Moshe Koppel: Artificial Intelligence and Torah

David Bashevkin: 
Hello and welcome to the 18Forty Podcast, where each month we explore different topic bouncing modern sensibilities with traditional sensitivities to give you new approaches to timeless Jewish ideas. I’m your host, David Bashevkin, and this month we’re exploring prayer. This podcast is part of a larger exploration of those big juicy Jewish ideas. So be sure to check out 18Forty.org where you can also find videos, articles, and recommended readings.

There is a fantastic short story that somebody told me in the library of YU. I think one of the reasons why I remember it is because I was so enchanted with the author’s name. Many of you may be familiar with him. It’s the famous author, Jorge Borges. I think I’m pronouncing that name correctly, though if I didn’t hear somebody else pronounce it, I never would have thought that that is in fact how it’s pronounced.

And the story, which is one of his most popular, he’s a really phenomenal thinker, philosopher, one of the unique thinkers to live in the last a hundred years. He died in 1986. And Jorge Borges has this story called Funes the Memorious. Funes the Memorious is a story about a person named Iranio Fumes, who I believe gets into some like horse riding accident. And because of the accident gets hit on the head, he’s able to remember everything, does not forget a thing. And the story is really a rumination about a world where there is no forgetting. What would life be like if we never forgot everything, if we always had access to all information?

This is actually not just a case in science fiction. There was a real life story that was actually carried out by a neuropsychologist named Alexander Luria who wrote a book about his interactions. I’m curious from listeners if Alexander Luria is related to the famous Luria family. I assume that he is in some way, but I don’t really know much about his Jewish roots. But he wrote a actual book about his interviews with a client, with a patient, I’m going to butcher his name. I believe he was also Jewish. His name was Solomon Shereshevsky. Solomon Shereshevsky was somebody who did not forget, was able to remember everything.

And in the book, which Luria published later on, I think in 1968, which is called The Mind of a Mnemonist, he talks about what would it be like to literally not forget. And one of the things that always stuck with me, is that both in Borges’ story Funes the Memorious and in Luria’s, The Mind of the Mnemonist, they both highlight the same thing. And that is if you remember everything and you do not forget anything, it actually hinders conceptual thought.

Part of conceptual thought, part of being able to kind of group different categories together and form these broader concepts is hindered if you do not forget. If you do not forget, every single thing that you learn is kind of in its own category, and drawing those connections becomes nearly impossible. And in both the Minds of the Mnemonist and in Funes the Memorious, both subjects in the fictional story and in the nonfiction story, the Minds of the Mnemonist, the protagonist in each of those stories complains about the fact that they’re unable to actually generalize and create conceptual thought. In the words of Borges from Funes the Memorious he writes, “To think is to forget a difference, to generalize, to abstract. In the overly replete world of Funes there were nothing but details.” Nothing but details.

And the reason why I’m beginning with this is because I think in our contemporary world where we have Google at our fingertips, we now have AI at our fingertips. We’re able to offload so much of our memory, so much of what… We use brain capacity to remember where is this… And we offload so much of that. So what exactly are we left with? How has AI and the digitization of Torah learning affected the way that we relate to knowledge, and more specifically to Torah knowledge? I’m actually quite worried about this, I see this with my own students. I have found that the way that we compile and share Torah very often is just like, here’s 20 sources that more or less are saying the same thing. And we have these source sheets that get gobbled up in a lot of details. But I think a lot of the essential conceptual ideas of being able to bring in new creative ideas to Talmud Torah, to the study of Torah, has in many ways been hindered.

I’m trying to think in the last 50 years what incredible creative works of Torah have come out. And people have criticized me for bringing this up. I’ve brought this up online. I’m sure you could come up with examples for you that work. But frankly, I am worried that the digitization of Torah has actually hindered creativity. We should have seen a flowering of creativity. And instead, most of our published works just focus on gathering all the sources together. What’s become almost unnecessary with the advent of computers. And what really excites me are thinkers who are drawing new conceptual pathways, new doorways, new entry points into the engagement of Torah. And that is exactly what this episode is about, how the world of AI has affected, will affect the world of Torah.

How will our religious knowledge, namely Torah, the religious world view that we construct in our minds through sources, through the emergent in Torah, in Gemara, in all of the fine works that have come afterwards, how does not forgetting anything… So to speak, the world that all of us live in, we all have all of Torah at our fingertips. So where do we have room to grow, to build, to be creative? How do we make sure that we do not suffer the fate of Funes the Memorious, and just see details without that cadence and rhythm of conceptual thought? Because ultimately what makes us human more than anything, and this is an argument of Douglas Hofstadter in his fantastic book, Surfaces and Essences: Analogy as the Fuel and Fire of Thinking.

Douglas Hofstadter, which I mentioned in an earlier episode within this series, is the author of course, of Gödel, Escher, Bach. He won the Pulitzer Prize for that. His student, David Chalmers, is one of the forerunners of academic thought on consciousness. And Douglas Hofstadter, after writing Gödel, Escher, Bach wrote a phenomenal book called Surfaces and Essences: Analogy as the Fuel and Fire of Thinking, where he talks about conceptual thinking, drawing distinctions and parallels between different categories. Creating analogies in our own life to the things that we learn is actually the very engine, the very root of what makes us most human. The ability to draw those analogies in our creative thinking is actually at the very heart of what makes human knowledge distinctly human.

And in a world where we feel sometimes that we’re trying to keep up with the breadth of AI, the breadth of what computers offers, maybe our eyes should be diverted in something else. Maybe the pinnacle of human knowledge is something other than remembering everything, than knowing everything, than having everything at your fingertips, so to speak. But instead developing a cadence to your knowledge, learning how to integrate the knowledge that you have into your life, drawing analogies and building new doorways and entry points into existing knowledge. And I think in many ways, the opportunities of having so much information our fingertips only creates exciting possibilities for what Torah means, for what religious life means. So long as we get out of the rat race with computers and divert our attention to what makes us essentially human, and figuring out how to apply that, both to the world of Torah knowledge and of course to the world of Tefilah.

And that is why I’m so excited to share our conversation today with Dr. Moshe Koppel, who is a returning guest. Now, I want to surface something. When we announced this series, a few people, multiple people reached out and said, we have concerns that you are including Moshe Koppel in this series because of his involvement with the Kohelet policy forum and their role in the judicial reforms and all of the outbreak and protests that occurred in Israel because of good-hearted people, very decent people. On a personal level, I bristle at the notion that a disagreement, however serious, however real, however deeply entrenched in your personal ideology, should necessitate that we can no longer have a conversation, with few exceptions. You can think of a handful of exceptions of ideologies that I would no longer be able to have in a conversation.

But I know Dr. Koppel personally, I know he’s a decent person. And just the very notion that we can’t be in dialogue on a different topic because he was involved and took a stand on something that you don’t appreciate ideologically in the political realm, I really don’t like that idea, and I want… The people who reach out are very good people. But you should know people reach out in every guest that we have and say, “Well, they’ve said X, Y, and Z. Are you sure you want to have them?” Both people who we have on the right wing, on the left wing, on the political spectrum, on the religious spectrum. There is always someone whispering in my ear on whichever side and say, “I don’t know if you should really have this person. I don’t know if this is the right…” And I think that if we insist upon that ideological purity, we will all be stunted because of it.

I think part of the charm and beauty of what makes us human is our ability to engage and have the friction of ideas. But I did take the concern seriously. And the conversation I had with Dr. Koppel was actually recorded many, many months ago. But I reached out in advance of that conversation, parts of which were actually published in Jewish Actions quarterly magazine, the quarterly magazine of the OU. But before I released that conversation, and we get into that. I actually reached out to him more recently and spoke to him “Should we be concerned about your involvement? How do you want the listeners to think about that?” So let’s begin with my introductory conversation with Dr. Moshe Koppel.

So I am so excited to really begin my interview with Moshe Koppel, with an additional interview with Moshe Koppel. And the reason why I thought this was necessary, that was because in the months since we first recorded this interview, when you were on a trip, we had a lovely time together in Lower Manhattan. We were talking about developments in Israel, and AI, in computer science. I’ve always been enamored with your work. But since then, your name has been associated with things far beyond our conversations in AI, or our initial conversation that we had you on for 18Forty regarding Halacha, and that is your role in judicial reform.

And I wanted to begin, we’re not going to unpack what happened with the judicial reform. But for a lot of people who connected to your work regarding Halacha, regarding computer science. I’ll be blunt, I know people who I respect and I love, who were so deeply disappointed that they saw you getting involved in something that nearly, in their eyes, caused the civil war in Israel. And I wanted to kind of begin the interview with, what would you tell people who felt that disappointment? Maybe they were viewing it from too distant a lens? But they don’t associate you with that type of activity or politics. Can we separate your work? Should we be separating your work from the judicial reform? What would your response be to people who were hurt by what they saw your suggestions regarding the Israeli Judicial Reform and their larger relationship with your work and ideas?

