TEXINTEL TALKS - EPISODE 099 - EXPLORING AI AND ML WITH INDUSTRY EXPERT KEVIN SURACE

Foreign.

Hello, everybody, and welcome to today's podcast. Today, it really is an absolute pleasure to be joined by Kevin Sarrace. Kevin, welcome. Thanks for joining us.

Oh, I'm so happy to be here. Thank you for having me.

All right. It's fantastic. Kevin, your Wikipedia profile we're just talking about the offer is amazing. So you are, gosh, a technology innovator, speaker, and entrepreneur. Goodness me, you have so many achievements, 93 patents. That takes some doing. So I think it's probably better that I hand over to you to highlight your accomplishments in your journey and what you're up to, but.

Oh, that's much harder for me to do it.

Sorry.

You know, look, I started with humble beginnings in upstate New York. Always loved two things, music and musical theater and electronics. And I went to school at Rochester Institute of Technology and got my main degree in electrical engineering and computer science. And my sideline was still music. And I continue to do both through my career, inventing all kinds of things. And on the technology side, all of the technology that is used today and the methods for all of our virtual assistants, from Siri to the Google Assistant, et cetera, were based on the work that I had done back at a company called General magic in the 90s. So we invented the first AI. Virtual assistants then was very hard to do. We had to build our own operations center. You had to build everything. There was no cloud. You had to build everything yourself. But that led to the last 25 years doing more and more with what we called applied AI. And that's different than someone who's designing algorithms, which is very cool, and they're way smarter than I am. But applying those algorithms is something that we can use in everyday life. That's called applied AI. And so that takes a different set of skills where you look at the business options and you look at the business opportunities, you look at productivity, you look at people's needs and pain points. You say, how can I apply AI in really unique ways here? And that's what I've been doing there. And in building management, what became the trained building manager, which is one of the top building management systems used worldwide to AI and QA today. That's AI that finds bugs in our software automatically. So continue to develop new techniques in applied AI and applying AI across the landscape. And it is true. I've got 94 now worldwide patents. I guess someone will have to update that. Wikipedia, they keep growing a little bit. But I'd summarize this. Every day I get up and I want to have a sense of purpose. And my sense of purpose is, where are people's pain points and can I solve them with technology? And so, you know, the virtual assistant came out of the fact that we shouldn't be typing and doing things in our car. We should just be talking to essentially a human. And we can't all afford our own actual assistant. So could I create a virtual assistant? Well, today that's obvious, but in 1998, that was not obvious. There were no smartphones yet. There was, you know, there was nothing available. So always trying to solve problems. Pain points.

Fantastic, really is. So today we're going to dive into generative AI. And I think, you know, for our audience, for our listeners who predominantly reside in and around the textile industry with a focus on printing. Could you explain the term AI?

Sure.

Simplistic terms, how it works, and, you know, it has such a transformative impact for both the workspace and just everything to do with manufacturing. I wonder if you could just explain those terms.

Of course. So some people would say artificial intelligence is more of a marketing term than a real term. All of this underneath it is machine learning. Right. What we're doing is we're allowing a computer system to learn and continually learn and. Or learn from immense amounts of data in the past. Immense amounts. And there are hundreds of algorithms that we use today. But the ones that are getting the most attention, though, they're not always the ones that are most used. And in the textile industry, for example, I'm going to give you some examples of things we would do that might not use these. But the ones you hear about the most are all based on a concept called deep learning and neural nets, including generative AI based on the same thing. And I'm just going to. I could go deep into this, but I'm just going to summarize it this way. It's just math. All we're doing is learning from a bunch of data that's repetitive and to come to the same result. So if I put in the number seven, and I want to recognize that from the number eight, I want a machine to teach itself how to recognize seven from eight, and I've given it the right math to build its own model to do so. So it's building its own neural net and how to recognize seven from eight, and then it can recognize eight from nine, it can recognize nine from one, et cetera. Right. And then we expand that and we can recognize Debbie from Kevin. Right. And so it builds out, literally, a decision tree, if you will, or it's really a neural net. With a bunch of hidden decision points. And all that math does is look at what's coming next and say it's more likely to be this or less likely to be that. So now I'm going to expand that one step further to generative AI. So that's the basic underpinnings of everything. So with generative AI, I'm going to train sentences or phrases instead of individual words. So a phrase is built of many words. So for instance, in generative AI, if I've trained it well, and I say, today is Debbie's birthday, I don't know that it is. I'm making it up. Right. Probably isn't. But today is Debbie's birthday. What would you like to say to her? It would very likely come back happy, to start with happy. And then it would look at all the billions of words that could come up after happy. Happy glasses, happy headphones, happy birthday, likely birthday is next. And then if there were a third word, it'd probably be Debbie. Happy birthday, Debbie. And it simply does that by mathematical probability of what word should come after the prior word based on your request. Does that make sense?

Yeah.

And so I think when people think about this and they go, life is over as we know it. It's a very smart machine. It's going to be smarter than us, it's going to kill us. It's going to. That is completely wrong. Or it's sentient. It sounds sentient because its ability to create reasonable sentences is uncannily good. It doesn't say it's accurate, but it's uncannily good. And sentence structure is very good, May not always be accurate, but that's all you need to understand. That's all it is, is a math model doing probability on the next word after the prior word and building out sentences like. And that's how it works.

