Bloomberg Audio Studios, podcasts, radio news.

Bloomberg Audio Studios, podcasts, radio news.

You're listening to Bloomberg Business Week with Carol Messer and Tim Stenebeck on Bloomberg Radio. Look, I said we're going all in on AI. The author comes to our four o'clock hour today. Alphabet just reported we got that news just now, Caroline Hyde and man Deep Singh joining us Microsoft's GitHub agreeing to bank artificial intelligence models from Anthropic and Google into a coding assistant used by millions of software developers. And then there was that Wall Street Journal report earlier today Carol elon Musk's XAI is in talks to raise funding at a value of forty billion dollars.

You're listening to Bloomberg Business Week with Carol Messer and Tim Stenebeck on Bloomberg Radio. Look, I said we're going all in on AI. The author comes to our four o'clock hour today. Alphabet just reported we got that news just now, Caroline Hyde and man Deep Singh joining us Microsoft's GitHub agreeing to bank artificial intelligence models from Anthropic and Google into a coding assistant used by millions of software developers. And then there was that Wall Street Journal report earlier today Carol elon Musk's XAI is in talks to raise funding at a value of forty billion dollars.

Yeah, that's real money. So much of the reason we are talking about AI is because of Jeffrey Hinton, Nobel Prize winner in physics for work in AI, known as the godfather of AI. He was on the most recent episode of Wall Street Week with David weston.

Yeah, that's real money. So much of the reason we are talking about AI is because of Jeffrey Hinton, Nobel Prize winner in physics for work in AI, known as the godfather of AI. He was on the most recent episode of Wall Street Week with David weston.

These big eye systems, great sub goals. Now, the problem with that is if you give something the ability to create subcoals, it will quickly realize this one particular subcoal that's almost always useful. So if I have the goal of just getting more control over the world. That will help me achieve everything I want to achieve. So these things will realize that very quickly. Even if they've got no self interest. Then I understand that if I get more control, I'm it'll be better at doing what they want me to do, and so they will try and get more control. That's the beginning of a very slippery slope.

These big eye systems, great sub goals. Now, the problem with that is if you give something the ability to create subcoals, it will quickly realize this one particular subcoal that's almost always useful. So if I have the goal of just getting more control over the world. That will help me achieve everything I want to achieve. So these things will realize that very quickly. Even if they've got no self interest. Then I understand that if I get more control, I'm it'll be better at doing what they want me to do, and so they will try and get more control. That's the beginning of a very slippery slope.

A slippery slope. Indeed, perhaps our next guest work with Jeffrey Hinton developing the Boltzmann machine. It is the first learning algorithm for Mueller neural networks. And if you're scratching your et duh, you're going to get an explainer in just a moment.

A slippery slope. Indeed, perhaps our next guest work with Jeffrey Hinton developing the Boltzmann machine. It is the first learning algorithm for Mueller neural networks. And if you're scratching your et duh, you're going to get an explainer in just a moment.

We're very pleased to have a doctor Terrence Sanowski with us today's Francis Crick Chair at the Salk Institute for Biological Studies, Distinguished Professor at the University of California at San Diego. He's also president of the Neural Information Processing Systems Foundation. He joined us from San Diego. His new book out. It's called chat GPT and the Future of AI, The Deep Language Revolution. Terry, good to have you with us. Congratulations on the book. You have been thinking about this stuff for more than three decades. At this point, I think a lot of people have been talking about it and thinking about investing in it for two years. At this point, are you just saying every day, finally people are taking seriously what I've been looking at for a generation.

We're very pleased to have a doctor Terrence Sanowski with us today's Francis Crick Chair at the Salk Institute for Biological Studies, Distinguished Professor at the University of California at San Diego. He's also president of the Neural Information Processing Systems Foundation. He joined us from San Diego. His new book out. It's called chat GPT and the Future of AI, The Deep Language Revolution. Terry, good to have you with us. Congratulations on the book. You have been thinking about this stuff for more than three decades. At this point, I think a lot of people have been talking about it and thinking about investing in it for two years. At this point, are you just saying every day, finally people are taking seriously what I've been looking at for a generation.