Moshe Koppel: 
Okay, so first let me explain that I have a think tank in Israel called Kohelet, the Kohelet Policy Forum. We don’t do politics, we do policy. Policy means, well, what are the right policies for the state to have? We don’t get involved in the matter of political parties and what sacrifices they need to make in order to appeal to their base and get reelected or their coalition politics and things like that. That is not what we do. What we do is write policies. So in fact, the papers about judicial reform are papers that we wrote actually, some of them seven, eight years ago. We didn’t actually do anything lately. All that happened lately is that some politicians liked our work and used it as a basis for judicial reforms.

Now, policy is not the same thing as computer science or mathematics. In math prove a theorem and if you have a correct proof, well, it’s right. In computer science, sometimes you prove theorems, sometimes you… Does something. Policy isn’t like that. First of all, people disagree about what it is they’re trying to achieve. And secondly, they can disagree about how you can get there. So what we want to have in Israel is we want to have a vibrant democracy. A society in which people have all the basic freedoms, a Jewish country. And the question is, how do you reconcile all these things? What are the right… For example, in this case, what are the right checks and balances between the branches of government in order to achieve that?

So we wrote papers explaining what we thought were the problems with the Israeli system and how we thought adjustments could be made in order to get the right checks and balances. It doesn’t seem to me like the kind of thing where people should hate me for it. You could disagree with me or not disagree with me. But the idea that some people who opposed the particular policies that we advocated for. Or to be more precise, who opposed the judicial reforms that politicians brought that were based on the policies that we designed. Well, okay, you could disagree, but the fact that they kind of held me up as the demon in order to-

David Bashevkin: 
Moshe, allow me interrupt. I hope I’m not the first person who is coming and saying, oh, I know of people who-

Moshe Koppel: 
No, you’re not. I mean, there are people who tell me this to my face, so it’s not… I was at a wedding last week and some woman came over to me and said, “You’re a Nazi.” And then things got worse from there. I know that there are people who feel strongly about this, but… Me personally in a negative way because I… Haven’t contended with the argument.

I mean, there are plenty of people who understand these issues who came to me and said, “Well, we disagree with you for the following reason.” And said, “Well, we think you’re not going to achieve the goals that we both want to reach doing this.” And I said, fine, let’s see how we can make adjustments. And we discussed ways of making it better. It’s not like I think that everything we wrote in any position paper over the last 10 years in Kohelet is absolutely perfect, and nobody could disagree. But the fact that people read things in the newspaper and then hold me up as a demon. Or even worse, use me as part of their virtue signaling, well, that’s kind of disappointing. What can I say? I mean, it doesn’t increase my respect for them, but so be it. I have to live with it.

David Bashevkin: 
I’m just curious, just almost like friend to friend. I have a very hard time when my name is being used and I’m not in control of the narrative particularly. And this has happened to me fairly recently, when people look at work that I’ve done and they feel betrayed. Maybe it was an interview that took place in 18Forty, or something I wrote, or something I said. And they said, “We disagree so strongly that we feel a sense of betrayal.” And it has a very real personal effect on me because again, I’m a people pleaser. I love being liked. I guess it’s a human condition. I’m curious, just on a personal level, how did you cope when you were really a part of an international news story?

Moshe Koppel: 
Yeah, it’s not fun. As I said to you earlier, I said, I’d be delighted to have my anonymity back. But that’s not going to happen. What’s important to me is that the people that really are my friends and my family of course, and people that I have great respect for, they trust me. And even if they disagree with me, they’ll say, “Well, we disagree for this reason or for that reason.” And there’s no person in the world with whom I agree on every single topic, right. But it doesn’t have to be the end of the world if you disagree about something. The fact that you disagree with me about the exact composition of the judicial selection committee, and whether there should be six representatives of the coalition or five representatives of the coalition. I mean, okay, that doesn’t sound to me like the kind of thing that if we disagree about, we can never talk again for the rest of our lives, just we just can’t get past it. I think we can get past it.

David Bashevkin: 
And that is my hope as well, which is why, obviously we’re going to be sharing this interview that we’ve already shared, snippets of it in the pages of Jewish Action. I will continue to read your work. I agree with you. Very often, especially for people like myself, who want to be everything to everyone all at once, to paraphrase the title of that great recent movie. Being everything to everyone all at once is difficult, and it becomes nearly impossible to share ideas in the public square. There are people who criticize me for writing in this platform or in this place. It could be debilitating. And I hope that your ideas and whichever ones people are attracted to, people continue to give them attention and credence on the merits, on whether or not it resonates. But I have met with you, we have broken bread together. And I can attest to the fact that you are not, in fact, as far as I can tell, a demon. As far as I can tell.

Moshe Koppel: 
You know David, that’s how the demons work. They’re so clever.

David Bashevkin: 
They’re so clever. The greatest trick the devil ever pulled was convincing the world that he doesn’t exist.

It was so incredibly generous of him to jump on. I can vouch for the fact that as much as you disagree with him and as serious as those disagreements may be, he’s a person of decency, and integrity. And somebody who was thinking constantly about not only computer science, but specifically Torah. His book, which we discussed when we had him on last time during our Halacha series called The Meta Halacha, is absolutely fantastic. It actually talks about a lot of the principles of computer science, and thinking about the ways that we think about Halachic axiology, how we build the Halachic system. Is that the same as kind of programming a computer? Is that different? And factoring in our own humanity into the Halachic system plays a big role in what we talk about. Here is our original conversation recorded many months ago. I am delighted to share our conversation with Dr. Moshe Koppel. Moshe.

Moshe Koppel: 
Yeah, no, you have to start with, “It is my absolute pleasure.”

David Bashevkin: 
So you know-

Moshe Koppel: 
If you don’t say, “It’s my absolute pleasure,” I’m not talking.

David Bashevkin: 
It is my absolute pleasure. We’ve had a few returning guests before, but it really is my absolute pleasure to welcome you back in conversation to talk about the world of artificial intelligence. And I know when I think of artificial intelligence, what comes to mind? I’m always thinking about movies. I’m thinking about Space Odyssey. I’m thinking about robots coming to life and taking over earth. But I’m also thinking about AI. I’m thinking about writing on Gmail and when it’s able to fill in and know how my sentence is going to end. So I guess as a starting point, I want to understand and really build definitions. What exactly does it mean when we talk about the term ‘artificial intelligence?’

Moshe Koppel: 
Okay. So the truth is, it’s an unfortunate term. What happened was that in the early days of computers, back in the fifties. The kinds of things that computers were obviously designed to do, and do on a regular basis, is kind of boring things like sorting things in order, alphabetizing, or searching for something. You’re looking for a particular item in a long list, and that’s kind of the bread and butter of what computers do. Of course, they do it in more sophisticated ways. When you combine all these operations, they do a lot of important and interesting things.

And in 1956, there was a conference in Dartmouth, which is widely regarded as the founding of artificial intelligence. And the idea was that we were going to try to get computers to do, not the kinds of boring things that one imagines computers do, but rather the kinds of things that you associate with people doing. So the kinds of things they were talking about then were game playing. So could you get a computer to be a chess champion?

David Bashevkin: 
Yeah. And at that point, I remember it was considered almost unimaginable That a computer would be as good as a human being.

Moshe Koppel: 
Right. It was a big deal then. Exactly. Correct.

David Bashevkin: 
Why is there a difference between searching for an item, alphabetizing and playing a game? It’s interesting that at this conference, it’s like, could we get computers to play a game, that is tricky. What is it about games that would make it more difficult for a computer to master?

Moshe Koppel: 
Well, it’s not fundamentally different. But if you think about sorting things in order or searching for things, there are lots of ways to do it. You yourself can figure out what needs to be done. It’s just that it takes a long time. If you need to order-

David Bashevkin: 
It’s one principle. I know it goes A, B, C.

Moshe Koppel: 
Exactly.

David Bashevkin: 
It’s a whole… Yeah.

Moshe Koppel: 
Once you know the principle, then the interesting research was how can we do it most efficiently? But with the kinds of things like playing chess, you need to really think through how you could even get a computer to be good at chess. It’s not obvious what the right technique is.

David Bashevkin: 
This is 1956.

Moshe Koppel: 
This is in ’56. In ’56 there was a paper written by a guy named Samuels. He was a Jewish guy who had gotten the computer to play checkers fairly decently. That’s where it began. And as you know, eventually they got to chess, which is a much more complicated game.

David Bashevkin: 
I was a kid during Deep Blue.

Moshe Koppel: 
Right. So Deep Blue-

David Bashevkin: 
Beating Casper off. It was a whole-

Moshe Koppel: 
So at some point in the nineties, chess programs became better than grandmasters. Now are clearly the best in the world, certain programs for chess. But Go, even when computer became the world champion of chess, the Chinese game of Go was regarded as way more complicated than chess.