So when you get into something technical, like the textile industry and trends and analytics and all of those things, where does it source information from?

Great question. So generative AI models are general language models. They're called LLMs, large language models. And as a large language model, it mostly knows language really well. But in learning everything on the web, literally trillions of phrases by chance, it also came across lots of statistics, including textile statistics and retail statistics and color statistics and things like that. So it wouldn't be a surprise if you. So, for instance, it's not a surprise if I asked it who was the first president of the United States? You know, we might assume it says George Washington. You know, why it didn't go research history, but it probably knows that because it saw that many times. Does that make sense?

Yeah.

Saw it enough times where it can come back with that fact. However, it can also come back with a false fact.

So based on probability, but it's. It's fake based on probability. Probability. Yeah.

I. If you ask it the right way, most LLMs, if you ask it, who shot George Washington? Now we know that nobody ever shot George Washington, as it turns out. However, it did read a novel of fiction where George Tryon shot George Washington. And if you goad it in the right way, it'll come back and say, well, obviously in the way you're asking me, George Tryon shot George Washington. And that's true in a fictional novel. It fails to tell you that that only happened in a fictional novel because it actually doesn't know the difference between fiction and fact. It learned everything unsupervised. So these are the pluses and minuses. Now, if I wanted to really build out an AI model specifically for. Based on, let's say, sales trends over the last 10 years, I wouldn't use generative AI. I would build out a neural net based on the exact data that I want. I don't need a large language model. I need the data that I have or the industry data or whatever. And from that I might be able to predict the future based on prior trends. You can't always do that, as you know, but based on prior trends of people are moving to, say, more neutral colors, you know, we, we could predict that that continues for a period of time. Right. That might be one, one example. So. So as I said, there's lots of forms of AI. Everybody's excited about generative AI, but generative AI is primarily around language and it happened to learn of a lot of other things. And by chance it might be right on a lot of those other things, but it might also be wrong.

Yeah. Yeah. It's interesting, isn't it, how the way. It's the way. The way you read everything in the news, the way that, you know, the marketing that comes out, people are. We're going to come on and talk about this a little bit more later. I think that's why people have been so threatened about the use of AI, really, and that, you know, it isn't going to replace the human because we have to fact check. And I think it's probably open to. Would you agree? It's open to misuse by people who don't. Within technical realms that don't check. Yeah.

Yes. So so look, you're not going to approach something that you ask an LLM about. And I use, I use large language models every day. You're not going to approach it saying it's probably wrong. But, but, but you know, I don't go to it for sets of facts because I have better fact checking mechanisms by the way, including Google search, which is, it'll give you, you know, 20 websites and then I can go evaluate where those sources come from. It's the New York Times, I probably believe it, right? It's the London Times, I probably believe it, right? If it's the National Enquirer, I probably don't believe it, right? So I can, I can immediately check the value, you know, based on this. So we have had search for 25 years. That has been excellent. That allows us to fact check if we're smart enough, right. An LLM doesn't tell you where it got its facts from. So if you don't use it for that and instead you use it for instance for generating marketing content. So you might say I've got this new apparel coming out and it looks like blah blah, blah blah, and you give it all kinds of description. Can you generate advertising copy that is a summary of what I said that's exciting to the consumer and do so in 50 words or less. It is phenomenal at that. You just gave it all the facts it needs. You're leveraging its natural language processing, its ability to put sentences together and its ability to reason based on the information you gave it. You didn't ask it who shot George Washington? You said here's a bunch of content, please give me back something that I can use and if you don't like it, you can say I don't like that one, give me five more and it will. That is brilliant. Now that's like having a marketing assistant. Or it's like you instead of spending two days doing this, might spend 20 minutes, which includes getting the information and editing it and you're done. Making you 10, 20, 50, in some cases a hundred times more productive than you were before. So if you use the tools correctly, you have an opportunity for up to 100 brain power of you. It's still your brain, right? You're putting the inputs in, you're editing the outputs. But we've amplified your brain by a hundred x. When have we? We've only been able to do that a few other times in history. And I'll give you an example and this will level the playing field. People don't remember probably that the Calculator came out really in the 70s, and then broad use of the 70s, and then by the mid-80s, and certainly late 80s and early 90s, we had the spreadsheet. Okay, Those two things solved math for humanity. So much so that I could guess right now. Debbie, the chance of you having done long division with a pencil is about zero over the last 20 or 30 years.

I'd agree with that.

There's just no purpose because there's a machine that does it. And when those machines came, including the spreadsheet, if you were in finance and you were an accountant instead of accounting, what you literally did is you were adding up columns and rows in ledger books, right?

Yeah.

All day. Hundreds of people would do that. Well, this replaced that task, but not their job, because we employ more people in accounting today than we did then, except their jobs are quite different today. So we have now solved language the way we solved math 40 years ago. And that solution didn't kill anyone. The spreadsheet is something we all use today as a tool. We will soon look at large language models as just a tool. It's just a language tool for us. And it just took us a lot longer to solve language versus math because it was a harder problem to solve.

