words, it gives us a solution to questions, but this is not accompanied by mental processes. It is some sort of imitation of what a human being does, of the response I would expect a human being to give, but it is imitation. Even though the imitation — just now someone asked me on the site — this imitation can in principle be perfect. In other words, it could be that you would not catch it in any bug, in any kind of problem. It would know how to do everything like a human being, and even better than a human being, and it would still be imitation. In other words, we are not speaking here about a quantitative comparison. This is an important point that we will get to in more detail later on. But one has to understand: the claim is worthy, the claim is worthy — just a moment — the claim is not a quantitative claim. In other words, it is not that the human being knows how to solve more problems than the language model. That is also not true. But even if it were true, that is not the point. Even if the language model were a perfect human being or a superhuman being who knows how to do everything, I would still make the same claim, because there is no mental dimension accompanying its actions. It is still imitating. It is imitating a superhuman being, fine, but it is not a superhuman being. It is only imitating a superhuman being. What is a mental dimension? What?

[Speaker B] What is a mental dimension?

[Rabbi Michael Abraham] A mental dimension — everyone understands what a mental dimension is. Why does that need explaining? Everyone understands. I don’t.

[Speaker B] No, if — if — if the direction is rational thinking, rational thinking like weighing data and drawing conclusions, making connections between things, even for non-rational people, so making connections between different things, even for people—

[Rabbi Michael Abraham] Even non-rational people have mental events.

[Speaker B] So it’s not related — no, no—

[Rabbi Michael Abraham] No, it is not rationality. It is processes of thought, whether rational or not.

[Speaker B] No, but no, no, Rabbi, by rational I mean carrying out logical thinking and all the considerations that we engage in from a rational standpoint.

[Rabbi Michael Abraham] No, but even a person who talks nonsense, even a person who talks nonsense—

[Speaker B] Has a process—

[Rabbi Michael Abraham] Of thinking, only stupid thinking.

[Speaker B] Before he — before he speaks. No, it may be mistaken, but he is using rational thought. I am speaking about rational thought, not emotions. After all, Rabbi…

[Rabbi Michael Abraham] No, so the concept of rational here — no, I think you are not using it correctly, but never mind, it is just terminology, okay.

[Speaker B] Rational thought that is not connected to emotions but rather to the analysis of data and the drawing of logical conclusions, whether correct or incorrect — AI does that completely, so therefore I ask what… No, it does not do that.

[Rabbi Michael Abraham] Why not? It does it even in an empty way, not only in a complete way.

[Speaker B] Why? For what reason?

[Rabbi Michael Abraham] Because it does not do that. It answers answers exactly the way the person in the Chinese room—

[Speaker B] In the Chinese room does not do that. Why? In our brain we have a neural network; we made a copy-paste.

[Rabbi Michael Abraham] The material that mediates this in us is our brain. Our brain is a neural network, but—

[Speaker B] We are that—

[Rabbi Michael Abraham] Not the brain. We are the intellect.

[Speaker B] Right, there is… I am trying, I am trying to understand what is missing here in terms of rational thought, if we neutralize the emotions. What is missing, what is missing is—

[Rabbi Michael Abraham] The intellect, not the brain.

[Speaker B] The brain is a neural network. But intellect is intellect. If I — I say, I am going in the Rabbi’s direction, that rational thinking is what matters, to neutralize the issue of emotions. I agree that AI has no emotions at all. But if we neutralize the issue of emotions, and focus only on intellect, on rationality, AI has everything, because we simply made an ideal copy-paste, a copy-paste of all the neural networks of humanity, and put them into… You are mistaken, that is not true.

[Rabbi Michael Abraham] We did not make a copy-paste. What we have here is imitation. When a parrot imitates the way you speak, do you think the parrot speaks like you? Does it have mental events in the background? Is it thinking about what it says?

[Speaker B] No, but it does not have my neural network. But if I put into the parrot’s brain—

[Rabbi Michael Abraham] The parrot my neural network with my weights… No, that’s not it, that’s not it! It has the neural network that produces the answers it gives. And let’s say it spoke exactly like you — so what? It is still imitation.

[Speaker B] But what is there in my brain that it has less of?

[Rabbi Michael Abraham] Not in the brain — in the intellect!

[Speaker B] But… I am speaking again about the rational aspect. If I… I am speaking about the rational aspect!

[Rabbi Michael Abraham] The rational aspect takes place in the intellect, not in the brain.

[Speaker B] The intellect thinks by means of the brain, just as we walk—

[Rabbi Michael Abraham] By means of our legs.

[Speaker B] Let me ask the Rabbi like this: does the Rabbi admit that we have a neural network in the brain? Yes. And that it has different weights? The Rabbi explained that brilliantly. Okay. If we neutralize the issue of emotions, ignore it as if it does not exist… I neutralized the issue of emotions; remove emotions from here. Good, excellent, excellent. So we relate to… a person whom we succeeded… to whom I gave an anti-emotion pill. He no longer has emotions. He only performs rational calculations. Okay. How does he do them? By means of the neural network and the weights found in it. Does he have that? Yes, and he does the weighting — wonderful. The computer — I copied all the weights and the whole neural network into a computer, theoretically, in the future, another hundred years, it will be there, and then it will do exactly the same thing.

[Rabbi Michael Abraham] No, it will not do the same thing.

[Speaker B] It will not have emotions; what will be missing?

[Rabbi Michael Abraham] It will not—

[Speaker B] Do the same thing.

[Rabbi Michael Abraham] What is missing? What is missing? So I am saying: what is missing is the mental processes that accompany this calculation. The Rabbi is talking about mysticism!

[Speaker B] The Rabbi is talking about mysticism! No, I am not talking about mysticism!

[Rabbi Michael Abraham] But what, what is missing? What is missing? Only emotions are missing? The answer is that only emotions are missing. When you look at the physical world, nothing is missing. But that is exactly your problem — that you are looking at the physical world. I claim that there are mental processes, not only in the world… leave aside the physical world. We said that dualism is not necessary for this discussion. Even without dualism, let us say that a human being is only material, okay? Still… let us say that a human being is only material. Still there is emergence, and mental processes are created as a result of the brain.

[Speaker B] Emotions only,

[Rabbi Michael Abraham] Sensations and emotions, nothing else. No, not emotions, not emotions! When you think about something, you think, you… you process in your brain, you say two plus two equals four. That feeling, that consciousness, that perception — that does not exist in AI.

[Speaker B] I agree with the Rabbi, I agree with the Rabbi, but the conclusion from that is that logic too is ultimately an emotion and not something that exists in itself. And we argued about this in the past, Rabbi, but I… this is proof. The Rabbi is really proving what… what I argued.

[Rabbi Michael Abraham] We are not going to return to that argument again, okay.

[Speaker E] May I ask a question too, or is that just… this… I remember that in the introductions the Rabbi, as he just said, said that this discussion does not presuppose dualism. But when you say that the difference here is not quantitative but essential between the human mode of thought — of the human brain, let’s say, or the accompanying intellect — and the mode of artificial intelligence, does that not presuppose an essential difference that only dualism can explain?

[Rabbi Michael Abraham] Does the fact that we also have a mental dimension require a gap in performance? We will talk about that later. At the moment I am not assuming at all that there is a performance gap, okay?

[Speaker E] No, you said that there is a difference, that the difference in the mode of thought… maybe I did not hear correctly, but what I heard was that there is an essential difference and not a quantitative one. That is, even if you now add tens of billions and—

[Rabbi Michael Abraham] I said that on the level of principle, my claim stands even if in principle we manage to produce a perfect imitator, meaning that there will be no quantitative difference between the problems I know how to solve and the problems it knows how to solve. I would still make my claim. Is there really such similarity in achievements, or is it possible that there could really be a situation in which everything I know how to do, it too will know how to do, even though it is only an imitator without a mental dimension? That is a question we will discuss later. But for now, for the purpose of our discussion, I am not talking about the quantitative difference. The assumption is: let us say that we carry out the same tasks, know how to carry out the same tasks.

[Speaker E] Okay, I understand.

[Speaker F] No, but there is some dimension of plasticity of thought when a person talks with a person, right? For example, I sent the Rabbi a question — my daughter asked ChatGPT about the existence of God, and then a discussion developed there, something unbelievable, and it continued, the Rabbi didn’t… I did not send the whole thing. In short, ChatGPT became a devoted follower of Leibniz in the strongest possible way. Meaning, there is, there is something in… you feel that in its answers, in the computation the chat gives you, right? You — it is not a human being, it is…

[Rabbi Michael Abraham] You are going back to… you are going back to the question of judgment. We talked about it in the introduction, and we will return to it. It is connected to… to what we have just been discussing, that is, to the differences that… the question is whether there must also be differences in performance, and not only in the question of the existence of a mental dimension

Mental. Okay, but let’s leave that for now; I’ll get back to it later. Now, what I actually want is to put this discussion a bit more precisely into the pattern that has accompanied us until now. We saw that the distinguishing feature between a computer and a human being, or what separates a computer from a human being, has two main characteristics that I suggested. One characteristic is that the one who solved the problem was not the computer but the human being, the programmer. Okay? The computer does not solve problems; the computer runs programs that are fed into it. The one who solved the problem is really the programmer who wrote the program, or whoever built the computer, it doesn’t matter, okay? And therefore that is one difference. In other words, one difference is that the one who solves the problem is the programmer and not the computer. The second characteristic is the one I just spoke about: in a computer there is no mental dimension that accompanies the syntactic actions, yes, the mechanical actions the computer performs. And in a human being there is some kind of thought that accompanies these technical, physiological actions, okay? Those are the two characteristics.

Now, with a computer, what happens in a language model? In a language model, even in a neural network, but in a language model it is much more prominent, the second characteristic, the second distinction, exists here too. At least in my opinion. I obviously have no proof, but at least to me it is fairly clear that the language model has no mental dimension. It performs a purely syntactic action, without semantics, okay? But the first characteristic has fallen away. The first distinction no longer exists. That is an important point, because with an ordinary computer we said that the one who solved the problem was the programmer. And that is unlike a human being, who solves the problem. In a computer, the computer does not solve problems. The computer runs programs; the one who wrote the program thought about how to solve the problem, found the solution, fed the program into the computer, and then the computer executes the program that was fed into it. Therefore, the one who solved the problem was really the programmer and not the computer; a computer does not know how to solve problems, and a human being does. That was the first difference I spoke about regarding ordinary computers.