David Bashevkin: 
I’ve never heard of that game before.

Moshe Koppel: 
Okay, well, it’s… You remember Chinese checkers when you were a kid?

David Bashevkin: 
Sure. Yeah.

Moshe Koppel: 
It’s kind of like that. It’s a little bit more complicated. But that game was regarded as… No, we can do chess, but we’re never going to be a world champion at Go. And the reason they thought so was, you can calculate the number of possible moves. So in chess, it’s a zillion trillion moves. And Go, the number of possible moves was a zillion trillion times bigger than the number of possible moves in chess. So they thought, this is unconquerable, because really, what are you doing? You’re thinking about what are the possible moves that you can make? You’re trying to take a path down the different possible moves, and you’re trying to maximize your own strength and minimize your opponents’ strength.

David Bashevkin: 
Yeah. But bring it back to the foundational question, which is, is all of this artificial intelligence?

Moshe Koppel: 
Yes. The truth of the matter is, if you ask somebody, okay, where is boundary line between what we call artificial intelligence and what we call just computer programming, there’s not a big red line between them. Okay. So in a very colloquial sense, the kind of stuff that people do, if you can get computers to do them as well or better than people, that’s what they call artificial intelligence.

David Bashevkin: 
Do you see people using these chatbots?

Moshe Koppel: 
ChatGPT.

David Bashevkin: 
ChatGPT. And you can ask it to do a prompt, and there are people prompting it, write a sermon for the Parsha in the voice of this rabbi?

Moshe Koppel: 
Yes.

David Bashevkin: 
And they’ll really come up with something new and innovative and be able to imitate a voice.

Moshe Koppel: 
Just to clarify, for the readers, is when you say, voice, you mean in the writing style?

David Bashevkin: 
Exactly. It doesn’t sound like-

Moshe Koppel: 
It’s not spoken. You get back an output in text.

David Bashevkin: 
Correct.

Moshe Koppel: 
But you can actually say, give me something on the Parsha like in gangster rap?

David Bashevkin: 
Exactly. And they’ll really be able to do it.

Moshe Koppel: 
Which is… Yeah, it’s insane. But they actually can do it. Yeah.

David Bashevkin: 
So, when you see this now, what are the general trends that you see coming, in how AI is going to revolutionize the world of Torah study?

Moshe Koppel: 
Okay. There are many different areas in which it can do that. Let’s focus, first of all on things like text processing, how you handle rabbinic texts. Okay. Because one could imagine all kinds of virtual reality things for doing carbanote.

David Bashevkin: 
You can put on a headset and you’re in the Beitzah mezeh.

Moshe Koppel: 
I don’t want to talk. I want to talk more about what we can do with the kind of texts that people who are studying Torah usually are reading and how we can help them. And there are basically three different areas in which you would use artificial intelligence for this kind of purpose. The first one is just to make Torah more accessible to people. So this is not the kind of thing that an expert would need, but something that just makes it easier. So for example, you can take any cipher, any cipher at all. First of all, take a printed book and using what’s called OCR, optical character recognition-

David Bashevkin: 
Which people may know, anytime that you take a PDF and you turn it into a Word document, it’s using OCR technology, right?

Moshe Koppel: 
Correct. So the idea is if you take a picture of an old cipher, right. Now, it’s just a picture. You can’t search that, right? It’s just a picture. As far as the computer’s concerned, it’s black dots and white dots. All right? You can’t find a word that way. So the first thing you need to do is convert that picture of a page into a sequence of actual words so that the computer knows what the words are.

David Bashevkin: 
And especially an old print that gets harder and harder, it’s not as clear.

Moshe Koppel: 
Exactly. So that’s called OCR. And what’s being done right now, I can tell you this because it’s being done in my lab in Israel, which is called Dicta. We’re now able to take pictures of a book that was printed in the 19th century and some horrible Rashi font. It’s really hard to-

David Bashevkin: 
Close together, unclear.

Moshe Koppel: 
Yeah, exactly. Choppy and more spaces and less spaces and all kinds of weird stuff going on on the page. And old-fashioned letters sometimes, to OCR that to an accuracy of 99 point something percent.

David Bashevkin: 
And that’s already in existence? We’re there.

Moshe Koppel: 
We’re there, yes. And the next thing you could do, just talking about making texts more accessible to people. Is you could put the Nikud into a text, the vocalization. The Nikudot. It’s hard for some people to read a cipher that doesn’t have the Nikudot in it because words are ambiguous.

David Bashevkin: 
I remember when we first started learning Talmud, I think in fifth or sixth grade. I believe the top class, which I don’t want to brag, I was in in fifth or sixth grade. I think we had to draw in the Nikudots. But I think the other class had a special Gemara that only existed on a handful of chapters in the whole Talmud that had the Nikudots inside of the Talmud.

Moshe Koppel: 
Right. That’s pre-Stanza.

David Bashevkin: 
Yes.

Moshe Koppel: 
The idea is now for Stanzas to put in the Nikudots, that was a lot of effort, that was done manually. Somebody went through it and actually did it. But what you could do now, using some ideas from artificial intelligence is, you could take a new text that has never been seen by the computer. And without any manual intervention, just using automatic means, you can actually put in all the Nikudots.

David Bashevkin: 
Give me an example of a book you did this with a cipher.

Moshe Koppel: 
Okay, we’ve done it on hundreds of books, but I’ll just take one that’s particularly interesting. Because if you looked at it, you go, oh my God, this is taking my eyes out. Okay. If you take the Afkat Rochel. The Afkat Rochel, or the teshuvas of the Beis Yosef. The version of the Afkat Rochel that’s out there was… Actually, it was printed originally, I think in 1865 or something like that.

David Bashevkin: 
They did retype set it though. But that’s more recently-

Moshe Koppel: 
I don’t know if they retype set it. I have a version that came out a few decades ago, which is just a photo offset. It’s basically… They didn’t reset the book. Okay. Now there might be another more modern version-

David Bashevkin: 
But you took the old photo.

Moshe Koppel: 
We took the old one, scan it in. Just scanned the book in, and then OCR it. And the OCR is very accurate. By the way, we know more or less where there’s uncertainty.

David Bashevkin: 
No, there are certain words when you’re learning Gemara. Difference between Haka, Hasam, Hachi, you get certain Talmudic words mixed up. What does the computer get tripped up on?

Moshe Koppel: 
What the computer gets tripped up on is not similar concepts, but similar letters. It’s when a resh and a daled, hay and ches, they happen to be letters that look very similar. So you can do all kinds of tricks. And of course, we do all these tricks, which is… Well, if it’s a hey, then it’s a word that we have in our dictionary. We have a dictionary of all the words in the language. So if you’re not sure if it’s a hey or a hes well then you can check both ways.

David Bashevkin: 
The “you” in that sentence is the computer?

Moshe Koppel: 
The computer. The computer will actually look it up in a word list and give preference to a word that’s in the word list.

David Bashevkin: 
So, on the first level is OCR. That is a fascinating possibility. And that’s already… It’s not a possibility. That exists. It’s exciting to me my brain did not explode.

Moshe Koppel: 
Right. Your brain didn’t explode. Wait, so now the next thing you do is you put in the Nikudot.

David Bashevkin: 
Super helpful.

Moshe Koppel: 
Yes. That’s working. That’s working. We’ve got that running. We got it running. The next thing you do is you add punctuation. If you’re looking at the Afkat Rochel, there’s this book with… And there’s horrible old print. Well, there’s no punctuation in there. There’s these very long paragraphs.

David Bashevkin: 
There are very few that had any punctuation like-

Moshe Koppel: 
Exactly.

David Bashevkin: 
There’s just periods. There were… Yeah.

Moshe Koppel: 
There’s just periods and those are few and far between.

David Bashevkin: 
Yes.

Moshe Koppel: 
Right.

David Bashevkin: 
Lot of run-on sentences, if you’re a copy editor reading rabbinic literature.

Moshe Koppel: 
Absolutely. If you look at the Egris Mosha, he can write a five page Shuvah in a single sentence. It’s astonishing. Now, it’s really hard to do this. First of all, in the Nikud, one is right and one is wrong. I mean, there are rare instances where there are two possibilities and you’re uncertain which is actually right.

David Bashevkin: 
I’m actually intrigued, just for a moment. I know that it wasn’t real polemics. But there were always battles over how to pronounce certain, even masechtas, certain words. In Yeshiva, there’s a word that appears many times in the Talmud about an argument. So some people will pronounce it bmay kmifalgi

Moshe Koppel: 
kimifalgi.

David Bashevkin: 
Exactly. kimifalgi. Or there’s a masechta called Erchin. And it’s properly enunciated Archin. Or minachos, is probably minchaos correct?