It's interesting, isn't it? Because as you just said, you know, in life, everything's always about input and output and clarity and fact checking as well. Really, one of the things that, you know, you use the example of long division, you know, I've used a calculator for that long. I don't think I could even remember how you do it manually. And I think the same would be said of a more recent technology, I suppose. Sat navs. You know, we. We all. I mean, I do it. I do it routinely now, but in the past, you'd have had a map and you would have kind of remembered where you were going, and you would have found your way back to that point on your second visit. Yes, I, you know, I absolutely am so excited about everything that AI has to offer. But I also worry whether it stops our learning in some way. It stunts our learning. A good or a bad thing.

Well, we have changed our learning probably for as long as human history has been around, right? So from the invention of the wheel, look, there were people who had to essentially move large objects by having lots and lots of people. And they learned to do that, right? They learned to do it in very unique ways. And then the wheel came about and all the learning that they had before was no longer needed. Right. And this Is true throughout history. So the need to do long division has ceased since the 1970s. We still teach it in school. And there's some question whether we should. I mean, that's a great question. I don't know the answer, but. And many of your listeners will say, this guy's crazy. Of course we have to teach that. Okay, But. But we don't teach probably basket weaving anymore. It would be hard to find a course on basket weaving because no one actually weaves a basket. Now there's some engineer somewhere in a factory that teaches a machine to weave a basket, which we've done for 70 years now. There's no humans weaving baskets, probably as a way of making money.

Yeah, but being quite brutal about that, though. Language is about learning basics, isn't it? You know, and the constant repetition of writing, whether it's an email, whether it's marketing content, whether it's a novel, whatever it is. I think you could desensitize yourself, I suppose, really, and kind of forget the rhythm of the written word.

These are the same arguments. I hear you.

I'm not arguing, I'm just saying. No, no, no. Putting it out there is like. I just kind of. Because it happens to me with, you know, SAT navs and stuff like that. And you think, well, I'm not. I'm actually. Because I'm not living about it anymore in the moment anymore.

Right.

Exactly.

That's right.

That's a better way of putting my point.

But. But I can only say this. Those. And I know you're not arguing, but those are arguments that were made when the spreadsheet showed up. We will. We will lose the ability to do addition and subtraction and multiplication. We'll lose the. We won't need to do it anymore. And I get that. However, it made everyone more productive and I don't know, humanity seemed to survive. Even though most of us will not be able to do that math long term, in the very long term, the machine will do all the math and is doing. So that is starting to happen. A language. Now, here's what's interesting. We can still choose, even today to do long division. We can choose to pull out a map. You could still buy them, or print one out and figure it out and pencil our way there and pencil our way back. Right? Nobody's stopping you from doing that. We don't have to use SAT Navigate. And nobody's stopping anyone from writing the next novel or writing their blog post or doing anything by hand. You can still choose to do it, however, if you choose to do that, you are at a disadvantage compared to your competitor who has chosen to use AI.

Yeah, that's a great segue into the next question actually. How do you think quite a lot of our audience work in the fashion industry or the interior industry or the spirit sportswear industry? Or they might be printers. You know, they, they, they have manufacturing workflows. They're starting to look at automation in a deeper format. They're starting to look into robotics. What are the implications of AI for them at any scale of business? I think that's really important that AI and its application is incremental. You don't have to be a huge company to be able to afford to do this.

Well, the nice thing about Gen AI like ChatGPT is that it's available to a one person company and you can use its lowest end model for free. Well, that's pretty interesting. Now what can you do with that? Well, you're gonna write blog posts and advertising copy and maybe generate PowerPoints and things like that. Right. Maybe do a little market research. That's amazing. For free. Now you probably were doing a lot of this anyway using Google, but you were searching, you were finding what other people wrote. It just took you longer. This is taking hours down to min. That's fabulous. As you get to be a manufacturer today, it is all about reduction of labor costs because your costs are really materials, maybe your infrastructure, your real estate, but almost everyone, it's labor. Okay? And there is no question that there is a move worldwide to onshore manufacturing, to reshore manufacturing for, you know, all of the reasons. We know. And I think the easiest, you know, the easiest way to say it is that I know this is a little bit of political story, but I don't think anyone's going to disagree with me. If you choose to do business with a communist country, eventually it probably ends badly. It just does just this. Throughout history that has never worked out. Right.

Yeah.

And we all chose to do business in Russia for 30 years and we've chosen to do a lot of business in China. And we know over time this probably ends badly. And we only did it because the labor was cheap. I mean, that's the truth.

It was, yeah.