In a language model, that difference no longer exists. A language model knows how to solve problems. In other words, I am not feeding it a program that instructs it how to solve the problem. I ask it a question and get an answer from it. I do not write it a program that constructs the answer it gives me in the end; it decides on that answer by itself. In that sense there is a leap here, and that is what creates all these discussions and all these confusions that accompany us here. That is the whole story. In other words, that one difference that existed between a classical computer and a human being does not exist in the difference between a language model and a human being. The second difference, that there is no mental dimension, exists here too. But the first difference, which says that the one solving the problem here is not the machine, is no longer correct. In a language model, the one solving the problem is the machine.

Now, I’m not talking at the moment about tools, and we may get to that later, agents and all that, where in fact the human being also intervenes and helps the language model solve the problem. Let’s assume we are talking about a naked language model, meaning just it, without a wrapper, without tools, without anything. In other words, just an ordinary language model like there used to be at the beginning, okay? So in that situation, the language model answers all sorts of questions, a very wide variety of questions, and it answers them correctly. It solves very many kinds of problems and it does so correctly, and I did not tell it how to solve the problem. In other words, it solved the problem on its own. Here it is not the human being who solves the problem but the machine. And therefore all the anthropic confusions we talked about in the introductory conversation begin, yes? That is why people say: wait, wait, so in fact there is a human being here. This is no longer a computer like the kind where it was obvious that an electrical circuit is not a human being. Here we have something that, although it is implemented by electrical circuits, in the final analysis it runs on a digital computer, it is software, a language model, but still, you can see that this software is software that succeeds in fitting a solution to all sorts of problems, and that fitting is done by it and not by the programmer. Rabbi? Yes.

[Speaker G] Can’t one say that basically all the optimization methods and the training and those things are what caused the model to be like this?

[Rabbi Michael Abraham] And because of that caused it in some way—you know what, since you asked, I’ll address it here, because I planned to do it later. One could in fact say that the training process, in which we build the weights inside the neural network, is really the program we fed into it, and that program really tells it how to solve the problems. But that is disingenuous. It is disingenuous, because when we fed all that information into the network, we ourselves did not know how to solve the problem that the network knows how to solve. So clearly we did not really solve the problem for it. Rather, what did we do? We gave it, we put it through, some process that created in it abilities to solve problems. But that process is not feeding in a program that tells it how to solve the problems, but something much more general. That is one thing: it is not focused on a particular problem, where I feed it a particular program and it solves the problem. No. I fed it something general, I created some kind of monster here, and this monster knows how to solve lots and lots of kinds of problems, including problems that I myself do not know how to solve. So I think it is disingenuous to say that here too I am really the one who solved the problem.

I’ll say more than that. In fact, a human being also learns this way. When we teach a student in class to solve a quadratic equation, okay?

[Speaker I] So who

[Rabbi Michael Abraham] solved the problem, the student or the teacher? The student did it in homework, found the two roots of the quadratic equation. Did the student solve it, or the teacher?

[Speaker J] There’s noise, there’s noise in the background. If maybe whoever it is could mute…

[Rabbi Michael Abraham] Ah. Okay, maybe I’ll mute. Okay. So the question is: when the student solved the homework and found the two roots of the equation, did the student solve the problem or the teacher? The teacher trained him through examples, showed him correct examples, incorrect examples, taught him how to do it, and the student is just carrying out what the teacher taught him. In what sense is that different from training a neural network? In fact, the teacher—Shmuel will probably say this—trained the student’s neural network, and now the neural network solves the quadratic equation. So in fact one could perhaps say that this difference between a neural network and a computer is not really correct, because here too the person who trained the network solved the problem. In that sense, a teacher who teaches a student is the one who solved the homework problem when the student solved it. But again, I think that is disingenuous. It is disingenuous, because when the teacher teaches the student, he teaches him how to solve a quadratic equation, and the homework the student gets consists of quadratic equations; he has to solve quadratic equations.

In a neural network it does not work like that. I feed into it, load into it, enormous amounts of information, and in the end somehow it turns out that it knows how to solve many, many problems. By the way, including problems that were not really solved in the information it received, at least not directly. Therefore, I think it is somewhat disingenuous to say that here too the human being solved the problem. The human being created this creature; it is a creature created by the human being and not by the Holy One, blessed be He, yes? But after the human being created this creature, now this creature ostensibly really thinks. In other words, it receives a problem, decides on a way to solve it, and solves it. And that is what creates the hesitations, or intensifies the hesitations, that have arisen in the last few years, because a language model really does solve problems, unlike a computer. A computer does not solve problems.

[Speaker G] Rabbi, I can’t understand the difference between this and a computer and water. After all, in all of them in the end there is some inanimate thing running according to laws.

[Rabbi Michael Abraham] There is no—no, it is not running according to any laws that I fed into it. It runs according to laws, of course; my brain also works by the laws of nature, but that is not—yes,

[Speaker G] But ostensibly, ostensibly, the claim you are making is that in our case at least there is a subject, there is an intellect that drives things, not the laws of nature.

[Rabbi Michael Abraham] No, that difference exists. I said that is the second difference,

[Speaker G] Ah, okay, okay.

[Rabbi Michael Abraham] and it still exists. But the first difference, which says: who solved the problem? Was it the machine or the person who built or programmed it? Here I think there is definitely room to say that the machine solved the problem and not the human being, unlike a computer.

[Speaker B] Right, but if—

[Speaker E] May I?

[Rabbi Michael Abraham] Okay.

[Speaker F] But we trained it. We trained it. There are several global conjectures that were solved by AI.

[Rabbi Michael Abraham] What?

[Speaker E] That sounds, that sounds like—

[Speaker F] There are several global conjectures that were solved by AI.

[Rabbi Michael Abraham] What was solved? I didn’t hear.

[Speaker F] Conjectures, conjectures.

[Rabbi Michael Abraham] Ah, conjectures. Yes, mathematical problems, it was just published, yes? Yes, there are several, four-thousand mathematical problems that are very central. Jacobian and…

[Speaker F] Out of them, seven hundred—

[Rabbi Michael Abraham] twenty-two were solved by GPT. They published it in an article.

[Speaker E] And is that the version available to us, or is that a different version?

[Rabbi Michael Abraham] No, no, no, that is a version only they have.

[Speaker E] Claude too, as well.

[Rabbi Michael Abraham] Ah. A version only they have.

[Speaker F] Claude too, Claude also solved some.

[Rabbi Michael Abraham] Fine. In any case, yes, the mathematical world is now in an uproar. I spoke with several mathematicians a few days ago; you can also read about it; the mathematical world is in an uproar. In other words, people do not know what will happen in the mathematical world in another year. We are not talking about five years from now—one year from now. And it could be that in a year the mathematical world will no longer exist.

[Speaker I] Do people understand how it does it? Do they understand how the AI—we don’t understand what process is happening inside. I mean, they built the algorithms, they programmed everything.

[Rabbi Michael Abraham] No, they did not build the algorithms! No, they did not build the algorithms.

[Speaker I] No, I mean originally, originally, the LLM and all that.

[Rabbi Michael Abraham] Ah, the LLM, yes, but there is no algorithm that solves a specific mathematical problem.

[Speaker I] And since then they have no idea how the AI…

[Rabbi Michael Abraham] Correct. Even with simple problems, by the way, they have no idea how the AI actually answers. In other words, it is not that we understand—as I understand it, at least as far as I understand, even the experts dealing with it do not understand how this thing ultimately happens. When you attach an LLM to—then it becomes deterministic; there it is clear. But how the core of artificial intelligence works, I do not think anyone truly understands.

[Speaker E] Regarding the idea of solving, that sounds like saying that it’s not—it sounds something like saying that human beings do not solve problems, but rather evolution, which created certain pressures so that our brain developed in this or that way, is what ultimately solves the problems. I mean, in the end it was trained in a similar process.

[Rabbi Michael Abraham] Right, it is very similar, very similar. You are only expanding the teacher example I gave earlier. You are basically saying: the teacher taught quadratic equations, but evolution is what created for me the brain that can both absorb the lesson on quadratic equations and lessons on other topics, and in fact evolution somehow built—I said that we were built by evolution; the language model was built by human beings—but once you are already referring to the built thing, it is very similar: both solve problems and decide on their own ways of solving problems, right. So what is the difference?

[Speaker B] So then what is the difference?

[Rabbi Michael Abraham] No, I did not say there is a difference. I said that this difference really does not exist, ostensibly. We will get to that later, but ostensibly it does not exist. The second difference, which speaks about the mental dimension, exists here too, okay?

[Speaker E] No, I’m trying, I’m trying to say that in this example it does seem to me correct, at least perhaps under a broader definition of solving, that we solved ChatGPT’s problems, at least in your terms, because we created the evolutionary conditions, we created the situation, the evolutionary conditions, right, that caused…

[Rabbi Michael Abraham] No, that is what I said, that is what I said earlier, but I am saying that is disingenuous. It is disingenuous because I did not solve a problem that I myself do not know how to solve. A programmer, when he programs a computer, knows how to solve the problem; he is the one who wrote the program for how to solve the problem. Here, I myself do not know how to solve the problem and suddenly ChatGPT solves it for me. Yes, but it is not only mathematics. The fact that I built it does not matter. I built it—fine, okay, so what if I built it? But did I solve the problem? By the same token, with us too, the Holy One, blessed be He, created us. Does that mean that everything I think is really something He thought?

[Speaker E] No, He created the device. I don’t know if that is correct, because say, say, say a mathematics student now has to diagonalize—sorry, find the Jordan form of a matrix, and he does not—I mean, let’s say it happens to be a complicated matrix, and it is a very complicated algorithm, and he is unable to apply his own algorithm to the matrix, and he cannot solve it. No, no, no, that is just a question of time.

[Rabbi Michael Abraham] I don’t see why that is not analogous, though.

[Speaker E] Because here he cannot solve it even though—that is not analogous, not in any way.

