Artificial Intelligence -— Groks Science Show 2023-08-23

I'm Frank Ling.

And I'm Charles Lee.

And you're listening to the GROK Science Show.

That's right. It's a weekly look world of science, technology and their effects on our daily lives. Coming up on today's program, Kevin Serrace will join us to discuss generative AI.

So stay tuned for all this, plus the Grokotron 5000 and our world famous question of the week coming right up here on the GROKS Science Show.

Foreign.

Welcome back to the GROX Science Show. Well, remarkable transformations are taking place with the introduction of AI. Who hasn't heard of ChatGPT and what does it mean for us today? Joining us today to discuss this issue is Mr. Kevin Serrace. Mr. Surice is a Silicon Valley innovator, serial entrepreneur, CEO, TV personality and edutainer. He has been featured by sources such as BusinessWeek, Time, Fortune. He has led pioneering work on the first cellular data, smartphone and generative AI. And he joins us today to discuss the remarkable innovations taking place in the field of AI. Mr. Surese, thank you so much for joining us today.

Yes, so happy to be here.

Certainly a pleasure. Certainly a topic that everybody is concerned with. What is going on in the field of AI these days?

Well, first of all, AI has been with us and we've been making advances for 70 years, literally since, since the 1950s. Right. And we've worked very hard and scientists have worked most mostly mathematicians. Believe it or not, AI is mostly a math field that people don't, don't realize math and then coding to encode that math. But we had worked on solving lots and lots of different issues. And of course over the years we've solved issues in math and science and all kinds of things. And only recently, over the last few years, language has gotten to a point where we can actually interact with something. Now I invented, me and my team years ago invented the first AI virtual assistants. Eventually all of that technology went on to make PM OnStar and went on to make Siri and Google Assistant and Cortana and all the others, right? Alexa, et cetera. So we worked with this human language interface very, very hard. And it was a large team. This is in the 90s, probably two dozen of my, almost a hundred patents in that field doing that kind of work. And what's happened since then is we've gotten better and better and further away from rules, engines overall and more to this transformer based technology that is really human like but not always accurate. But I will, you know, my friend Elon and others are trying to tell people out there, that AI is going to kill us all. Look, this is a large language model. It's no different than a math model that we would use in Excel. Okay? And Excel is going to kill us, made us more productive. Large language models like ChatGPT are making us more productive in certain fields. Do not hook the darn thing to the nuclear arsenal and we will be fine.

Concern the amount of information that's generated by generative AI for languages and do you think that there are the proper safeguards in place for its implementation to make us more productive and not destructive?

Well, that's a great question. Are the proper safeguards? Well, here's what people have done with most of these models. They put a rules engine on the output of it so that can't go to it and ask it how to build a nuclear bomb. By the way, it knows and it will try to tell you. And then it's stopped by a rules engine that says no, here's the proper response. And those kind of rules, which now are into the millions, have taken over a thousand people, mostly overseas to do that work for over a year. Right? This is true with ChatGPT, true with Bard at Google. It's true with really across the board. So everybody's putting rules on these things to try to keep people from jailbreaking them. And you don't really want these models. Unlike the models we did in the 90s where we recognize people immediately say, do you love me? We actually coded in and allowed it to say, oh yes, of course I love you, but I can't love you the way you want because I'm just a machine or I'm already married or things like that. Right? Well, now they're just saying no, we don't want you to respond that way at all. So why does a large language model want to respond saying I'm sentient or I'm in love with you or thinking the reason is it also trained on fiction. It didn't just train on fact. It read thousands, tens of thousands of fiction novels. And so it knows fiction and it knows that those are reasonable responses. When you're placing word after word after word, trying to place the response when someone says, I've really grown to love you or care about you or I'm worried that you're sent eater. It has learned to respond because it's read all those books and now they put rules in front of it that said, no, don't respond to that. People, people are bothered by it. So look, any new technology can be used against you. Excel spreadsheets can be used in bad ways. Right. You can calculate things you shouldn't calculate. Large language models certainly are being used by ransomware people to generate more complex phishing schemes than you could generate before. But it's true with all technology, the wheel has been used for good and has been used for bad as well.

What about the protection for the these models are being trained in and the people whose jobs might be affected because of the models potentially going to replace them?