Nothing against the people. I'm not, you know, I'm not making a bad statement. It's just we did it because the labor was cheap. Okay. If we can drive labor out of our processes and thus drive transport out of our processes, now we can start thinking about onshoring at about the same cost. So for a very long time Robotics was far, far more expensive than cheap labor. But cheap labor has been getting more expensive and it's 300% higher than it was a decade ago in China. Right. So labor has gone up and robotics has continued to come down. And the accuracy, because of AI, which I'll get into a second of robotics, has changed everything. So we had vision systems on lines decades ago where we could do basic things, not very accurately, but we could align a robot to some cross hatch and crosshair on the line and do something. Well, today with these vision systems we can recognize things incredibly fast and more accurately than human eyes can. Well, that changes everything. And they're cheap, These vision systems are cheap. Right. So I can train up a vision neural net, basically, right, a vision system to recognize defects. I'm going to pick that, pick on that because that's easy. Defects in some, some printing, let's say, okay, and put a crosshatch on that one so that it gets cut out at the end of the line or whatever the case is. And it's a defect that maybe a human would not see and could not see at that speed, but it will see it. Okay, this is pretty interesting. And so then you start to say, where are all the places I do use people on a line? And can I stop that? And the truth is we could probably replace virtually 100% of humans on almost every line with robotics. And some of those robots are quite expensive. But if they're going to run for 20 years, that million dollars might not be so bad. And it's very fast and it's very accurate. And it runs 24 by 7. Right. So that might be replacing three or four people for 20 years. And you go, okay, well that actually is a pretty good deal then. Right. Especially if you're leasing it. So, so I would highly encourage people to look at a lot world class robotics that's available, a lot of that's out of Germany. As you know, they, they, they make a lot of this equipment still. And, and a lot of it, all of it can be customized to your needs today. And so as you scale up as a larger manufacturer, it's not that you're not doing the marketing with LLMs and having it right, blogs and ads, but now you can really impact your business drive costs down. So when I look at pain points and I'm, I'm involved in a couple of companies that do manufacturing, I look at pain points there. It's all about how much can I do in parallel at what speed. I like to strive for a lights out Factory. I went to my first lights out factory in South Korea about 20 years ago. It was a printed circuit board factory. And I go in and for as far as I can see, there's no humans. And the boards are flying out and there's all this equipment and it's picked up and it's moved and it's all clean room. I go, there's no humans. They go, well yeah, it's a lights out factory.

Oh, and does that actually mean it can operate in the dark?

Yes, it can operate in the dark because since there's no humans, you don't need any light.

No lights on?

Yeah, no lights on. Lights out factory because there's no humans walking the floor. So the, the robotics have their own lighting for their own needs. And that might even be ir, might not even be lights that we would see. So you literally turn the lights off. Why would you even run the lights? There's no humans. And, and this is fascinating. So since then I've always been a fan of the lights out factory. Now I know people are going to say yes, but we've got to, you know, we need factory workers. What I would say is if we're bringing back manufacturing, maybe when it left there were 200 people in that factory. But to bring it back, probably there can only be 10 or 20 or else it won't be possible.

Exactly. That's exactly what's going on in the printing industry.

But that's fine. I want to bring it all back. I want. And by the way, printing is coming back to the United States right now. I'm in a suburb of Rochester, NY and I know the team at Kodak and Kodak makes these huge, huge printers for packaging primarily. Right. And other things. And they run so fast and what they're doing is unbelievable. I just, I was with the CEO Jim, a week or two ago and it's unbelievable, unbelievable what they're doing. And it's a multibillion dollar company still. It's Kodak, but they're in printing. You know, they're not as known for film even though they're still one of the world's only filmmakers. And, and they do make a ton of film. In fact, they're sold out for the next few years on film is what.

I read that actually. Yeah. Don't know why I know that. That's an odd fact now. But no, I absolutely, totally agree with you. Digitization is the whole way and it is actually reshaping the landscape of the global economies really. We're right at the Beginning of that journey with near shoring and on shoring. And in some ways it's an ideal, it's an idea, and it's only an idea that can't actually land because we have lost a lot of the skills that we need to support manufacturing. So that takes time to build those up. But as you say, that's building up new, a new footprint of, in my, from my life, the textile industry using new technologies and a newly trained workforce who work in a different way and have a completely different skill set. Really.

Oh, they are robot overlords. I mean, they're there really to program and manage the robotic line. They're not there to actually physically do something on the line. There might still be, you know, delivery of large rolls of textile or paper or whatever it is at the beginning of the line and some someone taking it off and shipping it at the end of the line. But again, you bring these back by reducing the workforce by 90%, but it had been reduced by 100%. So we're going to bring back some people, we're going to bring back these lines, we're going to get out of this geopolitical mess that we got ourselves into for low cost and get back to making things near shore. And that's good for the environment because all of this stuff we've been shipping on ships and airplanes all these years was never good for anyone. It's all bad. And there's a time value of money and the transportation cost and all that as well.

Yeah, we evolve, don't we? Yes, we absolutely evolve. Speaking of evolving, then, how do you think AI will change the future of retail particularly?

Well, look at Amazon. So Amazon's been using a variety of AI for maybe 20 years. I mean, I think the world today, or consumers think, oh, AI just arrived. So people in the field have been working on AI since the 50s. We made great strides in the 60s. Then there was a slow period of 70s. There were more strides in the 80s. It goes on and on. So when ChatGPT came, while consumers maybe were very shocked, it was the result of decades of work towards a model that could. Could really ingest a trillion phrases. Right. As opposed to a billion. And it turns out to make ChatGPT, you probably needed three or five or seven billion dollars of cloud services that Microsoft donated to them. Right? So to make a model that large, there's only a few companies in the world who can afford to do it. And they have. It's Google, it's Facebook, it is Apple, and it's OpenAI because of Microsoft, those are the only people who have $7 billion to spend to make a model.