Moshe Koppel: 
No, no.

David Bashevkin: 
That I made up.

Moshe Koppel: 
You made that one up.

David Bashevkin: 
I made the last one up. But I remember even Rosh Yeshiva getting annoyed About the smarty-pants students-

Moshe Koppel: 
Doing it correctly. Yes.

David Bashevkin: 
Doing it correct. Because a lot of that comes down to language, the rules for punctuation. Could they be overruled if the common spoken-

Moshe Koppel: 
Right. As you know, I’m a big fan of evolving language.

David Bashevkin: 
Yes. And it gets into that question, right?

Moshe Koppel: 
And that’s how people… Yes.

David Bashevkin: 
Descriptivist versus prescriptivist language.

Moshe Koppel: 
Correct. But for example, I have a guy on staff who is my official vocalizer. He’s the Nikud guy, right?

David Bashevkin: 
Yes.

Moshe Koppel: 
And he’s the guy who resolves like these questions. He’s a chassidishe guy, okay. But he wants to get it right.

David Bashevkin: 
He knows the rules.

Moshe Koppel: 
He knows the rules. But here’s the thing, he wants to do it right. Now he worked for another major publisher of Gemara with Nikkudot, okay. He said that he did the Nikud for there Gemara. And they said to him, no, no, no, you can’t do that. We don’t care about your fancy pants correct rules, okay? If the Olam says it like this incorrectly, we want you to vocalize it incorrectly, okay? They want it to be the way the Olam does it, right. For us, he actually did it according to the rules. I want to get the most accuracy I can squeeze out of the system where your gold standard is, whatever your gold standard is.

David Bashevkin: 
But take me to the next level now.

Moshe Koppel: 
Okay. First of all, on this accessible thing, so there’s Nikkudot, there’s punctuation, and there’s one more thing, which is opening abbreviations. If you see-

David Bashevkin: 
I love abbreviation games. Because there are ones that are very obscure.

Moshe Koppel: 
Some of them are really hard. Some of them-

David Bashevkin: 
An abbreviate, like an acronym. Which are all over rabbinic literature. Let’s give a very famous acronym. What’s a famous rabbinic acronym?

Moshe Koppel: 
kaf hay gimmel right? That could either be kahay gavna, or it could be kohen gadol. Exactly.

David Bashevkin: 
Right.

Moshe Koppel: 
There’s a very good example. And you need to be able to distinguish between them. Now we have this very advanced AI system that’s able to figure out, based on context, what is actually the correct opening up of an abbreviation. Now, once you’re able to do this, you can go back and look at other books and you can check whether they got it right or wrong. You can find amazing mistakes.

David Bashevkin: 
They got what wrong? There were other books they tried to open up-

Moshe Koppel: 
They opened up abbreviations, right.

David Bashevkin: 
And did it wrong.

Moshe Koppel: 
And did it wrong. So for example-

David Bashevkin: 
Did you ever catch mistakes?

Moshe Koppel: 
Yaakov avinu am haretz. Okay, that’s pretty bad. Can you figure out what it was?

David Bashevkin: 
Alav hashalom.

Moshe Koppel: 
Alav hashalom, right?

David Bashevkin: 
Was there really a work that published Yaakov Avinu am haaretz?

Moshe Koppel: 
Yes. I’m not going to tell you who, but yes.

David Bashevkin: 
That’s fantastic. So all of this is kind of setting the stage and making it more accessible, more easy, more approachable, the text.

Moshe Koppel: 
But now what about next scholars? Yeah. Okay, so scholars want something that’s actually, a scholar doesn’t need. The Nikkudot is-

David Bashevkin: 
Well, you’re presuming something when you say scholars want. What I’m curious, and I don’t want you to answer this now. First want you to tell me what this is about. I’m not sure scholars want this, because I don’t think scholars want to be replaced.

Moshe Koppel: 
Yes.

David Bashevkin: 
We’ll get there.

Moshe Koppel: 
You’re making a valid point.

David Bashevkin: 
What’s the next level?

Moshe Koppel: 
Okay, the next level is, I have a text, a book length text. I want to know every place where this book is actually based on an earlier source. I want to identify that source and I want to link to it. Okay.

David Bashevkin: 
Yes. This exists when you search in Bar-Ilan almost. No.

Moshe Koppel: 
No, it’s manually, they manually put in a lot of this stuff. And only like to Gemara’s or to psukim.

David Bashevkin: 
Correct.

Moshe Koppel: 
Now I’m talking about you’re learning the chiddushei haramban, and he has quoted a sifra or a sifri or a Gemara or a medrish or rishon that came before him or gaonim. He’s quoting tons of things. Every single time there’s a quote there that exists in an earlier source. I want to put in a footnote, and the footnote is going to say, this is a Gemara in psachim and you click on it and it opens up the Gemara in psachim for you. But it’s not just that. There are people who were quoting the Rambam subsequently. So if I’m learning a line in the Rambam, and I want to know every single achron, every single later commentator who has quoted this particular piece of the Rambam, right? I’m going to get it.

And not only am I going to show you for every single play, it could be 20 times on a page. Not only am I going to show you all of these things, but if you’re interested actually in knowing the different ways that Rambam has been quoted, how did people quote this line slightly differently from each other? Maybe Rav Moshe Feinstein brought it in a response when he quoted it one way. But if we look at the Minchas Yitzchak, he quoted it slightly differently. And I’ll actually put it in what’s called a synoptic form. Where you compare all the different versions and the differences are highlighted. So that’s something that’s of interest to scholars.

David Bashevkin: 
So it really shows you the depth of the text, everything that preceded it and everything that’s built upon it.

Moshe Koppel: 
And it’s built upon it. Exactly.

David Bashevkin: 
And that exists?

Moshe Koppel: 
And that exists as well.

David Bashevkin: 
Okay. Take me to the future.

Moshe Koppel: 
So now let me take you to the present, but that’s even more science fiction than this. Which is you’re learning some cipher of an actual physical book. You now take a picture of it with your phone. You open up the phone to illuminate.dicta.org dot aisle. You take a picture of it, your phone now tells you, oh, I’ll tell you what that is. What you’ve just seen in the book is the chiddueshei harashba on chulin daf daled amud aleph. And now let me give you the whole thing in digital form. You’ve got the book and this is what you’re learning now, but you want to get it all in digital form so that you can search it, so that you can at the push of a button add a Nikud or something.

David Bashevkin: 
This is like Shazam. There’s an app where if you listen to a song and just play like three seconds of a song, it’ll tell you what song you’re listening to and all the information related to it, who published it, et cetera. What albums it on. So this is Shazam for Torah.

Moshe Koppel: 
Exactly. Let me just tell you just one use case that my brother-in-law happened to mention to me the other day. You’re making a source sheet right. Now when you’re making a source sheet, if you have to use a picture from a book, it’s always awkward. Is it the font you want? And you could just take your phone now, take a picture of that, get it back in a digital form, and now you can change the font.

David Bashevkin: 
That’s so interesting.

Moshe Koppel: 
U can put stuff in, cut stuff out.

David Bashevkin: 
I want to tell you something, you’re not going to like this. But I’m not one of the leaders in AI, but I am one of the leaders in source sheet manufacturing. And there is a debate about do we like source sheets that has everything in a uniform font on the page? Or do you want the picture that has a little bit more of the personality and character? I’m sure it’ll come as no surprise that I actually like the picture. It’s a little messier. It’s not as clean. But I always say, I never want my source sheets to look like Bar-Ilan and a printer had a baby together. I want a little personality in there.

Moshe Koppel: 
Yeah, I know you want that personality. I mean, you’re a Surah Sajdah guy, probably too, right?

David Bashevkin: 
Yes. And I want you to know something about that. Surah Sajdah, meaning the way the page is laid out has been fairly consistent for many works. When the Mesivta Talmud, Oz Vehadar Mesivta Talmud changed the Surah Sajdah of the riff and the rhyme, I said, I’m out.

Moshe Koppel: 
You’re out.

David Bashevkin: 
I stopped buying them.

Moshe Koppel: 
It’s not just the Gemara for you. You want everything.

David Bashevkin: 
No, I was furious. Because I used to learn, when I was in Shiva, I learned gemara,rashi, tosfos, rif, rash ran, and I was like, what is going? Don’t mess with that tzuras hadaf, it means something to me.

Moshe Koppel: 
You rushed through the Ron and you ran through the rush. Beautiful.

David Bashevkin: 
Yes. That is where my traditionalism is most manifest. I like to think that. This is another area-

Moshe Koppel: 
But now let’s get to something even wilder. Okay. You’re familiar with search engines, so you’re studying a particular topic, you’ll put some keywords into a search engine like Bar-Ilan or Sefaria, and you’ll get a whole lot of material on that, right? But it’s not organized in any really-

David Bashevkin: 
No. There’s usually either chronologically or based on-

Moshe Koppel: 
It’s either chronological or it’s based on the quality of the match, which you may not care about. If you’re doing a topic. Now, you don’t really care whether every one of those words appears or just almost all of those words.