Yeah, yeah. So let's talk about training. So there's a complex area, and it's complex. Because of this, humans are able to read novels and learn from them. We can read novels, we can read books, we can read patents. That is how we learn. We read work, the work of others. We learn from it. We are not supposed to plagiarize it, but we can learn from it. This is true for patents. We can read 20 patents and find some holes where we can carve out something new. That is how we learn. That is exactly how these AI models are learning as well. They're going out and learning literally billions of phrases from all of humankind's writings, Right? Learning how to put those phrases together, learning how to tie those phrases, reasonable answers to questions, or you ask it to write a screenplay, or you ask it to do something, Right? It has learned from all of human, just the way humans do. So as long as it is not specifically plagiarizing, without giving attribution. So you can, you know, you can use a sentence or two from things for news stories and for other things, but you have to give attribution. So I think attribution has to be dealt with in some areas. But, but in general, these aren't plagiarizing. You could try to jailbreak them and try to get them to do so, but they're not doing that. They have learned, absorb like humans, and then come up with their own sort of outcomes of basic sentences and phrases. So we allow humans to do that. And the copyright law, they does not say you can't learn from anything. You can learn from everything as long as you have access to it. You can go to the library, borrow a book without even buying it, learn from it and bring it back. That's how it's done. So that's what AI models are doing. And our copyright law doesn't keep someone from doing that. So Sarah Silverman and others obviously have sued to say, oh, I think, I think it used my book and it learned from my book. Well, I learned from your book too, Sarah. I'M not against Sarah. I met Sarah, but it doesn't matter. I mean, I'm not good. I learned from your book. That was the point of publishing a book is so that others could learn from your work or your jokes or your novel or whatever it is. Right. So I think this is going to be a difficult legal ground for people to pursue. Clearly, the news organizations are starting to cut their deals with these large language models saying, if you really want to learn from the New York Times, that's fine, but you actually have to pay us for that access. And you can't pay us the way a single human would because you're learning en masse. And so we'd like a different payment structure. So I think they're going to get that. I don't know that novelists are going to get that, per se, but we'll see. Okay. The second thing is taking jobs. Really. Two questions here. And what I like to say in my keynotes is the following. And I do 40 or 50 keynote notes on generative AI a year. What I like to say is the following is that it's about the tasks, not jobs, that are going to disappear. So certainly if you have the task of writing blog posts for companies, first of all, jump head over heels into generative AI because it's going to make you 20 or 30 times or 50 times more productive than you were. But the task of actually writing rather than editing, maybe going away, just like the task of doing long division went away with the advent of the calculator. We don't. I mean, when's the last time you did long division? Fourth grade, something like that? Right. None of us do it because we have a calculator. We have spreadsheets, we have computers. We don't do long division. There's probably no need to teach us how to do long division. How many of us write literally by hand with beautiful cursive? Almost no one. Why? Well, there's the advent of the computer and the typewriter, and so we don't use that function anymore. So I think we're going to see some tasks go by the wayside simply because using these tools makes us 20, 30, 50, 100 times more productive, amplifies our brain power. And we're going to use that amplification again just like we do a calculator. We just. Calculator and Excel are tools that we're used to now. Yes, they probably took. They took a lot of tasks away. They took entire teams out of accounting. Yet in the US there's more people employed in accounting today than there ever have been. Even though they all use spreadsheets now rather than tally up things, you know, in ledger books. Right.

It really is going to benefit those who begin to adopt a technology to improve their productivity and eventual period of embracing the technology for what it can do.

Yeah, kind of an adopt or die situation. Right. You know, again, I think if. If you're an artist and what you used to do is create art and illustrations for people's PowerPoints and companies, you know, you better embrace this so that you have 10 times more clients and that use product for your clients much faster. Your clients don't really want to learn how to do multimodal large language models that, you know, generate exactly the right kinds of line drawings or pictures or whatever it is. They still would like to rely on you, but they don't want to wait three weeks for you. They want to write an hour. Right. So you need to master the tools or they will master them, or your competition will master them. But you can't sit there and go, the tools aren't going to be here. Yes, you know, I hear the plea of the actors and hear the plea of the writers, but two things, you know, if I'm not at the writers union, but if I was in the writers union, I should not get up every morning and say, I know what I'm going to do. I'm going to keep studios from using AI. Well, that's not going to happen. It's not going to happen. Nobody's ever stopped the march of technology, ever. So said you go, how do I be the master of this AI and how do we include writers to be sort of the robot overlord of AI so that you don't have a producer just using AI and writing around you? Right. There's a way to embrace that. But it might mean there aren't eight writers in the writers room anymore. Maybe there's two. And you're mostly editors and you're leveraging AI to do a lot of the heavy lifting because that's where things are going. Right. And so everyone's got to think through this and say, it's here to stay. How do I embrace it? How do I maximize my productivity? How do I make the most money? How do I make me the most valuable to my employers by having this knowledge?