Well, it's great to be here to have this opportunity. My book launched today, and I'm really excited about it. I think that what's happening right now is really unbelievable in terms of the breath and the depth and the excitement. And so I was there at the beginning. Jeffrey Hinton and I collaborated in the nineteen eighties. All the learning algorithms that are being used today for these large languages models and deep learning were developed by us back in that era. And of course what we didn't have back then were computers that were fast enough that could scale up these models to be you know, be able to solve these very difficult problems and artificial intelligence. But this you mentioned, the Neural Information Processing Systems Foundation on the President reorganized the biggest AI meeting and in December, we're expecting sixteen thousand researchers to descend on Vancouver. After by the way Taylor Swift, she's the big headliner on Sunday, but we have the rest of the week.

Well, it's great to be here to have this opportunity. My book launched today, and I'm really excited about it. I think that what's happening right now is really unbelievable in terms of the breath and the depth and the excitement. And so I was there at the beginning. Jeffrey Hinton and I collaborated in the nineteen eighties. All the learning algorithms that are being used today for these large languages models and deep learning were developed by us back in that era. And of course what we didn't have back then were computers that were fast enough that could scale up these models to be you know, be able to solve these very difficult problems and artificial intelligence. But this you mentioned, the Neural Information Processing Systems Foundation on the President reorganized the biggest AI meeting and in December, we're expecting sixteen thousand researchers to descend on Vancouver. After by the way Taylor Swift, she's the big headliner on Sunday, but we have the rest of the week.

So I want to ask you, how did you think about neural networks in large language models in the nineteen eighties. How do you think about them today.

So I want to ask you, how did you think about neural networks in large language models in the nineteen eighties. How do you think about them today.

Well, we actually had a premonition that these large language models, I should say not the large they were small language models. Back we're really good at language. And that was a particular project, a summer project for a gratitude in my lap called net talk, which was trained on a dictionary to be able to pronounce English text. You know, if you give it an article from the Wall Street Journal and they would pronounce it in an understandable way. And this in linguistics is a very difficult problem because English is so irregular. There are a lot of regularities, but you also have irregularities, and then you have rules for the irregularities. But it really was amazing that a small, tiny network with just a few hundred units and tens of thousands of weights, the parameters, the connections between the units could do that. It was like an amazing compression of complexity. And now we know that these large language models, the deep learning networks, they love language, and they are capable of things that we never could have imagined.

Well, we actually had a premonition that these large language models, I should say not the large they were small language models. Back we're really good at language. And that was a particular project, a summer project for a gratitude in my lap called net talk, which was trained on a dictionary to be able to pronounce English text. You know, if you give it an article from the Wall Street Journal and they would pronounce it in an understandable way. And this in linguistics is a very difficult problem because English is so irregular. There are a lot of regularities, but you also have irregularities, and then you have rules for the irregularities. But it really was amazing that a small, tiny network with just a few hundred units and tens of thousands of weights, the parameters, the connections between the units could do that. It was like an amazing compression of complexity. And now we know that these large language models, the deep learning networks, they love language, and they are capable of things that we never could have imagined.

Well, we're gonna we have a few minutes with you now, and then we're going to come back and have even more time with you. But that's really what I wanted to talk about the idea of super intelligent AI. What are we not thinking about? What's the threat out there?

Well, we're gonna we have a few minutes with you now, and then we're going to come back and have even more time with you. But that's really what I wanted to talk about the idea of super intelligent AI. What are we not thinking about? What's the threat out there?

So you know, my good friend Jeff is very concerned, and I think he's one of the smartest people I've ever met. And if he's worried about it, then there's some as a concern. However, I think that even if you're concerned, it's very difficult to know when that's going to happen, if it ever will happen. And there are super forecasters out there, and this is from the Economist magazine, who are much better at people who are experts at predicting you know, if and when there may be a catastrophic or existential threat, and it turns out that in fact, they're not as a concerned as the experts in AI. I'm happy that someone is thinking about the worst case outcome because if not, then if it ever happens, we're in trouble. But right now, I'm more concerned about trying to understand how they work mathematically and also to learn more from the brain. After all these were designed. Back then, Jeff and I looked at the brain the only existence proof you could solve any problem in AI. And you know, we tried to build something that was based on similar principles. So now we can continue. There's a lot more in the brain we can learn from.