Of now they're not going to miss.

Yeah. I mean, this isn't something that a startup can do. So it took immense amount of compute horsepower. So people say, well, why didn't we know how to do that 10 years ago? Even if, by the way, we sort of did in 2012, we understood that you could build potentially a large language model. Although we were still doing it word by word, not by phrases. Until 2017 when Google invented the idea that converting English to French wasn't working very well word to word because it would create a mangled French sentence. Right?

Yeah, yeah, yeah.

Because the sentence structure is different. So someone had a brilliant idea that said, how about we learn phrases now, learning a phrase instead of instead of a word isn't a brand new idea. But finally, around 2017, we had enough compute horsepower to do it. And so all of these have been governed by that. So then they said, we have enough computer horsepower, so let's do that. And that's what led to these large language models. So 2017. Right. So if you had followed this progression, like you're not shocked that you finally have a model with a trillion phrases, it's just you knew it would take billions and billions and billions of dollars to do so. When you look at retail, you have to say, what is Amazon doing right? What are they doing wrong? And obviously from an online retail perspective, they are very, very, very good at predicting what it is that you want and what it is that you. Every Amazon page that you see is very different than what your neighbor sees. It is all customized to you. It is true one to one marketing. And like I said, they've been using different forms of AI for probably two decades. They also use it in their warehouses. Okay. Those Kiva robots are amazing. And they have done more to automate warehouses than anyone and push that technology forward. They have no choice, Right? They have to deliver a $6 item to my door this afternoon for $6 and that included the cost of the item. How do you do that without immense amount of automation? Doesn't say they're perfect. Doesn't say people don't think they're the enemy. So now that you look at whether it's big box retail or local clothing stores or whatever it is, again, you're going to be looking for efficiencies throughout the entire supply chain. Throughout the entire supply chain. And also efficiencies. On understanding the market. Okay. And market trends, one of, you know, this better than I do. But, you know, one problem we have is, is we make too much of the wrong thing and nobody wants it for that season. That's certainly been an issue in, in, in, in the clothing world for a long time and sportswear and other things. And, and I can really help predict that and see those trends ahead of time. Look at prior trends, look at future trends. Prior trends were tied to, were they tied to weather? Were they tell to tied to an El Nino or whatever? Right. All of these things affect how people are going to react and what they're going to buy. Is it going to be, you know, in England, in the uk, is it going to be really, really warm winter? Is it going to be really, really cold? That will have a huge impact. And if it's really cold and everybody sells out of winter coats, well, you can't make more winter coats in time. They'll be, they'll come in, in the summer. Right.

So yeah, exactly.

You have to predict. And so now you get into weather prediction. So AI is really good at taking in hundreds or thousands of those different variables and looking over time and making that tie and say, well, based on this, it will be a cold winter. We're going to make 30% more winter clothing than we might next year because that's a different ocean flow and we're going to look at something different. Right?

Yeah.

So I think that's going to affect all of us. I think we can also look at, you know, brands as peloton going up in a brand or going down in a brand. Right now they're trying to recover, but you know, they're a, they're a Covid darling and now a non darling and there's a lot of those. So again, AI can help predict some of these. But AI could not have predicted Covid when it hit because that is a black swan event. AI can't deal with something that falls out of the air.

Yeah, exactly. It can only process the data.

And all of a sudden in Covid, nobody wanted any of the clothes that we had. They wanted sweats.

That's true.

And you couldn't make enough sweats. They wanted to just be home in sweats and they would be in sweats on their calls and, and these things. And, and then we couldn't predict the day that everybody would go back and say, oh, I need a whole new wardrobe. And there was no new wardrobe. The wardrobe was in the star store was the wardrobe from two years ago when the store closed. And then you open the store, you go, that I don't want that wardrobe. So those could not have been predicted. But I think a whole new economies.

Have been spawned because of that. You know, the whole rewear upcycle. Absolutely vintage. You know, the gosh, there are so many platforms now and people are really circulating their wardrobes, especially amongst the younger generation.

Well, you know, the younger generation, including the United States and much around the world is much more sensitive to that. They are not, I think like we were, oh, we've got to have the next new thing and I, I've got to be seen in this new thing and I don't want someone to see me and, and they would rather say, oh, you know, this, this is something that someone else wore. It's in lovely shape and now I want to wear it. And my daughter is very much like that. She gets most of her things from these sort of used, you know, so does mine. Websites.

Yeah, same thing.

And takes great pride in it. Look, this was worn by someone else for only a few times and then I got it at a quarter of the price and I'm wearing it and I'm going to wear it for quite a while and then when I'm done with, I'm going to put it back on there and someone else will get it. Right. This is pretty fascinating because. Yeah, I don't think my mom or my wife would have ever been caught dead in any of that. You know, they had to go to the store and get the latest thing.

Well, because it was seen as a signal of, not of the opposite of affluence really, wasn't it? To be wearing secondhand clothes way back, especially post war generation.

That's right.

We're sending out all the wrong signals to the neighbors.

That's right. That's right.

Oh, the preschool meals, it's. Yeah, it was terrible.

That's all they can afford is the used stuff. Well, now it's a badge of honor.

Yes.