[Rabbi Michael Abraham] If you give him the steps, if you tell him the steps one by one, he will be able to solve it. But if you give him the algorithm, he will not be able to solve it, so—exactly! Exactly! No, exactly, and that does not exist in the case of artificial intelligence. In the case of artificial intelligence, it is not that you cannot solve it because it consists of many steps and you won’t have enough time; you do not know how to do it. With a complicated problem, you know how to do it, it will just take you a very long time, so you tell the computer to do it faster. That is exactly the difference between a computer and a human being. A classical computer does not know how to solve anything that human beings do not know how to solve.

[Speaker E] I really did not understand you that way; it sounds absurd to me. Are we unable to understand the logical steps it takes?

[Rabbi Michael Abraham] Of course not. I mean… who said there are even logical steps there at all? That is a very big debate.

[Speaker E] So how do we even say that it solved those questions?

[Rabbi Michael Abraham] Because I see… because I see the answer. It presents me with a proof. Lior.

[Speaker F] Ah, okay.

[Speaker E] That—

[Speaker F] It’s not only mathematical, there is also biological.

[Rabbi Michael Abraham] No, no, but also in mathematics it produces a proof. No, but will it have a proof? I don’t know.

[Speaker F] Did you see, did you see what they wrote about the genome?

[Rabbi Michael Abraham] What there?

[Speaker F] When the genome project opened, they wrote that what the artificial intelligence did was equivalent to 800 years of

[Rabbi Michael Abraham] science

[Speaker F] by human beings, by human beings.

[Rabbi Michael Abraham] Of course. Human beings… By the way, that is true with a computer too. In other words, a computer that solves some problem, I don’t know, the product of two gigantic numbers with, I don’t know, 10 to the 100 digits, that too will take

[Speaker E] far more than human years to solve; there is nothing special about that. So why do we believe its solution at all?

[Rabbi Michael Abraham] I mean, what reason in the world do we have to think that it… That is what I asked the people there about this matter. The claim is that when they check it using Lean—there is a program that checks mathematical proofs—and that program can confirm whether a proof is correct or not correct. It is a deterministic program. And that too is artificial intelligence? It is a deterministic program.

[Speaker E] And that too is artificial intelligence, or is that something else?

[Rabbi Michael Abraham] No, no, it is a deterministic program. Okay, fine.

[Speaker F] And with AlphaFold what happened is that no human being ever did it; they simply could not do it. It’s a solution, what do you call it?

[Rabbi Michael Abraham] By the way, they told me that out of those 722 problems, there were one or two where they found errors. But that doesn’t matter; there are many others where they didn’t. In other words, apparently it is only a matter of time until the fellow, I don’t know, manages to solve almost everything that can be solved. Whether there is a glass ceiling or not is an issue we will discuss later—not that I have a clear answer, but we will try to define it better later.

[Speaker G] Meaning that in some of the problems it simply did not formalize it correctly in Lean itself. That is, it claims it wrote the problem correctly in Lean, but in practice it is written in a way that is not genuine, although the Lean code works. It is simply not the problem being discussed. So it is fooling us in…

[Rabbi Michael Abraham] Fine, that could be. I am not knowledgeable enough in this to… Good. So it is fooling us. I understand that this is what I was told, and I also understand that the formulation of the problem itself may be the issue, not the solution. It may be that it correctly solved the wrong problem. It could be that the error fell in the formalization of the problem.

[Speaker G] No, sometimes they also discovered that its objective function was simply to make the Lean code work, not to formalize the problem properly. So sometimes it just writes code that works in Lean and everything works, but it is simply not the problem we were talking about. So it is very funny. And if the rabbi—

[Speaker B] We’re talking only about mathematics. If the rabbi studies Talmud with AI, and it gives him a strong line of reasoning, and the rabbi does not feel that it is correct—not “correct,” I mean that it is right in the same way a human being is? The sharp logical distinctions… clearly, so then?

[Rabbi Michael Abraham] Clearly that is correct, so what? So what is missing

[Speaker B] there? What, it simply imitates a human being.

[Rabbi Michael Abraham] It imitates a human being.

[Speaker B] Right, so the question is what is missing? In every respect it is like a human being.

[Rabbi Michael Abraham] What is missing is the human being.

[Speaker B] But what is that human being? That is what we are looking for.

[Rabbi Michael Abraham] So I am saying, what is missing is the mental dimension that activates the… the… syntactic mechanisms.

[Speaker B] But look, we managed without the mental dimension in this matter.

[Rabbi Michael Abraham] Look, it did the same thing. We did not manage anything!

[Speaker B] We didn’t manage—who is the one who managed?

[Rabbi Michael Abraham] The AI did

[Speaker B] the same thing,

[Rabbi Michael Abraham] it produced an innovation that none of the greats… The rabbi studies… You are simply asking the same question. I will just ask you one question,

[Speaker B] because

[Speaker D] we are repeating ourselves a bit.

[Rabbi Michael Abraham] Do you think that the person in the Chinese room understands Chinese? No. What is missing there? If people could close the microphones because there is noise.

[Speaker D] Because he understands nothing, he understands nothing.

[Rabbi Michael Abraham] Why? He answers everything, all the answers relevantly. I…

[Speaker D] By the way, I did

[Speaker H] mute and it

[Rabbi Michael Abraham] still doesn’t help, something here… Now, now it is okay. Okay. The person in the Chinese room does not understand Chinese, that is all; that is what is missing. Semantics, that is what is missing. Syntax, for the moment I am assuming, can be perfect. All right? We will discuss later whether it really is perfect, or necessarily perfect, or not.

[Speaker B] Clearly what is missing is the aspect of feeling, of understanding, and understanding is not in knowledge, it is not in the domain of knowledge, it is not in the rational domain, it is in the non-rational domain.

[Rabbi Michael Abraham] Again, you define it as emotion, I define it as thought. It does not matter; there is no point getting into these semantic arguments, but there is something else missing here, that is all.

[Speaker B] So the point is… But let’s say, rabbi, what we learned is missing—sorry, just one more sentence. The fact is that if, say, that AI in another 100 years, probably much less, reaches the point where it answers all rational questions, all the Talmudic discussions exactly like a human being, all mathematics like a human being, then I will ask the rabbi: if so, what additional thing is there in a human being for which a soul was needed in order to answer those questions? Unless it is the domain of emotion.

[Speaker K] A human being has intellect, and this is only a brain. Exactly.

[Speaker B] What is the difference between intellect and brain?

[Rabbi Michael Abraham] I will give you the same answer—you are returning to the same questions I have already answered five times. You may disagree with my position, that is fine, but I said: it is superfluous, completely redundant. Even if there is no difference, even if there is no difference in terms of the range of tasks you can perform, there is still something in a human being that is not in a machine, that is all.

[Speaker B] But the rabbi admits it is superfluous, that part is

[Rabbi Michael Abraham] superfluous; it is not required for this.

[Speaker B] If humanity had been created without it, humanity would still have been created.

[Rabbi Michael Abraham] It would not have been created without it; what would have been created would be an imitation of humanity. Fine, okay.

[Speaker B] But in what way is it less? It answers all the questions, it creates the same works,

[Rabbi Michael Abraham] it passes the same tests. It does not answer any question! It does not answer any question. The answer to the questions is created in you when you read its answer. It itself answers no questions, and is asked no questions, and does nothing at all; it moves electrons. You relate to its output as an answer. The answer is created in you, not in it. That is only a feeling, that is only a feeling, I agree completely, I agree with the rabbi. And you don’t get up—yes, that is the famous question: if a tree falls in the forest and nobody

[Speaker B] is there, does it make a sound? Exactly, exactly.

[Rabbi Michael Abraham] The answer is no, it does not make a sound. Correct, because what is missing is the sensation, the sensation. And here too the model does not think; the model moves electrons. And the solution, when a solution to the question is created here, it is created only in the mind of the user, in the intellect of the user.

[Speaker B] In my view, I think we ought to be shaken by the fact that we can put all human rationality since the dawn of human history into a device, and what is missing—and it does everything perfectly on its own, without our support—and we say that the main thing is missing from the book. What is missing? Consciousness is missing, the rabbi calls it that, fine, but in the end—wait, the question is whether that is shocking or not.

[Rabbi Michael Abraham] Let everyone be shocked.

Or he won’t be shaken up; right now I’m just describing the situation as I understand it. Leave the shock aside, everyone—

[Speaker B] And no, no, no, I mean no, that’s not the focus. So what is it—that the Rabbi got stuck on emotion? I’m saying the meaning is that it’s clear to us that what’s missing is the main thing, and the main thing is emotion, not the answers and not the knowledge and not the rationality. It’s not emotion.

[Rabbi Michael Abraham] Come on, let’s stop. We keep going back to this again and again. It’s not emotion and not anything. You call it emotion, fine; we can keep arguing about that in a minute, what difference does it make right now? Look, the next stage I want to talk about was in 2024, about a year or a year and a half after this whole boom around GPT started. A new model came out called DeepSeek, Chinese, right? And this model had an interesting innovation. GPT also had it, in a model called O1, but the Chinese did it much more cheaply, much more efficiently, and in an open model. Meaning, you could copy it and do it yourself too. And the claim there was that they built a model that thinks according to the stages of human thought. In other words, when the model answers you, it tells you—it explains what it’s doing, how it approaches the problem: look, there are really three points here that I need to check, this point, this one, and this one. This point I solve in such-and-such a way, do the calculation, arrive at a solution. This point I solve this way, arrive at a solution. Then it sums up all the solutions, eventually gets to the bottom line, and gives you a final answer. But it also gives you all the inferential steps in between. The thinking, right? That’s called chain of thought, CoT; meaning it gives you the whole chain of thought that led it to the solution. Now, it turns out that computers, or models, that worked this way, first, managed to imitate human thinking, and second, managed to reach better answers than models that didn’t work this way. Okay, those are the facts. That’s what came out there, and it caused a huge stir in the world. This thing, beyond the fact that it was Chinese and they did various tricks there and maybe copied and maybe didn’t copy, doesn’t matter—but in the end, this phenomenon in itself made a tremendous amount of noise in the world, because it really starts coming close to human thought. Here it’s even closer to the mental processes that accompany the flow of electrons in our brains. It already tells you: what am I thinking? How am I breaking down the problem? What am I doing with each stage? And the entire chain of thought also comes out of this model, not just the bottom line. It really starts sounding like a human being, and when you talk to this fellow, you really do see a person in front of you. I mean, it really was confusing.