David Bashevkin: 
Because sometimes it’ll appear to text called kol kavua vchulay.

Moshe Koppel: 
Right. What you want is to find the central texts on the topic, right. So now the way that you would do something like that is, improve search engine technology. This is the basic thing that you would do. You would say, wait a minute, let me give you an idea. We haven’t quite implemented this yet, but it’ll give you the picture of how this is supposed to be done. Let’s it comes back that there’s a teshuva of the Rambam on this topic, and there’s a chiddushei harashba on this topic. And there’s some basic sources that you get back just doing a regular keyword search. But you know that they’re not totally typical. They may not be the best sources. They’re certainly not comprehensive, but they’re pretty good right.

Now, supposing that you discovered that of these sources that you found on a first pass, somebody like Rav Ovadia Yosef has quoted all three of those in the same paragraph, right? Okay, now what do you know. You know that Rav Ovadia is clearly writing about this topic here? Even if he never used the words kol kavua, but if he quoted these three sources, then this is a paragraph that’s about my topic, right? It’s nice you would add Rav Ovadia to your list. But here’s where the big benefit comes in. If he quoted those three sources in that paragraph, he probably quoted a dozen other sources in that same paragraph, that you had no chance of ever finding because they didn’t use your keywords.

You can now give the user all those 15 sources that Rav Ovadia quotes and be fairly confident that these are spot on because they’re curated, right? I mean, they were curated by Rav Ovadia. So you can really take search up to a whole other level once you’ve got all of the links. In other words, if I know that these three things have all been quoted by Rav Ovadia here, right? Because I’ve already developed this gigantic graph that tells me who quoted who. That’s what I was describing to you before. I know who’s quoting… Once, who’s quoting who, and you’ve got this big picture of whose quoting. You can play these kinds of games with the links on the graph, which are kind of Google like games, right? Google started this way. But in a more sophisticated way so that you can really get a picture of a sugya. That’s what we’re shooting for at the next stage.

David Bashevkin: 
The question that I think on a lot of people’s minds, and they do it when they’re playing with the chat box. And this is what I said the ambiguity of, do scholars really want this? Do you think a time is going to come in our lifetime where the role of the posek of the Halachaic decision maker is going to be replaced? Already now, I mean, there are some things I don’t need to call the rav of my shul to find out what bracha to make on a bowl of Cheerios. It’s already evolved. But there are things that are contingencies where, let’s say I want to know, is my knife still kosher? So there are a lot of follow-up questions that are involved in that. What did you slice with it? Or what were you washing it with? How hot was the water? Et cetera, et cetera. How many hours ago was it? Do you think in hour lifetime a computer is going to be able to be taught the rules of Halacha and be able to paskin sheilas?

Moshe Koppel: 
I would divide the world of sheilas into two parts. I mean, it’s really a continuum, but let’s talk about two parts. One is, okay, what brachado I make on a bowl of Cheerios? Or can I put a teabag into a kli sheini on Shabbos?

David Bashevkin: 
And this is the alphabetizing list. Those answers already exist in digital. And it’s just hard to find, and a computer makes it easier to find it. In our analogy, that was the first early computerizing, put things in alphabet order. Give me the answer to this because it’s too annoying to get off my tush and get it off the shelf.

Moshe Koppel: 
So short version is with the existing technology and search engines, that is not what’s existing in the labs. But the kind of stuff that’s out there, whether it’s Sefaria or Bar-Ilan or ozer chachma any of those, you would not be able to get the answer even to the most trivial question. You just wouldn’t, because you would put in a keyword like teabag, right? Or kli sheini or something-

David Bashevkin: 
And I’d get a thousand answers.

Moshe Koppel: 
And you get a thousand things and spend hours just kind of sorting through it-

David Bashevkin: 
For a very basic… The computer makes it more chaotic, not less.

Moshe Koppel: 
That’s not helping much. But it’s clear that not only in our lifetimes, but in short order, you’re going to be able to get answers to elementary questions. Anybody who’s played around with ChatGPT and seen how astonishing it is, right? Knows already at this point, you know could put in probably, I haven’t tried it. But you could try some very simple shavas there that are just a matter of knowing where to look in the books and you’ll probably get a fairly reasonable answer. I mean, I should warn you that ChatGPT is really, really good at giving very clear and coherent answers. It’s not very good at always giving true answers. That’s the thing you need to worry about. You would not want to trust ChatGPT with anything that’s really important, whether it’s Halacha or it’s anything that’s really important to your life.

ChatGPT is astonishing as a parlor trick, first of all, in the sense that… You know, you say please explain to me the idea of Nash equilibrium in game theory? And it will actually give you a more or less reasonable Wikipedia like answer to it. It’s amazing that it’s a text generator. It’s not a text copier. The paragraphs that it gives you don’t actually exist anywhere. It has actually created them. That is an amazing, amazing thing. But you wouldn’t want to rely on everything it says as being absolutely true. It tends to be way more accurate than you would’ve guessed is possible at this stage, right? But it’s not 100% reliable in terms of being truth, but we’re going to get there.

It is clearly the case that within a few years you’ll be able to ask the kind of elementary sheilas about which there’s not going to be any heated arguments among Rabbi. You’ll be able to get that. You’ll put in a question that says on shabbos, can I do this, that, or the other thing? Can I pour the hot water onto the cold water onto, and it’ll answer it fairly well. But the fact of the matter is that’s never what rabbanim were really for. Okay. I mean, what you really need rabbanim for are questions where-

David Bashevkin: 
Contingencies-

Moshe Koppel: 
The Rav needs to see the person in front of him and understand, right. It’s mostly like what I’d call shaas hadachak issues.

David Bashevkin: 
There’s a pressing circumstance with life situation.

Moshe Koppel: 
It could be a matter of health or even life and death. It could be a matter of… It would be very expensive for me to have to do this, right.

David Bashevkin: 
And it’s the difference if a millionaire’s asking this question. Or somebody who’s broke, and this is going to be the difference between-

Moshe Koppel: 
Correct.

David Bashevkin: 
Being able to send their kids to Yeshiva.

Moshe Koppel: 
People have issues within their marriage. Where say one of the couple has become less firm or more firm than the other one. And they need to resolve issues between them. It’s a matter of showing bias. And they need to know that maybe they could use a particular kulain order to save their marriage, right. You would not want ChatGPT answering that question.

David Bashevkin: 
Do you think there’s going to be a future where computers are going to be able to factor in or be able to explore people’s facial expressions or-

Moshe Koppel: 
Yeah. Oh, for sure.

David Bashevkin: 
You do think there’s going to be a time?

Moshe Koppel: 
Oh Absolutely. There’s no doubt.

David Bashevkin: 
So let me ask you this, given of what’s changing, do you think that the training and what the mastery that we insist upon in smicha curriculums, in rabbinic ordination curriculums, should that be evolving too, in your opinion?

Moshe Koppel: 
That’s a question that I haven’t really thought about. It may be the case-

David Bashevkin: 
Because in medicine, medical education is changing. They have to memorize a lot of stuff. But they’re so much more accessible on their phones, on databases.

Moshe Koppel: 
Correct. I think that it is important for people who are going to be paskening sheilas to have very, very broad and deep knowledge of Halacha, even if the material is easily accessible to them, and they could look it up just at the click of a button on their smartphone. I think it’s important because knowledge develops your intuitions, and a ravoften needs to pass it on the basis of intuition, because he’s not going to find the particular sheila he’s faced with in the shulchan aruch , in the mishnah brurah or in the igros moshe. He’s going to have to have strong intuitions for what’s right and what’s wrong. And the only way to develop strong intuitions is by having internalized a tremendous amount of information.

David Bashevkin: 
Let me rephrase the question. It’s a different question. I’m thinking of myself. I am not the biggest talmid chacham. I’m not, and I know that comes as surprise to you. I see your face. You look shocked. You look shocked. You look quite shocked.

Moshe Koppel: 
I think you’re just being modest. I think is false modesty.

David Bashevkin: 
I don’t know if the mic’s picked up. There was an audible gasp in the room. And I appreciate that, but I’m a vulnerable person. I’m willing to come clean and kind of really be honest right now. But I think that people have this sense, and I certainly have this, I don’t want to be replaced. I want to always feel like I am able to contribute to Torah, because I think a part of Torah is the human interpretation. The human partnership with the Torah’s Hashem, or like the Gemara in Kiddushin, of it begins as the Torah of Hashem uvatoraso. So your Torah yehgay yomam valayla darshened the passukim in Tehillim.