And in fact the younger generation, if it's something brand new, they go, why did you buy that brand new? You couldn't, you know, couldn't you found something that was available? You know, are you just trying to flaunt that you can buy something brand new? So it's become the opposite. It's really fascinating. Not good for the retail clothing industry, but interesting. Unless you're in used clothes, then it's good.

Yeah, it's all about balance, isn't it really? I think. Which kind of leads us on too. One of the things that also makes people worry about AI, I guess is how do we Is this even an issue? Kevin, how do we protect copyright and intellectual properties in the future? Should we be overly concerned about this?

Well, people are very concerned.

Yep.

But let me. Let me share a viewpoint, which is this. So humans learn by ingesting the world, right? We look at textiles, we look at colors, we look at patterns. We also read novels. And after we've read, who knows, 10, 20, 50, 100 novels, we may choose to write our own novel. When we do, those hundred novels informed us on the way we want to write, the style of which we want to write in. We may not remember it. And I'm not talking about plagiarism, right. I'm talking about we couldn't have written a novel had we never read a novel. That's impossible. Right. You'd have to read many before you'd say, I understand the structure and I'm ready to write now. Same with a playwright. A playwright might see 100 plays before they say, I think I'm ready to write one. And they might have read, you know, the style of the writing and how to do the staging and, you know, et cetera, Right? So what we've done with AI is we said, go train on these same things that humans would train on and then do not plagiarize, but learn from them. So not allowed to plagiarize. And there's a lot of rules in the good models to keep it from plagiarizing. And if it does quotes, if you force it to quote something, it'll say, this is a direct quote from. And it'll give attribution that's legal. So this is the question, because our copyright laws in the west, because there are none in the other side, but in the west only dealt with humans and allow humans to learn. Now, we also can't write a direct derivative work, right? So I just can't. I can't instantly go write a novel about Hogwarts, you know, and call it Hogwarts. I could write a different novel about kids in a totally different storyline that might go to a castle. I don't think they could take the train there. You know, I'd have to. Have to be. It can't be so close that it's derivative, but I can certainly learn. And by the way, I could write anything I want in the style of Harry Potter. I can't say it's in the style of Harry Potter, but I could learn a lot from Harry Potter and say, I'm going to write my own story and it's about witches that go away to a witching school, whatever. Right. I could do that. And as long as it's not too close, then it's no longer derivative, Even though I learned her writing style and I can write in that style. Okay. Totally allowed.

Yep.

Because humans have to learn from other words. Some people call it fair use. And there's fair use in education specifically, that if you were a teacher, you can teach about the writing style of. Or teach about the novel, and no one can sue you for quoting the novel because it's used in a learning environment. Right. So we are supposed to teach from our prior work so that we can learn from prior work. Now we've asked AI to learn from prior work. And. And then I can go to AI now and say, can you. I would like a. A. A short story in the style of Ernest Hemingway. Now, it will not. It's not supposed to, and probably won't do anything that plagiarizes Ernest Hemingway, but the style will surely be in a style similar to Ernest Hemingway, because it's doing nothing different than I would do. I read every Ernest Hemingway novel. Therefore I can write in the style of Ernest Hemingway. Should that be illegal? Now, if you are Ernest Hemingway's family, you're going, that's unfair. I don't want it to write in the style of my grandfather. That's terrible. That's stealing. Except Debbie can write and Kevin can write in the style of their grandfather. Right. I can't specifically say this is in the style of Ernest Hemingway, but I can write in that style. And as long as I didn't plagiarize, nobody can throw me in jail for it. And so now I've asked AI to do the same as I would be expected to do. To learn from a series of writings and then maybe write in the style of. Or. Or give me some examples of or whatever. And so our copyright law, I think, at least in the US If AI is treated like a human, there is. There is nothing in copyright law that suggests today that it's doing anything wrong. It's not plagiarizing. It's not specifically, you're stealing the storyline, but you could ask it to write in the style of. Or you could ask it for a quote from. And it will give you a quote because it read it, right?

Yeah.

And it'll say it's a quote from with attribution. So I don't see the problem, except for creators who are scared to death. And I understand that. And by the way, this is true with artists. It learned from every art that's out there. And now I can say, you know, give me create some art in the style of. And it'll do. Yeah, yeah, it'll do.

So create, you know, create a floral in the style of William Morris and this color palette.

Yes.

And put, and put it in repeats.

Yes.

And it will. And it will do it, of course.

But, but you could have also hired an artist to do that.

Yes.

Including today. You can go on to upwork or whatever, any, you know, your favorite thing is find an artist and someone somewhere in the world will paint that for you or draw it for you, or create in Photoshop or whatever it is and send it to you if you needed that. And they would charge you whatever, $1,000 or 5,000 or 10,000 or 500, depending on where they are and who they are. So you could have always asked an artist to do that and that's not illegal. And they would have studied William Morris and done so in the style of William Morris. Right, Totally fine. Now I've asked the machine to do it and nobody likes that. But it's not, I don't think it's.

Illegal is who owns it. So if the machine creates it, no.

One owns it right now.

Yeah, exactly.

There's no copyright on something a machine creates.

That was my understanding. Yeah.

