Amdocs Group President Anthony Goonetilleke

Speaker A: Is what you're describing just a really, really advanced if then chain?

Speaker B: No, because an if then chain assumes that you know what the if is and you know what the then is. What if you don't?

Speaker A: Welcome to How AI Happens, a podcast where experts explain their work at the cutting edge of artificial intelligence. You'll hear from AI researchers, data scientists and machine learning engineers as they get technical about, uh, the most exciting developments in their field, the challenges they're facing along the way. I'm your host, Rob Stevenson, and we're about to learn how AI Happens. Oh, yeah. Welcome back to How AI Happens, all of you wonderful AI practitioning darlings out there in podcast land. Machine learning engineers, data scientists, software engineers of every ilk and persuasion. I'm so pleased you are here with me again because I have a wonderful guest for you today. I know I say that every week, but this time I really mean it. He is a man who has had a ton of different roles, almost all of them with the same company, which is rare to have a career man in this day and age. They have all culminated in his current position as the group president and technology and head of strategy for amdocs. Anthony Guna. To like, welcome to the podcast. How are you today?

Speaker B: I'm great. Thank you for having me. Glad to be here.

Speaker A: Uh, how about that? A whole career, 20, 20 odd years, 20 plus years at Amdocs. I'm, I'm so interested when that happens because it's so rare these days. But also, you must have had plenty of opportunities to leave. I'm sure you've been tapped on the shoulder and recruited and in, every, every time that happened, you made the decision to stick around. So I gotta ask, like, when that happened, when a recruiter, or probably another technology person, a CEO, a president, whomever was like, anthony, come on, like, join our team. And you said no. What kept you at amdocs?

Speaker B: Look, I love the company. I mean, I've worked for Amdocs for almost 26 years. Started in Australia, moved to the US, moved back to Australia, moved to the US all with the same company. I think the great thing is, you know, I get to do a lot of different things. You know, I joke around that I've done almost every role there is to do in the company. Um, and it's constantly changing and I think the company culture of just getting stuff done and the impact you have on customers, you get to see it, obviously, and connectivity is such a core part today of society. It's, it's exciting and you Know, I think when you're in that place and you love what you're doing, you know, uh, makes things, makes uh, decisions pretty easy.

Speaker A: Yeah, that makes sense. And something else I enjoy about your career is that you've always been a tech guy. You know, like this, this role as president of technology at this large company is not merely a stop on the tour of duty to ascending to a business role. Like you have always been tech focused, which is somewhat rare for someone in your position of seniority, I believe. And I'm curious now, like uh, having a whole career and you know, decades of just experience around tech, when you are monitoring tech as just like, you know, a nerd, as like a 12 year old version of Anthony who just is excited about electricity and power and technology, what is it currently that really makes you excited when you sort of take on the landscape of what's happening in tech right now?

Speaker B: Yeah, I think the, I uh, think the, the, the 10 year old Anthony to the Anthony today still gets excited about uh, geeky stuff. You know, I think uh, that's real fun. And, and I think today it's just this rapid pace of how things are evolving. You know, um, if 10 years ago you thought about something and you thought maybe it'll happen in five years, you know, now it just happens overnight. You know, I was, I was on a long flight over the weekend and I had my laptop with me and I was, I had this game. I always wanted to like find a developer and get them to develop it, right. And I'm like, huh, huh. I wonder if I can just do this myself and do a bit of vibe coding. I haven't coded in like 10 years. More than that, right? I pull my laptop on the plane, connect to WI fi. Two hours later I have a game up and running. You know, like this is the world we live in, right? Like, how cool is that? And so that's, that's, I would say that's kind of what keeps me excited. Just this rapid change, keeping up with it and uh, just seeing how it impacts, you know, the world around you. I think being able to see that transition from not just, you know, bits and bytes in my computer, but how it impacts people's lives. I think that's the thing that kind of keeps me going.

Speaker A: Tell me more about this game. I have to know.

Speaker B: Just, it's just, just a little puzzle kind of game I was messing around with in my head. But more than anything I wanted to see if there was like, you know, using these gen AI platforms today. Like if I could just code it and build it and I'd never built iOS application or never coded on a Mac. You know, my programming background back in the day was, you know, Unix and Linux and all of that kind of stuff. But just goes to show, you know, today if you want to do something in technology, it's very straightforward.