And what I’m wondering is what advice would you give for an emerging Yeshiva student, an emerging talmid chacham, who’s like, I see where things are headed. That massive breadth of knowledge is going to be so easily replaceable. Maybe not on Shabbos but who knows where things are going to be headed. People always say, I’m a big talmid chacham, except I’m an am haaretz on Shabbos. But what are the skills that are going to be most difficult to ever be replaced by a computer that we could be investing in?

Moshe Koppel: 
The answer is, as I said, it’s developing healthy intuition. You need to understand what’s going on behind the Halacha. What are the principles that underlie this whole big mass of Halacha’s, right? What moral ideas are there, right? What’s really going on, even if it’s not a particular moral idea, but what are the principles of shabbos that are going on in it?

And more importantly, you need to understand people. A real talmid chacham doesn’t treat somebody who’s, everything’s going fine for them and they’re wealthy and their family is happy. The same as he would treat somebody who’s down on their luck, who doesn’t have a parnasa who has shalom bayis problems. And they understand what each person needs. You can’t make a halacha’s okay? You can’t do away with halacha. But there’s a certain amount of give, and you need to know where is the give and where do you have to be rigid. And what does this particular person sitting in front of me need in order to be able to flourish and live a healthy Torah life? And what does some other person need, which might be very different, right.

David Bashevkin: 
Your program is called Dicta Maivin, and you could use some of these technologies on your phone. Can you download some of this stuff already?

Moshe Koppel: 
Sure. Just go to dicta.org.il and there’s a whole long list of products there, which you can-

David Bashevkin: 
Already use.

Moshe Koppel: 
You can use them. You can vocalize your text in there. You just put your text in. Yeah. As you’re typing in Hebrew, it’s actually putting in the Nikkudot.

David Bashevkin: 
People who have been listening to me in 18Forty know that I’m always interested in pushback, in concerns, in the heresy hunters or the people who are worried about the direction. Have you gotten any concerned people reaching out to you saying, I don’t like what you’re doing here. Or pushback saying, we got to draw a line. There’s so much work now being done in AI safety to make sure that we’re not literally replaced by computers. The work of people like Elizer ZYudkowsky who we’ve mentioned several times before. Have there been people in the Torah world who are concerned about the AI Torah safety?

Moshe Koppel: 
Interestingly enough, that hasn’t been the case. I’m not one to shy away from controversy. I’m always happy for a good argument. But the fact of the matter is that we cooperate all the stuff that we’re doing dicta, we cooperate with a lot of Haredi groups and with Sefaria of course, and with other organizations. We really haven’t gotten any pushback at all. So far nobody has told us that we’re really frightened by what you’re doing. For the most part, you’re just providing a service to people. If you don’t want to use it, don’t use it. But we haven’t presumed yet to try to replace poskim or something like that. So far, it’s all been good.

David Bashevkin: 
Let me ask you, on a personal level. Do you like learning digitally or do you-

Moshe Koppel: 
No, I never learn from a computer.

David Bashevkin: 
You never do?

Moshe Koppel: 
No, no. I open up old-fashioned Gemara and just learned from the Gemara or… Yeah.

David Bashevkin: 
The one question I didn’t ask, and maybe it should have been my first question, but instead it’s going to be my final question before I do rapid fire questions.

Moshe Koppel: 
I already told you when I wake up and go to sleep.

David Bashevkin: 
No, we’ll get there. No, no, no. But my question is this. I know you in the context, we’ve spoken before about Halacha. You’ve written books about Halacha. What on earth qualifies you to be talking about AI?

Moshe Koppel: 
I’ve been a professor of computer science for 42 years.

David Bashevkin: 
That’s a good answer. That’s a good answer. What is your field of study that brought you to this?

Moshe Koppel: 
Okay, so I began as a mathematician. My doctorate was on the interaction of mathematical logic and number theory. You can’t get more useless than that.

David Bashevkin: 
I happen to be fascinated by this. I am terrible at mathematics. I love the philosophical implications of mathematics. I’m fascinated by this stuff. It happens to be.

Moshe Koppel: 
Well, it’s a fascinating topic. So eventually I kind of slowly moved more in the direction of computer science, from mathematics to computer science. Within computer science, my area is an area called machine learning. Machine learning is a branch of artificial intelligence that has more or less taken over artificial intelligence.

David Bashevkin: 
Is there a difference between machine learning and artificial intelligence?

Moshe Koppel: 
Yes.

David Bashevkin: 
What is the difference?

Moshe Koppel: 
Artificial Intelligence is this broad area where you try to get computers to do the kinds of things that usually people do. Now machine learning is one method of artificial intelligence. The idea is, let’s say you want to get a computer to give medical diagnoses the way a doctor does. So in the eighties, the idea was you would speak to a lot of doctors and try to figure out what’s going on in their heads, and then you try to reproduce that as a computer program. It never worked. One of the reasons it never worked is, that a doctor kind of, you have a lot of experience. You kind of look at some-

David Bashevkin: 
It’s that intuition that you mentioned.

Moshe Koppel: 
You have intuition. And you go, obviously this guy has hepatitis, right? And then you say, well, why? And the guy just starts to mumble. I don’t know. I’m looking at him and I… Right? So machine learning is where you just get tons and tons of examples. So to take the medical diagnosis example, you would take lots of examples where you have a patient, you know what the patient’s symptoms are, and you know in the end what the correct diagnosis was. If you’ve got a million examples of this, now you could just reconstruct the rules. You don’t have to speak to the doctor, you don’t have to ask him how you got to the conclusion, because the computer will itself figure out how you get from these examples to the general rule. That’s the idea of machine learning. And that’s what I’ve been working in for the last 25 years or so.

David Bashevkin: 
And that’s been an exciting time the last 25 years?

Moshe Koppel: 
Oh, it’s been a very exciting time.

David Bashevkin: 
Give me one moment in your career, maybe it’s with the Torah, where you saw a computer do something and you gasped. And you said, wow, I did not think I would see this in my lifetime.

Moshe Koppel: 
Well, that happened this week with ChatGPT. Okay. I mean, I just put some questions into ChatGPT-

David Bashevkin: 
And you were shocked?

Moshe Koppel: 
And I was shocked. I mean, it’s jaw dropping. I mean, you literally put in a thing and say, tell me the halachas of kalay habishul on Shabbos and do it in the style of rap.

David Bashevkin: 
Or rhyme or whatever.

Moshe Koppel: 
Or rhyme or anything. It does it. That really is jaw dropping. There’s another thing that came out of a few months ago. There’s a few of these programs, Stable Diffusion and Dall-E too-

David Bashevkin: 
Sure. Artistic programming.

Moshe Koppel: 
You say I’d like to have a picture of, right? A man jumping over the moon wearing a purple cape in the style of Rembrandt.

David Bashevkin: 
Correct.

Moshe Koppel: 
And it actually produces a picture like that.

David Bashevkin: 
Man, if you ever visit my house on almost nearly every wall, are self-portrait of myself in the Rembrandt style.

Moshe Koppel: 
Yes. That’s what I would expect for a talmid chacham such as yourself.

David Bashevkin: 
In the living room. It’s everywhere. 25 years from now, what’s going to be? What’s next?

Moshe Koppel: 
I think that we’re going to be facing the kinds of Halacha questions that we really haven’t thought about a lot. If you consider the use of grama, right? Of indirect action on Shabbos, right. So it comes up because places like tzomet have wheelchairs and other things like that that they use grama as the basis for allowing certain things that people need for medical or other reasons. Now, there are very few examples of that now compared to what there’s going to be 25 years from now. 25 years from now, you are going to think that you would like your driverless car to pull up in front of your house and take you somewhere and it will do that. You’ll have cars coming exactly when you want them to come and taking you where you want.

Now, to put that more broadly, what you’re really doing over there is separating actions from consequences of actions. What we’re accustomed to now is if you want to get a certain consequence, you want to get yourself in your car from here to there, you need to take actions. And what’s forbidden on Shabbas as far as we know, is those actions, right. You can’t drive your car because of a combustion engine or whatever. But what happens when you can get the consequences? You could get you in your car from here to there without taking any of those actions just by thought or just by pre-programming something before Shabbas, right? So now I get the consequence without the action. Do we want that to be forbidden on Shabbas as well because it’s really these are the consequences we’re trying to avoid? Or is it no, it’s just purely a technical matter. If you are not taking this action, then it’s perfectly okay.

David Bashevkin: 
And this is something that poskim will absolutely-

Moshe Koppel: 
And people are going to have to deal with this because it’s going to appear in every aspect of our lives.

David Bashevkin: 
Yes. A lot of very, very exciting things coming up. I always conclude my interviews with more rapid fire questions. If somebody wants to understand more about the principles of artificial intelligence. A layperson, to understand what has developed, where are we headed? Is there a book you would recommend on this topic?