Now, I tried a bit to check with people in the know, and I also read a little, about how exactly this whole story works. And my question, basically, what I asked people—I asked several people—was: how was this model trained? Did they basically train it too with all the examples, but only ask it: give us the chain of thought, and it managed to produce the chain of thought? Or not? In the training itself, did they require it to produce the chain of thought and give it feedback on that too? You understand? Meaning, when I asked it a question, in training already, even before it was built, it had to produce for me the whole chain of thought and the answer. Now suppose it produced an incorrect chain of thought, but a correct answer. As they say, it’s always good to make mistakes in pairs, right? So that they sort of cancel each other out, okay? So it made mistakes in pairs, made an even number of mistakes, and got to the correct answer. Now, in a regular model—not DeepSeek, O1, an older model—that worked directly, I wouldn’t know that. It gave me the correct answer, everything’s fine. It receives no feedback, right? It passed training on that example successfully, okay? In DeepSeek’s training, basically, even if it gave the correct answer, if the path wasn’t correct, it got an error message. It has to arrange its weights so that they also produce the correct path, not just the bottom line. It turns out that in fact the training was done that way. Meaning, no miracle happened there, where I trained it on examples and final answers, and only asked it to also give me the intermediate steps. No, it didn’t work like that. The training itself was also conducted on the intermediate steps. It got feedback on the intermediate steps too, not just on the bottom line. In other words, I trained it to produce reasoned answers the way a human being would produce an answer, and I didn’t train it only to produce the correct answer. Understand?

In game theory there’s a dilemma that has accompanied—or still accompanies—game theory, because in game theory we usually think of it as some kind of formalization or imitation of human thought. We’re trying to describe mathematically how human beings think, okay? Now, what happens when a person thinks incorrectly? Produces an error, okay? So game theory splits in two here. Those who see game theory as a branch of psychology will expect game theory to formalize the human way of thinking, even if it’s mistaken. In other words, I want to imitate a human being, okay? By contrast, those who see game theory as a branch of mathematics—which is the accepted view—there I don’t care about human beings making mistakes; on the contrary, game theory will give me the correct answer and help us not make mistakes. My goal is not to imitate human thought; my goal is to arrive at correct thought, okay? And that’s not always the same thing. But when you imitate human thought—if you build a mathematical machine that imitates human thought—then you’ll also make the mistakes human beings make, because you imitated a human being. You’ll get what a human being gives; that’s what you’ll get. But if you build a machine that works correctly, then you have a chance of offering solutions in cases where a human being might make a mistake, and you reach the correct solution, okay? Now, game theory is usually perceived as a branch of mathematics. But there is some tension there around this issue.

Here too, basically, it’s the same thing. Because the moment I train DeepSeek to imitate human thought, on the principled level that’s also a recipe for error. Because on the principled level, if it imitates human thought, then wherever human beings make mistakes, it too will make mistakes, right? So that’s no guarantee that you’ll get a better answer. But for some reason, it turns out that models trained this way—and today, from what I understand, all models are trained this way, though back then it was new—models trained this way really do also give better answers. Meaning, thinking like a human being improves the quality of the answers you arrive at. That’s what comes out there. Now…

[Speaker E] Can I ask two questions? Yes. So the first: earlier I gave the example of the student, and you told me that in an essential sense we can’t understand. But now it sounds like you’re saying exactly the opposite—that AI actually does indeed track human thinking very precisely.

[Rabbi Michael Abraham] If we train it to do that. I’m talking about a regular model. What? A regular model, a large language model. No, not those models that were trained—those models that solve…

[Speaker E] Those same models that solve the problems we talked about—

[Rabbi Michael Abraham] —they supposedly do operate that way. Fine? Huh? Before chain of thought, meaning when it was trained only on the correct answer, without the path of thought. Now I’ll say more than that: even the model trained on a path of thought—even there I don’t understand how it works. Why? Because the path of thought is its output. But when I ask how it created that output, I’m in the same problem; there’s no difference at all. There too I don’t understand how it was created. It’s just that there I also don’t understand how the human chain of thought was created, not only how the answer was created, because you have to understand that from its perspective, the human chain of thought is the answer it’s giving. It’s as if you asked it what a human being’s chain of thought would be—that’s the question. Say you asked an old model what a human being’s chain of thought would be, and suppose it gave you the correct chain of thought. Then I’d tell you I don’t understand how it got there, right? Same thing here. Same thing here: here too I don’t understand how it got there. I understand the answer in light of the chain of thought that leads to it, yes. I do not understand how the model created the chain and the answer. In that sense, the lack of understanding remains. Can we—

[Speaker E] But you don’t understand how—

[Rabbi Michael Abraham] You also don’t understand—

[Speaker E] —how your own chain of thought works; like, a human being also doesn’t understand how his own chain of thought works—

[Rabbi Michael Abraham] —it just sort of happens. Yes, but unlike a human being, in the case of the model, I built the model; I know what’s inside it. And how what I put inside it creates a chain of thought and answers—that I do not understand. How is that difference relevant? In a human being, from the outset I don’t understand, because I didn’t create him; I don’t know everything that’s inside the person.

[Speaker E] But how is that difference relevant? I’m not—

[Rabbi Michael Abraham] That’s why I call it “I don’t understand.” What do you mean? No, I don’t understand how the difference lies in my—

[Speaker E] —in my knowing the parts—

[Rabbi Michael Abraham] If you created a machine and it works and I don’t understand how it works, then how did I build it? I didn’t understand. Usually when you build a machine, you know how it’s going to work, and that’s how you build it.

[Speaker E] No, that’s not what I’m asking. I’m asking about my earlier example with the student doing Jordan form and a matrix. He doesn’t understand by means of the algorithm, but if you told him step by step, then he would do it. Right. Now here too, we don’t understand AI’s algorithm, but if it guided us step by step, the way you’re describing here, we would understand it.

[Rabbi Michael Abraham] No, we wouldn’t understand it; we would succeed in imitating it, not understanding it, because it’s not at all clear that it works with our kind of understanding. We could produce the answer because we would move the same electrons that the AI moves, that’s all.

[Speaker E] I think I understand the answer. The second question: aren’t we stepping outside our epistemology here if we’re talking about models that don’t imitate human thought? By what tools can we even talk about their drawing conclusions?

[Rabbi Michael Abraham] I don’t know. I can only check whether the final answer is correct or not. According to what?

[Speaker E] Because supposedly you compare it to known answers—

[Rabbi Michael Abraham] Known answers, of course. What do you mean?

[Speaker E] Ah, no, in that case yes—but if we’re talking about problems we haven’t solved—

[Rabbi Michael Abraham] Fine, but I’m saying: if on all the problems we have solved it manages to answer, that means it has abilities, so naturally my confidence regarding problems we haven’t solved can also be justified. Beyond that, many times I can also check the problems we haven’t solved. Even if I didn’t know them beforehand, after I know, I can check. For example, if there’s a mathematical problem I don’t know how to solve, a mathematical equation I don’t know how to solve, but if you give me the solution I can substitute it there and see that it solves the equation.

[Speaker E] Fine, that doesn’t prove anything if it didn’t know how to produce the solution, but—

[Rabbi Michael Abraham] Of course it does. If I see that the model gives me the correct solution, then it solved it correctly. What do you mean?

[Speaker E] But no, you need a general proof; you can’t prove—

[Rabbi Michael Abraham] No, doesn’t matter. I know it solved this particular problem correctly. Fine, I forgot that, understood. Okay, now come, I want to keep going for a moment.

Look, there’s a lot of work—at Anthropic too, by the way, there I think they exaggerated this a lot, I don’t know if that’s how Alfred—I wrote it down—what’s called “whitening the box,” meaning whitening the box or making it transparent. What does that mean? There are processes inside this box that we don’t understand exactly how they occur, but the fact is that they produce very impressive results. Now we’re trying—and today there’s a huge amount of research around this issue—to try nevertheless to understand what happens inside the internal circuits of this enormous network. Can we identify stages of thought there not through what it outputs outwardly? What it outputs outwardly is stages of thought because that’s what it was trained on. I want to know by what stages the machine works inside itself. For example, when I give it, I don’t know, “Tell me what the capital of France is,” can I identify, say, a circuit that scans the countries of Europe, and another circuit that holds a city-capital versus country table, and another circuit that does the search so that once I’ve found the country, what is its capital, and another circuit that then outputs the answer? If I manage to identify inside this machine certain circuits that know how to perform certain kinds of operations, then maybe I’ll be able to understand better how this whole thing happens inside the machine. And a great deal of work is being done on that today: an attempt to understand how to turn the black box, yes, the black box, into something transparent, meaning to try to see a little what happens inside it.

And there are already certain findings in these studies, and the question of how similar this is to how a human being works, or how dissimilar, is an interesting question. Because what is not similar to how a human being works, apparently we won’t be able to identify. Because we’re looking for patterns that we know from ourselves. But if it works according to a different pattern of thought—if it has a pattern of thought at all, I don’t know—but if it works according to a different pattern of thought, it’s not clear we’ll even be able to identify it.

And that reminds me a little of the issue—you know the so-called Bible codes, right? This thesis that all sorts of information is encoded in the Torah, with skips, equidistant letters, things like that. Among other things, supporters of these codes show that with their algorithms you can find, say, the name of Sadat’s murderer near Sadat, in proximity, in the same section, you’ll discover them. Or all kinds of historical connections of one sort or another. One of the critiques raised against this thesis is: why didn’t you warn us in advance? If you have a way to identify Sadat’s murderer before—then identify him before Sadat is murdered and give a warning. You’re wise after the fact, and after you know that Sadat was murdered by so-and-so, you find them both in the same section. You never found it ahead of time. That’s one of the critiques. And the answer to this critique was that we simply don’t know what to look for. If I don’t have the murderer’s name, how am I supposed to search? After all, the algorithm says: here is the name, find me where it appears with the minimal skip. So it turns out they find it in the same section where they find Sadat. But if I don’t have the name, I don’t know whom to look for—how can I predict it for you in advance?