So you know, my good friend Jeff is very concerned, and I think he's one of the smartest people I've ever met. And if he's worried about it, then there's some as a concern. However, I think that even if you're concerned, it's very difficult to know when that's going to happen, if it ever will happen. And there are super forecasters out there, and this is from the Economist magazine, who are much better at people who are experts at predicting you know, if and when there may be a catastrophic or existential threat, and it turns out that in fact, they're not as a concerned as the experts in AI. I'm happy that someone is thinking about the worst case outcome because if not, then if it ever happens, we're in trouble. But right now, I'm more concerned about trying to understand how they work mathematically and also to learn more from the brain. After all these were designed. Back then, Jeff and I looked at the brain the only existence proof you could solve any problem in AI. And you know, we tried to build something that was based on similar principles. So now we can continue. There's a lot more in the brain we can learn from.

But paint that picture for us, because I think a lot of people are worried about doomsday scenarios here, and if Jeffrey Hinton is worried about that stuff, I mean, should we we should be worried about it.

But paint that picture for us, because I think a lot of people are worried about doomsday scenarios here, and if Jeffrey Hinton is worried about that stuff, I mean, should we we should be worried about it.

You're saying, I think that we should be cautious, that is to say, we should be uh constantly thinking along the lines that Jeff is in terms of what could possibly happen, and you know, be cautious and put in precautions so that it can't happen.

You're saying, I think that we should be cautious, that is to say, we should be uh constantly thinking along the lines that Jeff is in terms of what could possibly happen, and you know, be cautious and put in precautions so that it can't happen.

What sorry, I just yeah, go ahead.

What sorry, I just yeah, go ahead.

What I'm really concerned about are the unintended consequences, things that you cannot predict. Something may happen that you know, no one thought of, even Jeff.

What I'm really concerned about are the unintended consequences, things that you cannot predict. Something may happen that you know, no one thought of, even Jeff.

Yeah, and like you know, we have learned certainly right great financial crisis pandemic like the un the unthinkable can happen and you throw technology into it and you just kind of don't know where it's gonna go exactly. Okay, so now I'm terrified. Okay, Terrence Stoke go anywhere. We're gonna do some news. I do wonder what.

Yeah, and like you know, we have learned certainly right great financial crisis pandemic like the un the unthinkable can happen and you throw technology into it and you just kind of don't know where it's gonna go exactly. Okay, so now I'm terrified. Okay, Terrence Stoke go anywhere. We're gonna do some news. I do wonder what.

Yeah, I think we have time for one more question before we go.

Yeah, I think we have time for one more question before we go.

Well, so you know, okay, we have a minute and then we're gonna take a break and come back. But I just do wonder. You know, when you talk to people, do you say, wait, this is really going to be net net a good thing?

Well, so you know, okay, we have a minute and then we're gonna take a break and come back. But I just do wonder. You know, when you talk to people, do you say, wait, this is really going to be net net a good thing?

Look, all new technologies have good and bad consequences, and you try to mitigate the bad, and you you have to balance them, you know. Yeah, and right now it looks like the good is way way ahead of the bad in terms of the impact it may have on us and society and businesses. But you know, like I say, we have to be careful because we don't really know where it's heading.

Look, all new technologies have good and bad consequences, and you try to mitigate the bad, and you you have to balance them, you know. Yeah, and right now it looks like the good is way way ahead of the bad in terms of the impact it may have on us and society and businesses. But you know, like I say, we have to be careful because we don't really know where it's heading.

You know what worries me too, And we will talk about this maye when we come back. I feel like we throw around a lot of words, not you, but all of us in the general like you know, whether it's you know, AI, generative AI, and like you know, and I don't know that we really understand what's going on, and so it's hard to know what it could possibly become. So we're going to pick your brain a little bit more Terry when we come back.

You know what worries me too, And we will talk about this maye when we come back. I feel like we throw around a lot of words, not you, but all of us in the general like you know, whether it's you know, AI, generative AI, and like you know, and I don't know that we really understand what's going on, and so it's hard to know what it could possibly become. So we're going to pick your brain a little bit more Terry when we come back.