Moshe Koppel: 
I don’t think that books are the way to go. I think there are all kinds of instructional videos that you can find on YouTube.

David Bashevkin: 
Whose videos do you like the best?

Moshe Koppel: 
I don’t know the answer to that question. I don’t watch them.

David Bashevkin: 
You said books are not the answer.

Moshe Koppel: 
I think instructional videos… All the books I know are very, very technical.

David Bashevkin: 
You don’t like any of the popular books?

Moshe Koppel: 
I don’t read the pop stuff.

David Bashevkin: 
You don’t like the pop stuff? Okay. If somebody gave you a great deal of money and allowed you to take a sabbatical. You already have one PhD. To go back to school, get another PhD, write another book, what are the problems, issues that you would go back to study?

Moshe Koppel: 
You asked me this question the last time I was on, by the way. And I said to you probably something at the intersection of game theory and behavioral economics. That is still the case. I haven’t changed my mind since the last time you asked me this question.

David Bashevkin: 
That is absolutely fair. And I’m always curious, and we’ve asked you this before, but let’s see if it’s changed. What time do you go to sleep at night and what time do you wake up in the morning?

Moshe Koppel: 
Let’s just say that I go to sleep in the wee hours of the night. Okay. In single digits.

David Bashevkin: 
Single digits.

Moshe Koppel: 
It’s definitely single digits.

David Bashevkin: 
Same.

Moshe Koppel: 
Absolutely.

David Bashevkin: 
And what time do you wake up?

Moshe Koppel: 
I was going to say in double digits, but I’d be exaggerating a little bit-

David Bashevkin: 
But a little bit later.

Moshe Koppel: 
Let’s just say that the only place I could still catch a shaachris minyan would be chabad.

David Bashevkin: 
I am with you. That is the life I also like to lead. Professor Moshe Koppel, thank you so much for your time.

Moshe Koppel: 
Thanks a lot. It’s been a pleasure.

David Bashevkin: 
Since I had already reached out to Professor Koppel, just to kind of touch base and reintroduce him to our audience, given everything that has been going on in his life, particularly related to Israel. I figured it would also be worthwhile to ask him, what did he think about the developments that have taken place since we first spoke? Here is Professor Moshe Koppel discussing some of the developments that have taken place since our initial conversation.

We spoke when the first rendition, the first iteration of ChatGPT became public. We spoke about how that will affect the world of Torah learning, some of our religious sensibilities. My interest in this topic has only grown mostly because of the famed New York Times article with the early versions of Bing, where it seemed that the AI that Bing had presented was almost aware of itself and reached sentience. And since we have spoken, there have been other versions of ChatGPT, and the world is moving quick in this area. I’m curious, since our earlier conversation, have you adjusted any of your thoughts? Are there any trends or capabilities that you have seen AI exhibit that you did not necessarily anticipate?

Moshe Koppel: 
Three things have changed. The first one is, well, we were at GPT three than GPT 3.5, then GPT four. Now the main difference between them, you need to think of these models as being the extent between them, it’s humongous right now. Maybe when we talked, the models that were being trained were trained with, we call those node parameters. On order of the early models were like 10 billion parameters, which sounds like a lot, but then very quickly went up to a hundred billion parameters. And now the latest one that came out a couple of weeks ago, GPT four, they don’t say open AR. But it’s estimated that there’s about a trillion parameters in there.

Now, it keeps getting better. The ones that are at a hundred billion are way better than the models at a hundred billion, way better than that 10 billion. And GPT four at a trillion or so is way better than a hundred billion, in terms of the quality of the information it gives you, the quality of the answers. They have all these downstream tests that they do. Can you pass the bar exam? Can you pass some medical school tests? Can you pass an economics final? And they’re just getting better and better and better.

Now you need to understand something. Initially, in the early days of machine learning, one of the things that you worried about was when your models were too big, it means there were too many options. And there was always the fear, you’re training on examples that you were fed in advance and you’re going to use that to apply them to new examples. The problem was that if, think of it as your brain too big, or your model too complex, or whatever, that it would over fit to your training examples. So it would just form these crazy hypotheses of over fitting. So the assumption was if you get too big, the performance is going to tail off because it’s going to over fit to your training. And that hasn’t happened yet. They just keep getting bigger and bigger and bigger and it doesn’t flatten out. The learning doesn’t flatten out. So that’s an astonishing thing. The question is, when will it flatten out? Okay, there’s going to be a point you’re going to have diminishing returns on making this brain bigger, but that hasn’t happened yet.

So that’s the first thing that happened. So the models are bigger and better. The second thing is that the whole world discovered this ChatGPT thing, and they’re building apps on top of it. So you’ve got the basic ChatGPT, but now people are building all of these systems that incorporate ChatGPT as a component and they can do all kinds of things. You give it a book and it will mark up the book and summarize it for you and do all kinds of things. It will write a script for a movie and then make the movie, right. GPT is generating a script, which is then feeding into something like Stable Diffusion or one of those image creating things, and it’s creating images and you can create a series of images. And then you can generate sound using some text to voice or whatever, and people are doing wild things with it. So that’s the second. It’s happening really fast.

And the third thing that’s happening, and this is probably the most important to them, people appreciate it the least. People thought, as recently as a few months ago that the game was that open AI and Google and maybe a few small players like that were going to completely dominate this market because you needed tremendous computational power, tremendous resources in order to build a model like this, right. So ChatGPT ran for whatever, a year and a half or two years on these humongous networks of computers.

So only the big guys could do it. But it turns out a model was leaked, not the algorithm that needed to running it, but the actual model. What we do now is we use the model, but we don’t actually have the model. But Facebook Meta’s model was leaked out, so programmers were able to sit on top of it.

And it turns out that once you’ve got a model, you can now do all kinds of fine-tuning to it and just post-processing, where now the whole world has the full power of these models at their fingertips. It is no longer a monopoly or a duopoly of the big guys. Once you’ve got something to sit on and you can start fine-tuning it, just making incremental changes and stuff. People are taking this all over the place.

So for example, open AI did this whole processing, they call it reinforcement learning based on humans telling you this is an ethical answer and this is an unethical answer. And then they did this post process, and now if you ask some question to ChatGPT, that might lead it off into doing something harmful, like telling you how to design a nuclear weapon or something like that, it won’t do it, right. It recognizes these and it will not give you an answer that violates its rules. But everybody’s past that now. You can put in any rules you want in this post-processing phase. You can create an evil GPT. If we were doing this for psak halacha, you could do one that’s always makel.

David Bashevkin: 
Always lenient.

Moshe Koppel: 
Yeah, that’s lenient. You could create another one that’s always machmir, right? That’s always stringent, right?

David Bashevkin: 
Not the worst idea. Okay.

Moshe Koppel: 
Yeah, that would be fun. Okay, that would be fun. But you can imagine that’s harmless stuff. But you can imagine basically, it’s a nuclear weapon that’s out there and everybody is able to adapt it to their own purposes.

David Bashevkin: 
For some reason, and I don’t know why, the question of consciousness and sentience never resonated. I don’t know if it doesn’t attract you, if you find it boring or outside of your area of expertise. But I am curious, in that much shared New York Times article, where the reporter had a conversation with the Bing chatbot and chatbot shared its own inadequacy and fears and concerns and anxieties. Do you think that there is a notion of sentience that is taking place within these computers?

Moshe Koppel: 
No, there’s a fantastic simulation of it. That’s all. I don’t want to go into the weeds of philosopher John Searle’s Chinese room. But you know the experiment, right? He imagines-

David Bashevkin: 
Tell us.

Moshe Koppel: 
You get instructions, right. Somebody gives you Chinese cards and then you have instructions for which card to return, and you always give appropriate answers. But that doesn’t mean you understand Chinese, right. Well, you’ve given a computer appropriate instructions, right. It’s very complicated, but it has instructions for what to say when it gets a prompt. That doesn’t mean that it understands the way you understand. Now, the counterargument to this is, wait a minute, I, Moshe Koppel, do not know that you David Bashevkin are exactly that kind of a robot, right. I mean, I only know that you look like me and odds are that we’re all human beings and we were born the same way. And right, we’re in some kind of symmetric relationship, in that sense, which is different than our relationship with ChatGPT or a robot, in which it might be embedded. But theoretically, I mean I know as much about your consciousness as I do about some robot-

David Bashevkin: 
The computers, correct. I could be a zombie and have an empty interior world as Professor David Chalmers has us explored. You’re a hundred percent right. These questions have always resonated with me. You didn’t know it at the time, but the conversation that we’re introducing is part of a series that is actually on prayer, because to me the religious touring test that we’re involved in every day is the act of prayer. It’s what gets us in touch with our very humanity, what makes us most human. And I was wondering what do you think is the key to our humanity? What makes us most human? If we are being outpaced in our ability to play chess and complete the bar exam and pass the medical boards, where do you turn to, Moshe Koppel, when you try to get in touch with your humanity? What makes you human different than a computer? You’ll never be more efficient than a computer, I assume. So where now do you think we should be turning to really get in touch and access our humanity?