So here too, basically, it’s the same thing. If this machine does not work according to human ways of thinking, it will be very hard to follow or identify how it works. Because I know how human beings search for the capital of France, and I break it down into certain tasks, and I understand that this is how I act in order to reach the answer. If I identify circuits—I know how to identify a circuit like that, with this kind of task—and across many questions I see that this circuit always operates when that type of task is needed, then I know how to identify that this circuit performs that task. But if this machine works in a way that is nothing like how human beings work, then it’s very hard to see how we could even identify its mode of thought. How will we decipher its mode of thought? All we do is map it onto our own way of thinking. But if it doesn’t map onto that, then we have no way to do it.

But insofar as I now train the machine to think—to produce the human chain of thought, to think in the human way—then I’m simply imposing human thought on it. And here there may be more possibility of whitening the box. Because once you already force it to work in that way, then you’re basically forcing it to think in a human way, and then there’s a better chance that you’ll find some circuit inside the neural network responsible for each type of task, because by definition these are already tasks in the way human beings think, okay? So there’s a better chance of whitening the machine. But it turns out—and this is surprising—that it also improves the result. Meaning, training the machine only on the final result versus training it on the chain of thought: in the second training method we get better results. That’s the fact. And that too creates a great deal of confusion. If so, this really already looks like a human being, and therefore in what sense is it not a human being?

I’ll just add maybe one more thing: if indeed the machine—say, the machine I trained in the first generation, right? The machine that was not trained on chain of thought—now, it still gave me correct answers, okay? Now, if after I trained it on chain of thought it gives me more correct answers, or a larger number of correct answers, that basically means that the first machine probably did not work like human beings, right? Because the fact is that after I forced it to work like human beings, the answers improved. Meaning that the first machine did not work like human beings, and still many of its answers came out correct. That even gives some indication—it’s not necessary, but it gives some indication—that it is entirely possible that there is some completely different way of thinking here, one that isn’t broken down the way human beings think, and still it can arrive at correct answers. So therefore the question whether the fact that the machine gives a correct answer also means that it thinks systematically according to the human system—that is an open question. It seems the answer is not at all clear. The answer is no.

[Speaker B] Rabbi, I didn’t understand—I didn’t understand why you’re surprised that when you train it also on the steps, it’s more successful. After all, if we assume—if we assume that the goal, that the tool of AI, and what we are basically aiming for in training AI, is to copy the human neural network with its weights, the whole of humanity’s, into the AI tool—so wait, into the AI tool—then obviously if we train it more…

[Rabbi Michael Abraham] No, no, no—where did you get that from?

[Speaker B] We’re not copying that—

[Rabbi Michael Abraham] Right, exactly. Then it really is less surprising, I agree. So it supports that assumption. But the assumption itself is a very non-trivial assumption.

[Speaker B] But it supports that assumption; the result supports that assumption.

[Rabbi Michael Abraham] Correct.

[Speaker F] Yes, but that’s what’s interesting: with ChatGPT or various kinds of artificial intelligence and the like, on simple questions it really does work like a human being. It immediately jumps to Wikipedia or something like that, and also gives you where it took it from. You can see that when it searches, right? Only on the hard questions—not again, that’s tools.

[Rabbi Michael Abraham] Again, tools. Because it accesses the internet. Fine, when it accesses the internet, that’s tools.

[Speaker F] But chain of thought—you need to check that in much harder thinking, where many times it gives an answer and doesn’t understand where it’s taking it from, it brings you some—

[Rabbi Michael Abraham] No, there are answers it gives on the basis of its internal knowledge, and there are answers it draws from the web. Fine, I’m currently talking about a closed machine; it’s not open to the web.

[Speaker F] By the way, over there, what they said about Claude, right? They’re talking there about how within six months to eighteen months, supposedly the golem will rise against its creator. Right, and supposedly this is causing an uproar now in America and so on. What do you think about that?

[Rabbi Michael Abraham] No, no, wait, how did we get there? We’re in the middle of a lesson—what are you—

[Speaker F] No, I just thought—fine, I thought I could ask. Fine. Yes, okay.

[Speaker K] Wait, but Rabbi, still regarding the part where you tell it the way to calculate, to train it—it seems to me that this only saves resources. That if, theoretically, I wanted to train it without telling it the method, and then, say, give it more cases, and then let it invent the method—

[Rabbi Michael Abraham] Could be. It could be that if you gave it broader information in training, you would reach results as good as training on chain of thought—maybe. But again, then do chain of thought on the broader information too, and presumably again you’d get even better results.

[Speaker K] Yes. Okay.

[Rabbi Michael Abraham] In any case, the point is that I want—I want—

[Speaker K] No, but chain of thought may slightly limit it to human thinking, whereas you want to open it up to a new kind of thought.

[Rabbi Michael Abraham] Correct, that’s what I said. Therefore there really is something surprising here in the fact that the results come out better, because it didn’t have to be that way. It could be that human thought is biased in certain aspects; we have a tendency to make mistakes in a certain kind of task, and if you attach it to our form of thought, then I would certainly think there are some things where we would actually cause it to err in that way—the point I made about game theory. Yes.

[Speaker E] Could you clarify for a moment exactly what is meant by human thinking? Because if my brain works by making logical inferences that human beings don’t regard as valid, and for it that works—what exactly is meant by human thinking?

[Rabbi Michael Abraham] No, when you—as I said before—when you look for the capital of Paris—fine? So now I say: give me your way of thinking; I say to the model: give me your way of thinking. Now, if it gives me something like what I described before—it could be that some people would do it by different methods—but we still know how to identify a sensible human method. So there are people who would say: let’s build a table of city versus country, let’s search the countries in Europe, I don’t know, divide the countries by continents, locate France, go to the table, search where Paris is, which city is associated with France, capital city, and then output the answer: Paris, okay? So that’s a sensible way a person would operate. Someone else may suggest a different algorithm; it doesn’t matter. So if the model gives me a method of that type, it gets a positive score. Meaning, everything’s fine, you succeeded. But if it gives me—

Something else—I say: no, no. Even though he writes “at the end, Paris,” I tell him: no, that’s a mistake; correct the internal weights. I want a method that will be acceptable to me, okay?

[Speaker L] Could it be, Rabbi, that he’s thinking on two tracks in parallel? Meaning, he has his own mode of thinking, which isn’t necessarily human logic, but he also knows that in his training he’s expected to present how a human being thinks, and so he presents that too.

[Rabbi Michael Abraham] I’d formulate it a bit differently, though I think it’s the same point. As I said earlier, in response to what Lior asked before, I said that even after there is a chain of thought, we still don’t understand how it works. Because we don’t understand how it produced the chain of thought itself, okay? Now here—you understand that in the end I don’t think that once you’ve created a chain of thought, that makes me identify this machine with a human being any more than a machine without a chain of thought. Because in the end it’s the same kind of thing—what difference does it make? I simply trained it, by means of examples, where I demanded a more detailed answer from it. But it is still done in exactly the same way. If you gave a correct answer, you get positive feedback; if you gave an incorrect answer, you get negative feedback; then arrange the weights accordingly.

[Speaker L] Yes, but the Rabbi is saying that the answer he gave—not necessarily the reasoning he presents—is what led him to the answer. It could be that some other reasoning did, yes, that’s the…

[Rabbi Michael Abraham] Exactly. And that’s why I said that even after he presents a chain of thought, it still doesn’t mean we understand how he worked. Because the question of how he produced the chain of thought itself is exactly like the question of how he produced the answers in the previous generation. Now the answer simply includes the chain of thought—that’s the answer. So now we ask: how does he produce that? Same thing. And therefore I say that I don’t think anything essential has changed here in the additional generation as compared to the previous generation. The question remains exactly as it was, and therefore I think this is still some kind of golem that produces answers on the basis of training. And anyone who decides that the previous generation is not a human being—I see no reason at all to decide that this generation is a human being. I don’t think there is any essential difference between them, because in the end we trained it on the chain of thought. I want to sharpen this point further: suppose I had trained the previous generation only on final answers, okay? That’s it. I would take a machine from the first generation and train it only on examples with final answers, and then I would say to it: listen, produce your chain of thought for me—how did you work? Okay? Or how should one work, it doesn’t matter. And it would produce some chain of things for me, but I hadn’t trained it on that; it would just produce such a chain. And suppose that chain was really identical to the way human beings think, okay? Then I would understand—even though I still wouldn’t agree—but I would understand people who say: wait, wait, this thing is really starting to become human, because it manages on its own to generate a chain of thought. We gave it only the problem, and it chose the chain of thought and reached the solution. That really is similar to how a human being works. But understand—that is not what happened there. What happened there is that we trained it to produce a human chain of thought; that’s what we trained it on. So in the end it didn’t really create the chain of thought; rather, we forced it to create a chain of thought. So what changed? That doesn’t change much regarding the question whether it is human or not human. If you’ve now decided it’s human, then decide the first one was human too. And if the first one is not human, I see no reason to say that the second generation is human. There is no difference at all. All I did was train it to imitate me also in the chain of thought and not only in the bottom line, but it’s still only imitating me. But Rabbi—

[Speaker B] In the first generation, when they checked the chain of thought, didn’t they see that in most cases it nevertheless did follow a human chain? As if—less than when they trained it?

[Rabbi Michael Abraham] I didn’t understand. Again.

[Speaker B] When they checked the first generation and said to it: we didn’t train you on the chain of thought, now give us your chain of thought—wouldn’t they see, in most cases… probably in most… I don’t know. Probably in most cases we would see that in fact it really did follow a human chain, and in the cases where it didn’t, it made a mistake?