It's Terry Sanowski. He's a Francis Kirk Charity Sauk Institute for Biological Studies, Distinguished Professor at the University of California at San Diego, President of the Neural Information Processing Systems Foundation. The new book out now, Chat GPT in the Future of AI The Deep Language Revolution More with doctor Sanowski.

It's Terry Sanowski. He's a Francis Kirk Charity Sauk Institute for Biological Studies, Distinguished Professor at the University of California at San Diego, President of the Neural Information Processing Systems Foundation. The new book out now, Chat GPT in the Future of AI The Deep Language Revolution More with doctor Sanowski.

Right after this, I want to get back to our guests. We're talking with doctor Terrence Sanowski. He is Francis Crick Chair at the Salk Institute for Biological Studies, Distinguished Professor at the University of California at San Diego. He's also president of the Neural Information Processing Systems Foundation, and he's joining us from San Diego on this Tuesday. His new book Chat CHEPT in the Future of AI, The Deep Language Revolution. I got to ask you because I am still trying to understand and I get worried that we throw these words around. Certainly not you, but we as we try to understand this with now, you know, not having full comprehension of what art artificial intelligence, the large language models that we're talking about today, where it takes us. Is it as subtle at evolving in life changing as the Internet was for us.

Right after this, I want to get back to our guests. We're talking with doctor Terrence Sanowski. He is Francis Crick Chair at the Salk Institute for Biological Studies, Distinguished Professor at the University of California at San Diego. He's also president of the Neural Information Processing Systems Foundation, and he's joining us from San Diego on this Tuesday. His new book Chat CHEPT in the Future of AI, The Deep Language Revolution. I got to ask you because I am still trying to understand and I get worried that we throw these words around. Certainly not you, but we as we try to understand this with now, you know, not having full comprehension of what art artificial intelligence, the large language models that we're talking about today, where it takes us. Is it as subtle at evolving in life changing as the Internet was for us.

So this is something that is emerging. And I have since the book was sent to press in the summer, I have a sub stack where I have tried to fill in with, you know, the new things that are happening. And I'm preparing something a new twelfth version the blog on the question of whether AI is overhyped or under hyped, and and you know, I've thought a lot about this, and you know, I think that it depends on the timescale. I think that on the short timescale it is overhyped. There's no doubt about it. There's just so much out there. I mean, every day the newspapers are filled with AI and your program. But I think in the long run it's actually under hyped. I think the real change in the Internet, for example, didn't occur within the first ten years. It was much later. Again, unintended applications that merge that you know, have enormous impact on our lives, like social media, So I think the same thing's going to happen with AI.

So this is something that is emerging. And I have since the book was sent to press in the summer, I have a sub stack where I have tried to fill in with, you know, the new things that are happening. And I'm preparing something a new twelfth version the blog on the question of whether AI is overhyped or under hyped, and and you know, I've thought a lot about this, and you know, I think that it depends on the timescale. I think that on the short timescale it is overhyped. There's no doubt about it. There's just so much out there. I mean, every day the newspapers are filled with AI and your program. But I think in the long run it's actually under hyped. I think the real change in the Internet, for example, didn't occur within the first ten years. It was much later. Again, unintended applications that merge that you know, have enormous impact on our lives, like social media, So I think the same thing's going to happen with AI.

But is it is it different? Like I guess, I'm like, what do you mean, Like, the Internet is not I wanted to say comfortable, but it's not because there's some really bad things that happen and we know that, right, and that's the battle we have with social media. And we want to talk to you about kind of regulatory oversight of AI in a moment. But I just I'm just trying to understand. You know, it does feel so seamless and just such a part of everything we do. But it hasn't necessarily replaced a ton of jobs. It's created jobs, It's replaced some jobs. I guess you could say, I'm just trying to understand, Like on what scale do you put it? You mentioned the internet, So is it apples to apples or is it something else?