Moshe Koppel: 
David you asked me approximately the most difficult question in the history of mankind at the tail end of an addendum. Now we’re talking about AI. So thanks for that, David. Let me ask that question to you. Do you have an answer to it?

David Bashevkin: 
I mean, I’m talking about it in the course of this series. I genuinely believe our impulse to prayer is an act of getting in touch with our consciousness, which I’m a little bit of a pan psychism, where I believe that consciousness is the ether in which the entire world is formed. And I genuinely believe that when I daven when I close my eyes and try to get in touch with my own inner wants and desires. And just the difficulties and the opportunities of the world, that in and of itself is what I believe makes us most human.

That impulse, that inner impulse. I’m not saying the words of ashrei that you say before mincha I’m not saying the specific halachic formulation. But the pre halachic impulsive prayer that predates Sinai. That to me is where I return to, almost that like inner chauffeur, that unarticulated impulse and desire. I think our definition of humanity, rightfully so, needs to change. And I think this is a time of great opportunity for religious thinkers, religious ritual, religious act to kind of reclaim what it means to be human. Do you care to either push back or-

Moshe Koppel: 
That was a beautiful answer David. And I’m not going to try to go you one better.

David Bashevkin: 
Okay. I very much appreciate it. I’m going to let you off the hook for your friendship, for your kindness, and most of all for your Torah works. It really means a lot to count you as a mentor and a friend, and really appreciate your time today.

Moshe Koppel: 
I appreciate it very much. Happy to be here.

David Bashevkin: 
I hope in many ways that AI changes our relationship to Torah. I think that it can change our relationship to Torah in very good and very healthy ways. And I’ll list two of them and they both relate to things that I surfaced in an introduction. Returning back to that story of Funes the Memorious, and thinking about a world where imagine we remembered everything. Imagine you never forgot anything that you read, it was always at your fingertips. Imagine that you’re able to recall every line of Talmud that you studied, every word of Torah that you’ve ever read. It needs to be more. I’ve heard Rav schachter say many times that certain people, their big talmidei chachamim during the week, and they’re amei haaretz on Shabbas. And what he is jokingly referring to, is during the week they’re able to have computers and databases at their fingertips, which they’re able to search. And on Shabbas, when they’re unable to have that, they know nothing.

And obviously, I think even if we had a way to skirt around that. Even if we were able to remember and have everything at our fingertips on Shabbas, I kind of hope that AI changes our relationship to Torah. And I hope that it is in two ways. One of which in a world where we could remember everything, I think that it requires even more so the need to build new doorways, new analogies, new reference points into Torah. I’m not just talking about a hook, I’m talking about new ways that Torah is resonant in our lives. It reminds me of a Torah that I’ve quoted many times from Rav Tzadok, where Rav Tzadok says that the Gemara, the Talmud in Rosh Hashanah, which says yiftach bdoro kshmuel bdoro was one of the great judges in the Book of Judges.
yiftach and his generation was like Shmuel in his generation. And what Rav Tzadok says, is that yiftach is not just a person, it’s not just a noun, it is also a verb. Yiftach, someone who is able to be poseach, to open up new doorways into religious thinking, religious ideas, religious experiences, that is religious leadership. That is the Shmuel Hanavi, the great prophet of his generation.

And I think when it comes to Torah, building new entry points into Torah, building new connections into Torah, more creativity, more deep conceptual thinking. Thinking about how Torah informs the very interiority of our lives, I think it’s a new frontier. It’s something that I try to do. I don’t know that I do it successfully, but it’s definitely something that I try to do. The essays that I write, those of you who follow me outside of 18Forty know that I write essays on Talmud for tablet magazine. I’m not writing for the initiated. I’m trying to build new doorways, new entry points to think about tractate sotah, tractate yevamos, Tractate Chagiga.

Areas of Torah that were previously cordoned off with a kind of mechiza, a barrier of entry for somebody who maybe doesn’t have the right background. And be able to think about thematically, what is this about? How does this resonate in my life? I think there are really incredible opportunities. In whichever way I try in that way, and I see such incredible work happening in other areas. But instead of just kind of like, I don’t know, we have Bar-Ilan, we have Sefaria. There’s a great article by Zev Eleff that he wrote in the layer house, entitled Digital Discourse at the Democratization of Jewish Learning. Which really asks this question of, like if we’re able to remember everything and have things at our fingertips, then Talmud Torah should be more than just memory. It should be more than just trying not to forget. It should be building new doorways into our lives, into the lives of other people, to think about Torah in new ways. That’s number one.

I think number two brings us back to our very topic. And our larger topic, which wasn’t at the center of the conversation today, is the topic of prayer. And it is well known, and this is something that I discussed with Rav David Weinberg that Rebi Nachman used to turn his Torah into Tefillah. He used to literally make prayers out of his Torah ideas. And this is something that Rav David if you listen to that interview, said himself and did himself in his own work, a birth of the spoken word.

But I think it’s really so much more than that. What does it mean to take your Torah and turn it into a Tefillah? I think it means constructing Torah ideas that address the very existential questions of our lives. What Rav Tzadok talks about in Tzidkas Hatzadik in the 211th paragraph, Rav Tzadok writes that Torah and Tefillah prayer and Torah ideas, they need one another. Were supposed to daven] in the same place that we learn. And the way that Rav Tzadok talks about it is Torah is about creative production. Almost about leaning into our sense of self, our ego, our creative powers, our perspicacity. A word that I really only know from a Simpsons episode. Lisa says that,” I’m losing my perspicacity.” All of the things that make us unique and make us special. And Rav Tzadok writes, that Tefillah prayer, Davening does the exact opposite. It is about our vulnerabilities, our wants, our needs, the fact that we are inadequate and insecure.

And those two kind of link up together and one fulfills the other. If you just have Torah, you’re going to walk around feeling that sense of accomplishment, that sense of superiority, which the Talmud itself describes and warns us of, of walking around with that sense of entitlement and proudness that Torah accomplishment often leads to. While prayers about leaning into and focusing our needs, our vulnerabilities before God, before the melech malchei hamlachim, the king of all kings.

And I think that in a world where we can remember everything, what we really need to focus on is what we’re able to internalize. It’s incredibly cliche. It’s a story that so many people know and quote, but it’s a famous story where somebody approached the Kotzker Rebbe and said, I’ve accomplished, I finished all of Shaas. I’ve been through all of Shaas. Shaas, of course, referring to the corpus of Shisha sidrei. The the shisha sidrei of the Talmud. And he felt very accomplished. And the Kotzker famously responded back, you’ve been through Shaas but has Shaas been through you?

And it’s a great remark. It’s a classic line for the Kotzker. I’m sure many of our listeners have heard this. What does it mean? Has Torah been through you? I think the only avenue to create a world where Torah is able to penetrate the recesses of who we are, our religious character, our sense of self is by uniting the world of Torah and Tefillah. Having silent prayers in a moment where nobody else can see. Moments of real sincerity in our life where we ask for things, where we ask for ideas to be embodied in our lives. Where we ask Torah to seep into the recesses of who we are and actually transform our character. It’s that world where the chazon ish himself, Rav Yeshaya Korelis, in his second letter, he writes explicitly, halimud vhatfila keshurin zeh bazeh. Torah and Davening, Torah and prayer are integrated, are reliant, are dependent on one another.

And I think that dependency is about learning how to take our creativity, our accomplishments, and unite it with our vulnerability. To have the ideas that we produce that inspire, that nourish us, really address those inner vulnerabilities that we carry with us in the recesses of our heart. And that is the world where Torah and Tefillah, prayer and our Torah accomplishments are constantly in dialogue fulfilling one another. keshurin zeh bazeh, entirely reliant and dependent, informing and nourishing each of these respective worlds.

So thank you so much for listening. This episode, like so many of our episodes, was edited by our dearest friend Denah Emerson. It wouldn’t be a Jewish podcast without a little bit of Jewish guilt. So if you enjoyed this episode or any of our episodes, please subscribe, rate review, tell your friends about it. You can also donate at 18Forty.org/donate. It really helps us reach new listeners and continue putting out great content. You can also leave us a voicemail with feedback or questions that we may play on a future episode. That number is 9-1-7-7-2-0-5-6-2-9. Once again, that number is 9-1-7-7-2-0-5-6-2-9. If you’d like to learn more about this topic or some of the other great ones we’ve covered in the past, be sure to check out 18Forty.org. That’s the number one eight, followed by the word forty, f-o-r-t-y.org. You can also find videos, articles, and recommended readings. Thank you so much for listening and stay curious my friends.