The US's early court rulings suggest that there's no defense for something a machine creates. Only human created items are defendable. Right. So there's no copyright of the machine creates it. So that, that creates an interesting thing. If I have a machine, you know, if I have AI create a brand new script, I can't copyright that because the machine wrote it. Now could I copyright it? If I am the editor and I edit it heavily, well, that gets complicated. How much did I edit it? How much did I add?

Right, that was my next question.

Yeah. And no one knows. Hasn't been tested.

Yeah.

But we do know, at least for the moment, courts don't appear to be in the west anyway willing to grant AI copyright holder status. And this actually, outside of this was tried at the USPTO Patent Trademark Office where people tried to trademark some writing, copyright some writings, and sent it in and written by ChatGPT and it rejected it. They decided there. That's not what our charter is. Our charter is to copyright works of humans, not works of machines. We're not going to copyright it. So that's interesting and we'll see where.

It goes, I think as it goes then my understanding from my reading too is that if you generate. So if you're if you're in writing or you're marketing whatever, and you generate something using AI, you can't sign it as written by because you didn't write it.

The only thing you can say is written by ChatGPT or written by Bard or written by whoever. Right? Yeah, that's right. And I do that in my keynote presentations. Parts of it. I specifically, because of course, I'm talking about AI and generative AI. I want to give examples and I say written by ChatGPT or Bard or who or Llama on this date because I want them to see what row. And I said, this is what I put in. This is how I set up the prompt. This is the prompt engineering for that. This is my output, this is how I tuned it. And here it is. And I often give the very closing remarks. It says, my closing remarks. This is written by ChatGPT on this date. And frankly, its closing remarks are better than I could write.

Fantastic. That's great, isn't it? We've just got a couple of questions and we've pretty much touched on this second one. But if you just had to summarize then, and just to round up, should people be threatened by AI and how should they prepare to embrace the benefits of AI, which is what really should be happening? Well, just all too much too quick.

Well, yeah, in the public domain, sure, humans are always scared of that, right? It's too much too quick. And so I'm just going to go hide in the corner. But, but here's the truth. Your job is not going away just because of AI, but your job could go away because your competitor is using AI and you're not. Right? So, so, you know, wake up, smell the roses, and let's get going. Right? And you need to use AI and generative AI, you know, in general across your business and to some extent in your life, if it makes sense to do so. But it's very critical that you embrace this. You're not afraid of it. Use it for marketing, Use it for blog posts, use it for ad creation. Use it in your life. Look, I get texts from people that are better pros than I am capable of. And it's so beautifully written. A thank you. And they'll send it by text today. Thank you so much. And, and I will go to an LLM, go to a large language model and say, I just got this text, it's from this good friend and this is what it says. What. How would. How should I best respond in my voice? Right? And it'll give me a Response. And I may edit a word or two, or I may edit nothing. It's just gorgeous. It's. It's what I wanted to say but did not have the skills to say it. Right.

Yeah.

I mean, had I thought about it for two days, I might have gotten there. Right. But I. I go, that reflects my exact thinking. And boom. Right. And then I'll often say, by the way, I did leverage LLM to do this. And they. And they love that, actually.

That's your specialism. Of course you did.

Yeah. So you can use this in your everyday life. You can use it like, I just got in an argument with my wife over the following things. What is the best way to approach a response where she won't be mad at me for my response for the next year? Right. And it actually has very good advice. Now, why does it have good advice? Remember, it read novels, it read fact, it read fiction, it's got all kinds of stuff. Right. And so you can take the advice or throw the advice out, but sometimes you just need advice. And when you do have. Actually, my wife and I never argue, which is great. But if you do have an argument at home and, you know, couples get in arguments, often one of those people just makes it all worse in what they think is an apology, but it just makes the entire situation worse. And they're completely oblivious to why it made it worse. Because, you know, the couples have differences. Right. And so you'd be surprised how good an LLM is at this, at formulating the correct language and just say that and shut up.

Because it's like having communication aid. Really.

Yes. It's like having a psychiatrist who's an expert in marriage, family therapy or whatever.

Right next to you, living in AI.

Therapy that looks up at you and says, no, Kevin. I said, well, what I'm going to say is I'm kind of sorry, but really. No, no, you're not. You're going to say, I'm horribly apologetic. It's all my fault. I completely hear you and I love you terribly and just shut up. I'm making it up. Right. And it'll be more likely than. Right.

Well, it would be good if it also showed you the anticipated response. Quite interesting, wouldn't it?

Well, you could do that. I mean, it all depends on your prompt, if that's what you want, and give me what anticipated response. But I will tell you in just trying it for some of these things, because in my keynotes, I talk about some of this.

I think when you make it human, it's much More absorbable.

The advice is exceptionally good and a lot of it is just, you got to swallow your pride and not say what you were thinking and what your emotions were saying. Just be apologetic, accept your fate and you're fine. And you know what? An hour later, the other side's going to come back and say, you know what, that was really wonderful of you and that was very thoughtful. You know, it's partly my fault too. And, you know, and then there's some conversation, but there's no conversation when things are heated. Right. And this is true at work, too. How should I deal with a coworker who. And it's going to come back with some really, really solid advice and you're going to say, I would have never thought of that or I would never had the heart to do it. Just do it.

Yeah.