But is it is it different? Like I guess, I'm like, what do you mean, Like, the Internet is not I wanted to say comfortable, but it's not because there's some really bad things that happen and we know that, right, and that's the battle we have with social media. And we want to talk to you about kind of regulatory oversight of AI in a moment. But I just I'm just trying to understand. You know, it does feel so seamless and just such a part of everything we do. But it hasn't necessarily replaced a ton of jobs. It's created jobs, It's replaced some jobs. I guess you could say, I'm just trying to understand, Like on what scale do you put it? You mentioned the internet, So is it apples to apples or is it something else?

No, Well, first of all, it'll it uses the internet, So I mean that's like the the machinery that you need to reach to scale up and reach a large population. But it's more intimate than the Internet in the following sense that it talks to us, right, I mean It's as if an alien landed on the Earth and could talk to us in English, and it knew everything about you know, what, humans, history, everything, and the only thing we can be sure if it's not human. But it's really quite remarkable. Let me give you one example of something that I was really surprised at when they did a study of whether people who needed cognitive therapy preferred real humans or AI. They preferred AI, which was really quite remarkable. I didn't expect that. And part of the reason is that the AI is not judgmental like humans.

No, Well, first of all, it'll it uses the internet, So I mean that's like the the machinery that you need to reach to scale up and reach a large population. But it's more intimate than the Internet in the following sense that it talks to us, right, I mean It's as if an alien landed on the Earth and could talk to us in English, and it knew everything about you know, what, humans, history, everything, and the only thing we can be sure if it's not human. But it's really quite remarkable. Let me give you one example of something that I was really surprised at when they did a study of whether people who needed cognitive therapy preferred real humans or AI. They preferred AI, which was really quite remarkable. I didn't expect that. And part of the reason is that the AI is not judgmental like humans.

Well wait, but isn't it depends on the data, Like we talk about, it wasn't.

Well wait, but isn't it depends on the data, Like we talk about, it wasn't.

Getting trained on judgmental data.

Getting trained on judgmental data.

It was you know, Actually, it's a good question. What was it trained on. I think that it was fine tuned with you know, data from real subjects that we're that we're talking with a doctor. But even without that, I'll tell you something again, it's shocking is that it is actually empathy. These large language models can empathize with humans. And why is it? How is that? Well, it actually absorbed a lot of text out there, novels, you know, letters, and read it and so forth, and where empathy was being part of the discussion, and so it absorbed that too well.

It was you know, Actually, it's a good question. What was it trained on. I think that it was fine tuned with you know, data from real subjects that we're that we're talking with a doctor. But even without that, I'll tell you something again, it's shocking is that it is actually empathy. These large language models can empathize with humans. And why is it? How is that? Well, it actually absorbed a lot of text out there, novels, you know, letters, and read it and so forth, and where empathy was being part of the discussion, and so it absorbed that too well.

I wanted to hear a little bit of your thoughts on what we heard from Elon Musk a little earlier today. He actually participated in a surprise conversation at the Future Investment Initiative to discuss the future of AI.

I wanted to hear a little bit of your thoughts on what we heard from Elon Musk a little earlier today. He actually participated in a surprise conversation at the Future Investment Initiative to discuss the future of AI.

It's most likely going to be great, and there's this some chance which could be tense, that it goes bad. The chances on zero that it goes bad. But overall, at one point you said that covers eighty percent full is one positive way to look at it.

It's most likely going to be great, and there's this some chance which could be tense, that it goes bad. The chances on zero that it goes bad. But overall, at one point you said that covers eighty percent full is one positive way to look at it.

Maybe ninety percent, okay, eighty or ninety percent positive. The question I have for you, professor, is do we need an international regulatory body? Do we need the largest, most powerful governments around the world to create some sort of standard to ensure or help ensure that this goes the right way.

Maybe ninety percent, okay, eighty or ninety percent positive. The question I have for you, professor, is do we need an international regulatory body? Do we need the largest, most powerful governments around the world to create some sort of standard to ensure or help ensure that this goes the right way.

Well, as you know, in the UK they passed an AI Act which is like one hundred pages long and you know, incredibly detailed, and it's already absolute. I just moving blasting forward and you know you're trying to catch up with it. But I do believe that it's absolutely essential that it be regulated, and it should be regulated by people who a building it. The government, okay, is the business of protecting people. And we'll see how that plays out. But in for example, genetics, this happened, you know, back in the sixties seventies. They had a meeting where they came together at a solomar and they came up with a containment rules and regulations for doing experiments under the careful protection so that nothing leaks out, nothing gets out. And I think we need to do the same.