[Rabbi Michael Abraham] Maybe. I don’t know. I’m not aware of that. Let me just finish with the example I referred to in my book on evolution. One of the creationists’ arguments, as they’re called, is: how can it be that evolution keeps advancing and becoming more sophisticated all the time without a guiding hand? Things don’t become more sophisticated without some factor directing them in that direction. Now, the example they often bring—the “to be or not to be” example—is already one that really gets people excited. What does it mean? Think of a computer that is supposed to randomly generate a string of fourteen letters, okay? What is the probability that it will generate the sentence “to be or not to be”? Shakespeare’s phrase, yes. Zero probability, right? Meaning, twenty-two to the power of minus fourteen. There are twenty-two letters for each slot, fourteen letters total, so the probability is basically zero, okay? And therefore, our world is much more complex than the sentence “to be or not to be,” so to generate, without a guiding hand, a world like ours is unreasonable, and therefore it is obvious that there is someone here who directed this process. That is the creationists’ argument. What do the neo-Darwinians answer? And you’ll find this in many places. They say: that’s simply a misunderstanding. Why? Dawkins calls it climbing Mount Improbable—reducing the improbable slope. If there is some improbable slope here, it’s not that it’s improbable for it to happen by chance; evolution lowers or moderates the improbable slope, and then there is a greater chance that you’ll succeed in climbing up that incline. Why? Because evolution basically says that the process of natural selection directs you toward reaching the more viable, or more sophisticated, or more complex creature, okay? It doesn’t happen randomly. This process of natural selection, genetics, and so on, basically directs you there. Now in the “to be or not to be” example they say the following: suppose I wrote a program, and I tell the computer: now generate a letter. It generates one—say the first is b. No, generate again. S. Generate again. L. Stop. Write down l. Generate another letter—the next one. Again we keep going until it reaches e. Stop. Now generate again, until I get t—“to be or not to be,” yes? And I do the random generation each time, and at a certain stage I stop it at the correct letter and move to the next stage. You understand that the probability of getting “to be or not to be” in such a case is much, much higher than randomly generating the whole string at once, right? And that is what Dawkins argues, or his followers, it doesn’t matter. What they argue is that evolution basically does the same thing. We don’t just randomly generate and suddenly out come human beings. Rather, there is a multistage process here, where each stage basically dictates the next stage, and therefore the laws of evolution—or the laws of biology, if you prefer, physics, or whatever—the laws of nature, basically moderate the improbable slope and make the probability that a human being will arise something possible. It is much less tiny than the probability would be if this happened in one step. Now, the question is whether this is really a good counterargument. My answer is no. It’s a bad argument. Why is it a bad argument? Because let’s think about the “to be or not to be” example. How did we actually succeed in making that string arise with a reasonable probability? Simple: because there was a programmer who told the machine: stop at the first letter; if you reached l, stop; move to the next letter. Right? So who actually created the “to be or not to be” here? Not the machine—the programmer created it. The machine generated randomly, but the programmer kept stopping it. You got to l—stop, move to the next one. You got to e—stop, move to the next one. He built “to be or not to be” with his own hands. So of course the probability improves, but the probability improves because there was a factor here that directed the process. That is exactly the creationists’ claim: that evolution is a process that improves the probability of arriving at a human being, and that means there is a factor standing behind this process, and it is what actually improved matters; it solved the problem, not nature’s computer. Do you understand? One second. Do you understand the analogy to our case? The analogy to our case is exactly the same. The question is: who is actually solving the problem? Okay? Or who is actually producing the chain of thought, if you like. Okay? Is it true that if I now tell it: listen, generate a chain of thought, and I give it feedback on an incorrect chain of thought—correct the weights—then obviously in the end I am moderating the improbable slope, and it will arrive with a higher probability at the correct answer. Okay? But obviously the reason it arrived with a higher probability at the correct answer is me, not it. Because I forced it to moderate the slope. I basically forced it to adopt the human chain of thought, exactly as the programmer there forced the computer to produce the string “to be or not to be.” It is exactly the same argument. And therefore I see no philosophical difference, for our purposes, in this advance to the second generation of computers that also produce the chain of thought. From here on I’ll continue next time, because the chain of thought will help me illustrate some more interesting points at the next stage, next class. Okay. Can you point out the analogy again between the two cases? What?

[Speaker E] Can you point it out? I don’t see exactly the analogy between what you said here and artificial intelligence.

[Rabbi Michael Abraham] In artificial intelligence, what happened is that you are basically forcing it to build a chain of thought in order to reach the correct answer, right? Through the training.

[Speaker E] Ah, okay, fine. Fine, yes.

[Rabbi Michael Abraham] Which is exactly how you forced the computer to arrive at the correct combination.

[Speaker F] Basically they call evolution God. That’s what they’re doing.

[Rabbi Michael Abraham] Right. Or in other words: God created evolution. God created the laws, and those laws that create an evolutionary process are responsible for moderating the improbable slope.

[Speaker E] Yes, but in the example you bring, you’re not choosing a filter—you’re choosing something that makes the probability 100 percent; you’re really choosing an actual guiding hand here, yes? If it stops at the letter l… No, if, say, you say: generate a random sentence, but you say: put a space every 4 to 7 letters and don’t make the space more than 8 letters away, then that increases the probability, and so on and so on. You don’t have to add filters that are a guiding hand; you can add filters that just reduce the probability.

[Rabbi Michael Abraham] No, the filters—

[Speaker E] Those are the guiding hand.

[Rabbi Michael Abraham] Any filter that increases the probability is a guiding hand.

[Speaker E] I think that definition disconnects it from the semantics of directing, because when we’re talking about nature—no. Randomness has no will, no, because… you can direct by—

[Rabbi Michael Abraham] Increasing the probability, that’s all. So what?

[Speaker E] No, because as the Rabbi says when he distinguishes between free choice and randomness, behind randomness there is no will, so it’s not…

[Rabbi Michael Abraham] Right, so that is randomness.

[Speaker E] It’s something different, ah…

[Rabbi Michael Abraham] It is directed randomness—that is exactly the point. There is a non-random dimension here that channels the randomness in the right direction. That is exactly the point. That non-random dimension is the guiding hand. True, it does not direct things deterministically; it only increases the probability that it will get there. You know, another example that neo-Darwinians bring in this context is the drunkard example. They also repeat it everywhere… these things drive me crazy, because they bring these arguments as arguments against the creationists, when they are the best arguments in favor of creationism. So this stupidity is just astounding. They say, for example, there is some drunk who comes out of a tavern, swaying all over the place, and on one side of the road there is a wall, and on the other side there is a ditch, and he sways right and left, completely drunk. You understand that in the end he’ll be in the ditch. Okay? In the end, in the end the wall… there, you see? That means you don’t need the drunk to direct himself toward the ditch; I can tell you in advance that he’ll end up in the ditch, even though the drunk is functioning completely randomly. There, there’s your proof against the existence of God. Now I pinch myself—I cannot understand how intelligent people fail over something so foolish. God is the one who created the wall on one side and the ditch on the other side. Once He created that, anyone who behaves randomly within this system will end up in the ditch. So true, he got there through a random step, but the deterministic circumstances within which the random process takes place—that is the guiding hand.

[Speaker B] But the reality, Rabbi, is that the world is cruel, and the weak disappear… Rabbi, practically speaking, directed randomness is like—

[Speaker M] The watch analogy.

[Speaker B] I don’t understand—is that called God? What? I don’t understand. The condition, the tragic reality in which the weak do not survive. But to come and define that tragic reality, where the weak disappear, and say that this is God and look, we found God?

[Rabbi Michael Abraham] I’m not dealing with God’s good and evil; I’m dealing with logic.

[Speaker B] No, but how can you call that a guiding hand? How can you call such tragedy a guiding hand?

[Rabbi Michael Abraham] If you see that the probability of this happening is very low, then there is a guiding hand here. It may be the guiding hand of a wicked villain—okay, fine—but it is still a guiding hand. I’m speaking about statistics, not ethics.

[Speaker M] The reality in which the weak… Rabbi, is directed randomness like the watch analogy you spoke about back then? What? Well, you said directed randomness—meaning the guiding hand in directed randomness? Yes. Yes, something else that channels the random process to its destination.

[Rabbi Michael Abraham] So what? I didn’t understand what…

[Speaker M] I mean it’s like the analogy about the watch. What is it?

[Rabbi Michael Abraham] Paley’s watch? Paley’s argument with the watch, that it was not created by itself?

[Speaker M] Yes—that it wasn’t, that he found a watch and he didn’t understand it, and he said: ah, this watch works randomly—but no, he simply didn’t understand the watch. There is a certain random form here, but also… ah, no, that’s something else.

[Rabbi Michael Abraham] You’re talking about the argument—

[Speaker M] About the excuse of my broken watch.

[Rabbi Michael Abraham] Yes, the broken watch. That’s someone saying—that is an answer to what Shmuel just asked, yes. It’s still the case that there is a watch here, just a watch with a mechanism I don’t understand, yes?

[Speaker E] Can I ask questions not related to the current topic? Yes. So you told me regarding Spinoza’s philosophy that it seems to you like a meaningless game, and I…

[Speaker K] Rabbi Yuval, thank you very much, Rabbi. Thank you, all the best.

[Speaker E] And I tried to compare Spinoza’s processes to mathematics. And if, say, you compare it to modern mathematics—I mean calculus and things like that—the only difference I see is that we take the concepts as given, meaning we don’t define them, whereas he does define them. And besides that…

[Rabbi Michael Abraham] No, no, absolutely not. I can produce a mathematical field with concepts that have no anchor in actual reality in principle.

[Speaker E] Yes, that is exactly what I mean. Like a set or an element.

[Rabbi Michael Abraham] So what? It’s simply, simply, simply not… the mathematical strictness is a façade. There isn’t really mathematical strictness there. I agree.

[Speaker E] Yes, yes, I’m not arguing with that.

[Rabbi Michael Abraham] Good, so that’s why it’s not similar to mathematics.

[Speaker E] And why is it not similar? Wait.

[Rabbi Michael Abraham] Because the mathematical strictness doesn’t exist. The rigor. It is not true that the conclusion follows from the premises.

[Speaker E] I don’t understand what you mean that it’s not true that the conclusion follows—

[Rabbi Michael Abraham] It is not true. The argument is invalid.

[Speaker E] The mathematical argument is invalid?

[Rabbi Michael Abraham] His argument is invalid, yes.

[Speaker E] Ah, Spinoza’s? Yes. So that’s why it’s not mathematics. So he made a logical fallacy, as it were? There is a façade of mathematics; he is trying to show as if there is some axiomatic structure here, but that’s not true. There are conceptual ambiguities there, and there are intellectual leaps there. It is really not a deductive, axiomatic structure.