So look at this co pilot. Look at this assistant that you now have as such a miracle, not something.

That'S trying to take your job. Yeah, yeah, yeah. Your virtual assistant.

It's unbelievable. It's unbelievable. And he or she knows all. Why? Because it read a trillion phrases.

Yeah, yeah. More than a lifetime's worth, more than a lifetime's word.

We could read all that.

Absolutely. Yeah. Kevin, thank you so, so much. Just before we finish, then, obviously AI regenerating at an incredible pace. What's. If you had to summarize what's next? Where are the next leaps for this technology and its adoption?

In Gen AI, everything is going to multimodal. So that means you can input pictures soon as well as text, you can give it more information and it'll make a better judgment call. And also more of it will be connected to putting out pictures and drawings and illustrations. So everything's going to be multimodal. And then, look, we're in a hype cycle. And these hype cycles, this is true with the Internet, it was true with everything true mobile phones is that it gets overhyped and you think, oh, it does everything for me, and then you realize it only does some things for me, and then you fall into this trough of disillusionment and you're disillusioned by the whole thing. Well, that's an overreaction on the downside. And then you're going to come back and find it's very valuable and it sits right in the middle somewhere. Right? So I think get ready for this trough of disillusionment, but don't get disillusioned. Just use it for what it's good for. Use it for where it's valuable to you. And this is across the AI landscape. And I know you asked, should we be threatened by AI jobs? And this and that. One threat people think is, it's not only going to take my job, but it's going to kill us somehow. Because when I talk to it, it's all knowing, okay, do not hook a large language model to the nuclear arsenal. This is a really bad idea, Right?

Yeah.

But aside from that, it has no ability to kill you. It's simply doing what a spreadsheet does for math. It's just language. And so embrace this amazing language tool, and I guarantee you, three, four, five years from now, it's just going to be an everyday tool that we use, no different than our PC, no different than our smartphone today. Remember when smartphones were new? You go, what am I going to do with this thing? And then you get into it, and after a while, you, can you live without it today? No. Let me take it to the Internet. The Internet shows up in 1995. A few people get it on their desk, whatever, and you go, I don't know what this is good for. Okay, Today, if you're at an office and the Internet goes down, everyone goes home. They close the. They close the doors. You go home, you can't do anything. You can't process an order. You can't do research on who your competitors are. You can't create your next product. You're done. You shut the doors until the Internet comes back. I mean, that's how ingrained in our life it became. Okay, if you're in accounting today and Excel spreadsheets stop working, all accounting stops. We're done. Okay, in the next few years, if the LLM goes offline, you will stop working. You'll go, what do I do? I'd have to write this by hand, but that would take days. So why don't I just wait for the hour for the thing to come back up, and I'll be fine, Right? It'll be more productive. So it's just going to be a tool in our lifetime. And that's how we have to start teaching in school. We talk about this at college. I'm on the board of RIT quite a bit, which is, how should we approach this in the classroom? And some instructors are going, it is not to be used. It's verboten. Get it out of the classroom. It's a very bad thing. And others are going to. Actually, what we should be doing is teaching prompt engineering and editing skills, because it turns out these kids are going to use this. Anyway, this is now here. And you can tell them not to use it, but they're going to. So now let's embrace it and say, this is now part of your life and if we want them to write their own essay, we better give them a pencil and no computer and no Internet and let them write in class where we can watch them. Right. It's proctored. Otherwise, I'm sorry, they are going to go home and use a large language model to help them write their essay. And by the way, before that they were using Chegg, and before that they were using Google and stealing pieces. And then we put in all this software to say that was plagiarism. You know, come on. We've been chasing this problem, by the way. Before that we used parents. Parents wrote the essays. I'm sorry. I'm sorry to tell my teachers this, but sometimes parents help with the essays when the kid was stuck and it was due the next morning.

Okay, yeah, definitely.

And the math. So. So kids have been getting help all the way through now. The large language model is really, you know, heavy lifting. And it turns out in their, in their life in the future, they will never be asked to write an essay at work, ever. Yeah, they've. They're never going to be asked to do long division. They're never going to be asked to do an essay. So teaching that skill, I know some of your listeners are going to be horrified, but it's probably not a useful work skill in the future. If you find it useful at home, please write your own essays all day, write novels, write whatever you want. Nobody's stopping you at work. If you're going to hand write an essay. In fact, if you're going to hand write a blog post two years from now, you know, or type out of your mind, your boss is going to come up to you and say, what are you doing? We're not paying you to do that work. Ask the LLM, get three recommendations, pick one, edit it, and get it up in the next five minutes. That will be your job, not to write it from scratch. Who would pay you for that?

So the skill is going to be in the prompt of the line.

It's prompt engineering and editing. And understand, remember, you can ask for five versions and then pick the one you want and edit that. Right? So there is still a lot of creativity going on and a lot of checking, but to write every word yourself, I mean, certainly the people I know who have embraced this, they wrote their last word themselves, you know, three months ago. Yeah, I haven't done it since.

So interesting, isn't it? Kevin, thank you so, so, so much.

You bet.

Your insights and knowledge sharing. Wonderful. Thank you. Really appreciate you taking the time.

Oh, it was great to be on the show. Thank you.