Is it possible for the overlords to become AI?

Well, you know, that's a good question, but in the end, the bottom line is we as humans want to edit and put things into our voice and add our value. Right. Sort of what we do. So we have to just go back in history and say, you know, what happened all the way from the invention of the wheel to the invention of robots. The first industrial robot was installed at General Motors in 1961. It made GM more productive and allowed them to make a few more cars than they could before. Today they have tens of thousands, if not hundreds of thousands of robots. And yes, an awful lot of the people who work at GM almost certainly manage those robots, repair those robots, reprogram the robot. Nevertheless, GM is making as many or more cars as they have ever made and over time, arguably at a lower cost, cost in terms of sort of real dollars than they could even make before the advent of robots in 1961. You know, these technologies come along and make people and companies more productive, which then drives higher revenue and higher profits and lower prices ultimately for consumers, which makes consumer demand go up and raises the GDP of an economy. Now some people, I know what some people say is, oh well, you know, I'm not going to get the advantage of that. Well, two ways to get the advantage of it. But the best way is, you know, use these new techniques and technologies and make them just part of your life. It's what you do. And the second is there always is a delay as GDP goes up. You know, that everybody gets to participate in this. But in time, the best economy was the strongest economy in the world. Is United States, the highest GDP in the world. And where else would you want to live except the highest GDP in the world? Right. And so this kind of technology is going to drive overall economic opportunity higher by driving consumer costs down ultimately. And I know that's a long winded way of what's the impact on the economy, but in the end it's actually quite good because if we look through history, every other productivity increase, from desktop computers to robots to wheels to whatever, to excel, has actually driven ultimately prices down in consumer demand up.

Well, certainly encouraging. And maybe to close. Any final words regarding the future for AI and economy?

Well, look, I think this is one of the most exciting times to be alive because we've never had language models that were, by the way, in AI. We've had language models. They've, they've been around since the 60s, since a project called Eliza at MIT. So it's just been getting better and now they're so good you can actually start to use them a lot of the time with a lot of editing, with a lot of focus. Right. So there, there is work to do. But what an exciting Time because we've gotten so productive in math and accounting and other areas. We never had a lot of productivity given to us in language. And now these are just going to be. These sorts of things will be built into Word. And, you know, if you're writing something, yes, you're going to ask AI to write it first, then you're going to go through and edit it and you're done in 20 minutes instead of perhaps 20 hours when you had writer's block. Really good. Are there bad sides to it? That side to every technology? I said the wheel. But this is the best time I have ever seen to, you know, to be alive and embrace technology. And it's going to have tremendous impacts and they're mostly positive. And no, it's not going to kill us. It's not going to take over the world, so relax. By the way, in the end, these models, these large language models, are actually just math models. All they're doing is statistically putting weightings on the next word in a phrase that would be appropriate, choosing that word, placing in that phrase and delivering it to you. And it's all math. You know, it's a. It's an rnn, it's a neural net, but it's all math on the probability of the next word. That's it. It doesn't actually know what it's saying, by the way. It has no sentience whatsoever. It's just math.

It's like some people.

Really. Yeah, like some people we know. That's exactly right.

Well, a fascinating discussion, but I will just close by saying we were talking. Mr. Kevin Surese, on the topic of AI. Thank you very much for joining us today on the grokscience show.

Yeah, it was a pleasure. Thank you so much.

And that's all for this week's edition of the grokscience Show.

Make sure you tune in next week for more from the world of science and technology.

If you'd like to contact us here, you can email us@sciencegrocks.net For Grok Science, I'm Frank Ling.

And I'm Charles Lee. Make sure you also see us on the web@www.groks.net.

Have a great afternoon and keep on Grokkin.

It.

Sa. It.

It.

Sa.