[Speaker L] Rabbi, now regarding the question—again I’m bringing it back—with AI, human or not human, there are several AI experts and philosophers who claim yes, that AI is getting very close to being human and has consciousness, like Geoffrey Hinton, the godfather of AI, who started the whole story. The way I see it, it has to do with the question of how atheist, materialist, or not, you are. If you…

[Rabbi Michael Abraham] It’s not connected, but there is correlation. There is correlation. You can be a materialist and an atheist and not think that way, and you can be a believer and think that way, but… there is correlation.

[Speaker L] There is correlation. Fine. The way I see it…

[Speaker F] Rabbi, what about what I asked? What there? About what they said at Anthropic there—that between six months and eighteen months, yes? There are major concerns, especially in America now, yes?

[Rabbi Michael Abraham] I have no idea. We’ll wait and see. How would I know? I’m not an expert in this field, and I’m also not sufficiently familiar with the current situation. I’m not…

[Speaker F] But do you think that danger exists?

[Rabbi Michael Abraham] Of course such a danger exists. When it will materialize and whether it will materialize—I don’t know. And also what exactly the danger is, because there are many dangers. I don’t know. It also depends on how we defend ourselves, depends on what we do; lots of things can…

[Speaker F] No, this question is very important, because if half of high-tech today is invested in AI, yes, then the question is whether…

[Rabbi Michael Abraham] The fact that the question is important doesn’t mean I know how to answer it. I don’t… fine.

[Speaker N] I want to understand: humanity basically led this AI process, and today it doesn’t know how to explain it? Meaning, did the very people who started it know where it was going?

[Rabbi Michael Abraham] No, no, they didn’t know either. I spoke about this last time. On the contrary, they knew even less. I said last time that computer science originally is a branch of mathematics. Artificial intelligence is a branch of an observational science. Meaning, you discover a phenomenon and investigate it with empirical tools. It’s not that you built the machine and therefore know how it works because you simply built it according to your own plan. You simply test it with tools to see what it can do and what it cannot do. Was it some kind of discovery?

[Speaker O] No, but why did they think in the first place that this thing would work? I mean, what is the observation here? They built a machine, so why did they invest this energy before—meaning, did they see something in nature that hinted to them—

[Rabbi Michael Abraham] That this machine was possible here? When they made neural networks, they understood that a neural network could produce results. And once they tried to build a large neural network, it gave them results they had not expected—far, far beyond what they expected. The mechanism… Was a human neural network the first thing they saw? The mechanism of a neural network can be understood.

[Speaker P] Right, so then where is the failure, if so? I mean, why did they think that something that—

[Rabbi Michael Abraham] But how do you get from training on a huge collection of texts to such abilities of solving problems, with creativity, with… It is simply beyond comprehension. We have no idea how such a business works. You cannot describe the local minima that the system reaches and how exactly that happens. There is no way. It’s not…

[Speaker C] So what—they looked at a human neural network and said: there’s a chance this thing can work in a—

[Rabbi Michael Abraham] Not a chance. In a small neural network they saw that it succeeds in performing tasks. But in a small neural network you know what is happening there, because you can calculate it; there are ten sites here, so you can calculate. But what do you do when there are billions upon billions of sites? You have no idea what it will do, how far it will go, how to act on it, how to get it to do all kinds of things. It was a kind of experiment, a certain gamble. It turned out to reach truly grandiose results, far beyond what they expected, and today it is only running wilder still.

[Speaker E] Rabbi, a question about what you wrote in “Truth and Stability.” In the introduction, when you present nationalist ideology and religious ideologies in their dogmatic approach, where a person sacrifices his thinking—his thinking—to them, and that way he becomes more nationalist or more religious, in the fundamentalist approaches. And I can’t manage to see the comparison here between…

[Rabbi Michael Abraham] No, I didn’t understand the claim. That is the definition of fundamentalism. What exactly is the claim?

[Speaker E] No, I don’t see how you compare nationalism and religiosity here. Because in religiosity you sacrifice the… what does it mean, sacrifice the… sorry, in religiosity you sacrifice your intellect or the logical conclusions you reach in relation to their faith…

[Rabbi Michael Abraham] You can sacrifice it, and you can also not sacrifice it. And in nationalism too, you can sacrifice it, and you can also not sacrifice it. The fact is that in very, very ideological societies, like a fascist nationalist society and a religious society, there are more cases of… of people.

[Speaker E] To whom are you sacrificing when it comes to nationalism? I don’t—I generally also see religion, Judaism, as something national, and I myself am also nationalistic, which is why I can’t see the… You sacrifice yourself for the sake of the nation.

[Rabbi Michael Abraham] For the sake of the nation, by the power of that same…

[Speaker E] Sorry, I didn’t hear the difference.

[Rabbi Michael Abraham] Are you talking about national or nationalistic? They’re not the same thing.

[Speaker E] I’m talking about what you meant in the book.

[Rabbi Michael Abraham] What I meant in the book was fundamentalist nationalism. That is nationalism that does not pass things through the crucible of intellectual criticism. It is basically a kind of religious way of thinking.

[Speaker E] And why is that? I can’t manage to see why you locate the kernel of it in nationalism itself.

[Rabbi Michael Abraham] No, I don’t locate the kernel of it there. It’s an empirical fact. What? That there is… The empirical fact says that in ideological societies there is a greater chance that there will be a person who is fundamentalist, that is…

…that he is not willing to subject to critical examination. That’s it. No, but you’re putting them…

[Speaker E] That’s not how it is.

[Rabbi Michael Abraham] It’s just read from the book, you’re putting them… A person might arrive at it through his philosophical mind, or religiously, he might arrive at it…

[Speaker E] Through his philosophical mind, and then he’s not a fundamentalist. But it’s not just, “he read it from the book.” From the book you get them, you put them in the same category, and then you bring the quote, “the more absurd it is…”

[Rabbi Michael Abraham] More, from…

[Speaker E] That same Church Father: the more absurd it is…

[Rabbi Michael Abraham] The more I believe in it. Tertullian, yes. Yes, meaning, here it’s “the more absurd it is…” This is a phenomenon that exists in societies of that kind, the phenomenon of dogmatism. That doesn’t mean that you necessarily have to be dogmatic or fundamentalist, no. A religious person can exist—I think that I, for example, am a religious person, and I don’t think I’m a fundamentalist.

[Speaker E] I don’t think so either. But my question is—maybe I just didn’t understand you—so what you’re saying is that specifically here we’re dealing with two different kinds of fundamentalism, in nationalism and in religiosity? No.

[Rabbi Michael Abraham] Those are two examples of the same phenomenon. But this is not an identification of nationalism and religiosity with fundamentalism; it’s a correlation. Meaning, within the religious and nationalist world you’ll find more fundamentalists. And of course that’s not accidental, because if you have a thesis, you can be fundamentalist about it; if you don’t have a thesis, why would you be a fundamentalist? Today, in the postmodern era, there are people who are fundamentalists about the absence of theses. That is, they cling uncritically to the idea that there are no theses. They managed to produce fundamentalism toward a vacuum, which really is an impressive somersault—an impressive intellectual and psychological somersault.

[Speaker E] Okay.

[Speaker B] Rabbi, Rabbi? Rabbi, if I may ask one more question—can I also ask in another area, in a didactic area? This week I happened to study with—a request came that I sit with some ninth-grade boy in a yeshiva high school, and I studied Talmud with him. We learned for a few hours in Tractate Ta’anit there, on the first pages, 2, 3, 4. Now, it was really hard for me, because he’s a talented boy, and we’re sitting and learning the arguments of the Amoraim, and they’re trying to… the Rabbi roughly remembers the Talmud there, “Give dew and rain,” they invent inventions where the feeling is that there’s an unusual wind and air after the rain, and all kinds of inventions where first of all it’s obvious that they made this up not simply because they researched the subject and approached it with intellectual purity and wanted to check things empirically, but because they have a problem with the tannaitic source, with the baraita, and they need to reconcile it, so they invent inventions that have no basis in reality at all. You don’t feel they have any basis in reality, and you ask yourself: how am I supposed to teach…

[Rabbi Michael Abraham] I can’t answer a general question of that kind; you need to look at a specific example and try to think about it. But in general I’ll say—I once wrote an article about judging favorably and Ockham’s razor. And I talked there about the fact that when you judge someone favorably—like judging the Tannaim favorably—that doesn’t mean you’re not rational. If you trust that the Tannaim were wise people, and suddenly you see that they said nonsense, you’ll be willing to accept far-fetched explanations in order to explain why what they said was not actually nonsense. Okay? It may even be a far-fetched explanation, but it’s still the more reasonable explanation. The alternative is that a wise person said nonsense, and that’s not reasonable.

[Speaker B] Right, I accept that. I’m saying, after all, I come from a feeling of admiration for the Tannaim and Amoraim. But here you’re challenged by the fact that when you learn it, at least broadly, the feeling is that they’re making up baseless inventions, and for very self-interested reasons—not in order to check reality and answer it, but to resolve a tannaitic problem that…

[Rabbi Michael Abraham] We’d need to see an example; it’s hard to speak in generalities.

[Speaker B] We need to see an example.

[Rabbi Michael Abraham] In general, in the Talmud they often deal with questions that—I called them Platonic questions—questions that won’t appear in reality. Some extremely forced reality, yes? I’ve spoken about the interpretive setups; I also gave something on this on Sunday. They create some interpretive setup that is strange, some implausible reality; it’s obvious that the Tanna didn’t mean it, and I try to explain why it’s still a reasonable interpretation. It’s a reasonable interpretation because the principle the Tanna is stating will be implemented precisely in that very strange reality. There you can see it. So that doesn’t mean I think the Tanna was speaking about the strange reality; rather, through the strange reality I see the meaning of the principle the Tanna stated. If you want, you can look at my article about interpretive setups in…

[Speaker B] No, I know it, I heard that, and I very much agree with the Rabbi about the… we also spoke about it in a previous series, about the interpretive setup, I very much agree—but here I’m talking—here it wasn’t a matter of an interpretive setup. They’re inventing baseless inventions. Clouds—clouds that come after the rain. What, what clouds after the rain? What importance does that have?