Well, as you know, in the UK they passed an AI Act which is like one hundred pages long and you know, incredibly detailed, and it's already absolute. I just moving blasting forward and you know you're trying to catch up with it. But I do believe that it's absolutely essential that it be regulated, and it should be regulated by people who a building it. The government, okay, is the business of protecting people. And we'll see how that plays out. But in for example, genetics, this happened, you know, back in the sixties seventies. They had a meeting where they came together at a solomar and they came up with a containment rules and regulations for doing experiments under the careful protection so that nothing leaks out, nothing gets out. And I think we need to do the same.

Okay, So when does as you said, ten years for the internet to really kind of make its impact and presence really known and maybe you know, integrated into our lives. So is it a decade before we see LMS and AI at this level integrated into our lives.

Okay, So when does as you said, ten years for the internet to really kind of make its impact and presence really known and maybe you know, integrated into our lives. So is it a decade before we see LMS and AI at this level integrated into our lives.

We are at a stage that aviation was at when at Kitty Hawk the Wright brothers made the first flight. It was ten feet up and one hundred feet long, and that really was the you know, something that then took decades and decades to build. And the most difficult thing, by the way that airplanes, you know, design of airplanes had to solve was control. How do you how do you make it go where you want to go without crashing and that's something that again it's like we're going through right now with AI. And yes, it will take decades. It's not going to happen overnight.

We are at a stage that aviation was at when at Kitty Hawk the Wright brothers made the first flight. It was ten feet up and one hundred feet long, and that really was the you know, something that then took decades and decades to build. And the most difficult thing, by the way that airplanes, you know, design of airplanes had to solve was control. How do you how do you make it go where you want to go without crashing and that's something that again it's like we're going through right now with AI. And yes, it will take decades. It's not going to happen overnight.

I know.

I know.

It's just it's kind of fascinating. We have a million more questions. Is it going to take all the jobs? Is it going to is it going to create jobs? Is it going to take jobs? Just got thirty seconds and yes.

It's just it's kind of fascinating. We have a million more questions. Is it going to take all the jobs? Is it going to is it going to create jobs? Is it going to take jobs? Just got thirty seconds and yes.

Yes, you know, I get asked that question whenever I give a public talk, and my answer is that you won't lose your job, but your job's going to change and you're going to need new skills, and you know it will morph over time. Now, you know these are people who are in the workforce now, but you know young people coming up, they'll have no trouble whatsoever finding new jobs in this new industry.

Yes, you know, I get asked that question whenever I give a public talk, and my answer is that you won't lose your job, but your job's going to change and you're going to need new skills, and you know it will morph over time. Now, you know these are people who are in the workforce now, but you know young people coming up, they'll have no trouble whatsoever finding new jobs in this new industry.

We gotta get you back on the show, doctor Sanowski. Really appreciate you.

We gotta get you back on the show, doctor Sanowski. Really appreciate you.

Just say, Terry, what we always want to know is, like you know, people like us, you know anchors.

Just say, Terry, what we always want to know is, like you know, people like us, you know anchors.

We don't have time for him to answer that question. We don't have time for him to answer that question. Carol, I'm sorry, I got to give the book a play.

We don't have time for him to answer that question. We don't have time for him to answer that question. Carol, I'm sorry, I got to give the book a play.

Well, come back hopefully.

Well, come back hopefully.

The new book Chat GPT in the future of AI, the Deep Language Revolution, Doctor Terry Sanowski. He was there at the beginning. He knows it all. He's Francis Krickchair at the Salk Constitute for Biological Studies, among many other things. This is Bloomberg BusinessWeek.

The new book Chat GPT in the future of AI, the Deep Language Revolution, Doctor Terry Sanowski. He was there at the beginning. He knows it all. He's Francis Krickchair at the Salk Constitute for Biological Studies, among many other things. This is Bloomberg BusinessWeek.