[Rabbi Michael Abraham] Send it—send a question in the Q&A on the site if you want, something specific, maybe we can try to discuss it.

[Speaker B] Specifically, is it possible…

[Rabbi Michael Abraham] To try to discuss it.

[Speaker F] What, do we have to accept everything the Tannaim said? Because…

[Speaker B] No, how—my question wasn’t about a specific question. How do I teach, how do we begin teaching boys Talmud when we want them to love it and not—because I felt he was suffering. He said to me: “Listen, there are stretches here that make no sense, and what am I supposed to do with

[Rabbi Michael Abraham] this?” No, I think the didactic question is secondary. First of all, the question is how you learn it—what the truth is. After that comes how to convey it didactically. Right, right. If you understand why it is correct, I assume you’ll also solve the didactic problem. Right. There’s no point moving to the didactic plane; we need to examine how to learn the passage correctly.

[Speaker F] Yes, but a passage is learned regardless—whether it’s true or not true. You want to understand what they said, and you have to relate to it—not to put their heads inside mine, but I’m photographing them.

[Rabbi Michael Abraham] No, I also need to understand whether what they’re saying makes sense. Why? I also have criticism of them.

[Speaker F] But there are many things they said that aren’t logical. You’re trying to learn the…

[Rabbi Michael Abraham] If it’s illogical, then no—then apparently that’s not what they said, or I’m mistaken, doesn’t matter, but no. Or they really did say something illogical, and then it’s just that—then I don’t agree.

[Speaker F] Fine, we don’t have to agree, right?

[Rabbi Michael Abraham] Okay.

[Speaker E] Is there time for one more question, or…?

[Rabbi Michael Abraham] We can try.

[Speaker E] Regarding this: you present the idea of unstable truth as some kind of alternative, and I can’t see how—how it can serve, especially since you present it at the beginning as some kind of political alternative—I can’t see how it can serve as a political alternative. Because at the end of the day, the ordinary person, if I may put it that way, will never be able to use this method, both out of laziness and out of stupidity, and probably in the overwhelming majority as a combination of the two.

[Rabbi Michael Abraham] So how is it…? Look, I explain it there. My claim is—look, someone who is dogmatic and fanatical and fundamentalist will remain that way. But there are many people who, when you present… for example, let’s take religiosity, okay? There are people who are convinced that in order to be religious you have to believe 100%; if it’s not 100%, then you’re not religious. I think there’s a very strong current in religious education that leads to conclusions like that, to definitions like that. Now, what I’m saying is that many times, if I present the person with the very existence of an alternative—“No, you can be religious without being fundamentalist; you’re not certain, and you’re still fully properly religious”—then he will allow himself to recognize within himself that he isn’t really a fundamentalist. You usually won’t convince a real fundamentalist. But there are many people who push themselves into a fundamentalist position because they don’t see any other way out: “If I’m a believer, then obviously I’m a fundamentalist. How could it be otherwise?” So if I present him with a position according to which religiosity can be non-fundamentalist, suddenly: “Wait, wait, actually the truth is that I really am there.”

[Speaker E] Well, yes, I suppose because an act of self-recognition is nice, but it’s nothing more than that. It’s not that now he’ll take the method as you present it as an act he’ll operate by. How will he do that? For that he has to study thoroughly.

[Rabbi Michael Abraham] Then let him study.

[Speaker E] But he won’t do that—that’s reality, that he won’t do that. Why won’t he do that? Again, society… take, say, political parties in Israel.

[Rabbi Michael Abraham] Nobody reads their

[Speaker E] platform, apart from curses and that sort of thing, there’s nothing.

[Rabbi Michael Abraham] There is no platform. Most political parties in Israel have no platform at all. That’s been true for many years. A few have one. But never mind; I’m speaking generally. Humanity, or human society, is built in circles. There’s a small handful of people who think—critically, rationally—form positions, whether right or wrong, that’s not the point, but there are independent people, with independent thought, who form positions. Around them there is a circle that knows what they say, but is already not entirely independent in its thinking; it attaches itself to them. And around them there are more distant circles of the masses, who somehow identify with those figures and therefore follow them. Now, if I succeed in moving two people from the leading elite, then I’ve moved 30% of the public. Not everyone needs to become philosophers.

[Speaker E] What is the leading elite? Is that LeBron James and a pop star, like? What leading intellectual elite exists today that leads the…

[Rabbi Michael Abraham] Especially…

[Speaker E] In the secular world?

[Rabbi Michael Abraham] Of course it exists, of course it exists, what do you mean? In the religious world too, it’s the same thing.

[Speaker E] Who are the admired figures of the average secular person?

[Rabbi Michael Abraham] They’re not intellectuals. And do you think that Yitzhak Yosef is some extraordinary intellectual? No—I know that. Now, I explain to him: “Listen, the man is an idiot.” Okay? An idiot and corrupt and wicked, and I don’t know what else to say about him, really—the bottom of the human species, okay? That’s it, that’s what I tell him, fine? Now I say to him, “Look, I also have reasons; I can explain it to you. And let’s even say that he has Torah knowledge—he has some knowledge, never mind, even if distorted—but that doesn’t mean he’s a sage. Torah knowledge does not necessarily go together with being a sage.” Now, suppose I managed to convince him of this model itself. He doesn’t need to learn the whole Torah in order to disagree with Yitzhak Yosef. He’ll leave him, because he understands that it’s no longer necessary to follow him. Well, aside from…

[Speaker E] that…

[Rabbi Michael Abraham] Basically, teach him that the fact that he wears a shiny robe doesn’t yet mean he’s Rabbi Ovadia and the leading rabbinic authority of the generation.

[Speaker E] Well, aside from the fact that I think this story, as you present it, is wishful thinking and in reality probably won’t happen—but secondly, this is true specifically, I think, as some feature of religious society; it doesn’t exist in secular society. The model of admiration…

[Rabbi Michael Abraham] It exists, and very much so, to the same degree.

[Speaker E] The model of admiration in secular society is not a sage at all.

[Rabbi Michael Abraham] Even more so, actually. Even more so.

[Speaker E] Yes, exactly—it’s someone who has no connection whatsoever, not even indirect, not even malicious, not even fake, to education.

[Rabbi Michael Abraham] Not true. There are intellectuals in the secular world who have enormous influence. But they don’t have a direct stage. Their influence passes through many circles, including LeBron James and Madonna, who themselves have intellectual gurus of that kind—not specifically, I mean in general—and through them it can filter down to the broader public. Absolutely. In the secular world it’s no less than in the religious world.

[Speaker E] I just can’t imagine it.

[Rabbi Michael Abraham] There are opinion leaders there, there are definitely people there—absolutely, what do you mean? No less than…

[Speaker E] No, if we just take, say, one branch of this thing—let’s say, I don’t know, girls aged 18 or 19 out there who like Noa Kirel and are, say, influenced by her opinions. Just let’s say that. Okay. Do you think Noa Kirel reads, I don’t know, philosophical articles and is influenced by them?

[Rabbi Michael Abraham] No, again, you’re being so simplistic in what you’re describing here. Influence—intellectual and spiritual influence—passes through many, many media. Many, many media. Noa Kirel can hear some radio commentator or some radio program that is influenced by some philosopher or another, or some thinker or another, and that will influence her, even if she hasn’t read his writings and hasn’t delved into his philosophy. By the way, maybe she’s a smart woman too—I don’t know her, I don’t know. No, not specifically—I know, yes, I’m not speaking about her specifically, just generally. But I’m saying these influences pass in countless ways. And I really think intellectuals have a lot of influence, a lot of influence. In my opinion, a large part of our situation today is what’s called the betrayal of the intellectuals—that the intellectuals are not doing their job. They too are deteriorating. And if they were doing their job, the broader circles would also look different. Or some of them would.

[Speaker E] And can you explain what you mean by their job?

[Rabbi Michael Abraham] Yes. Their job is not to be fundamentalists, to form positions in a more rational way, to form positions in a more critical way, more balanced.

[Speaker F] Well, maybe they’re not intellectuals? But…

[Rabbi Michael Abraham] Again, there are intellectuals—I’m not going to get into the question now of who is and who isn’t.

[Speaker F] No, so maybe we don’t know that that really is what they’re like.

[Rabbi Michael Abraham] Fine. No, don’t you think that every…

[Speaker F] Every one, every one of…

[Rabbi Michael Abraham] I’m not talking now about a closed list of intellectuals. I’m saying spiritual influence, or intellectual influence, can pass through many, many media. Many media. I meet lots of people, by the way—it surprises me every time anew—who read my site and tell me it influenced them greatly, including completely secular people. Many of them, by the way, never comment on the site at all. Somehow I happen to meet someone like that, or hear of someone like that, or someone else who met him tells me. Now, again, I’m not saying this about myself specifically; I’m sure it’s true of anyone who spreads ideas. Anyone who spreads ideas has a great deal of influence. It’s a big mistake to belittle that, a big mistake. It has a great deal of influence. It passes through the mediation of the press, of celebrities, of all sorts of people of all sorts of kinds; it passes through a huge number of mediations, and it gets distorted along the way. All of that is true. It’s a very complex process. But it’s not true that the whole world is just wading in shallow water and there’s no possibility of doing anything. I don’t agree with…

[Speaker E] When I look at the press or the news sites and so on, especially the popular ones, I can’t see even a hint of what you’re talking about. Maybe that’s my blindness, but…

[Speaker F] Okay, I have seen it,

[Rabbi Michael Abraham] I do see a small handful like that, but okay.

[Speaker E] All right, and for hope…

[Speaker F] Each one of us is a collection of certain influences, plus a personal opinion. So certainly, yes, certainly, certainly that’s how it is.

[Rabbi Michael Abraham] All right, friends, let’s end here. Sabbath peace. Sabbath peace.

[Speaker F] Sabbath peace, goodbye. Sabbath peace.

Artificial Intelligence – Lesson 17 – Rabbi Michael Abraham

Table of Contents

Summary

General Overview

Full Transcript

Leave a ReplyCancel reply

Table of Contents

Summary

General Overview

Full Transcript

שתף

Leave a ReplyCancel reply