Speaker A: So were you using like uh, a low code or no code copilot to help you?

Speaker B: Yeah, yeah, I was actually using replit. I downloaded RELET on my Mac and I, you know, start going at it and it was funny, you know, I landed and then I Send a little WhatsApp to all my leadership team. I'm like, hey, you know what I did in the last two hours, I didn't look at. This is one person.

Speaker A: Yeah, I didn't look at the deck for the board presentation. But I. What, here's what I did. I love that, I love that you still have that uh, that desire to kind of get your hands dirty a little bit. And it is exciting. As for someone like you who despite having not coded in a while, you still know enough to be dangerous. Right. And I do think that's who is like most, most empowered by a lot of these low code and no code tools. It's not someone who has zero experience with tech, although that's happening too. But for the, the, the engineer with uh, a little bit really, you know, not break tags and enough to be able to generate their own print, hello world. You can do some really, really amazing things in, you know, in a domestic flight.

Speaker B: Yeah, for sure. I think it just democratizes the creative ability. Right. So I think, and I agree with you, you know, you need some basic level of understanding to actually go in there and do something. But then again, you know, I think if you have no coding experience and you wanted to build a prototype of something, you could very easily do that. And then you could go hand it off to someone and say, here's the prototype, now go build the code behind it. Right. So I think this whole democratization of the technology world, like I think previously it was kept for the geeks and only the people that knew it could play with it and they could tell you whatever. Now doesn't matter who you are, you could just, you know, I mean literally look at a, you know, two minute YouTube on how to do it. Go type it, use natural language to say what you want built, you know,

Speaker A: voila, you have something. Yeah, everyone's a geek now, Anthony. Yeah, that's right, you are, you are. Also, this is topical. So I'd be remiss if I didn't ask you about it. Uh, you're also fresh off the stage from Nvidia gtc. Um, I guess not fresh off the stage a couple weeks ago now, but I wanted to, to hear about this because you were speaking about agentic in, in your industry, in telecom in particular. But I just kind of wanted to get like a gut check from you. How did the talk go? And, and uh, you know, what was kind of the, the tenor of it?

Speaker B: Yeah, it was really good. I had myself a couple of other people from the industry and there was a professor from one of the universities and the whole notion here was where is agenti going and what does the future of this really look like? Right. And um, so we had done some research and our research showed that people are very willing to interact and talk to AI. Obviously it's better than the old standard chatbot that was there that really people would used to get frustrated with. But the interesting piece of the research showed that people would like to deal with more of a human type interface. So you'd come in and you know, um, I would come in and say, hey, like if I'm dealing with this service provider, I want to talk to an avatar or I want to talk to a female or I want to talk to a male and hear the characteristics. I almost kind of like when I was looking at this research, it reminded me of, uh, you know, back in the day when you used to have the Wii, right? Like the Nintendo and you used to make those, what were they? Me. The me. Me. Right, the me characters. Right. And people used to like go into, put so much effort to build these characters. Right? And you go, why? There was some attachment, um, some association with the character. Right. So I almost feel like the research kind of showed that, yeah, people would like to kind of build their own care agent and then have their own care agent deliver the results to them. And for whatever. Uh, you know, I'm sure some psychologists would have a good reason, for some reason they would believe this person much more than a random person that was given to them. And of course depending on age group. So we found that if you were like Gen Alpha or Gen Z, you'd just prefer to create your own avatar. You didn't even want to deal with a human like person, which also kind of, you know, makes sense. And then we found out, you know, 60, 70% of people would rather deal with a female avatar giving them an answer than a male avatar. So there's a lot to unpeel there But I think this is the evolution of going from simple chatbot type things that we've all used now and then and we're always like, hey, can I speak to a human? Can I speak to a human? To going to an intelligent information exchange where they understand the context of what you're saying to now going to like almost like a human like interface. You know, interesting topic because I just, you know, the announcement about Nvidia and Yum, uh brands for example, I don't know if you saw that, you know, they're working together on a partnership of kind of automating all of these drive throughs and things like that, which you know, this is the new era we live in. Right. So I think it'll be exciting to see where this goes.

Speaker A: Yeah, one doesn't need to conduct a ton of user research to know that we are really tickled by a digital facsimile. Right. Of an avatar. We're a narcissistic, we can't help it. And like it's even going on like as we record this, the one going around right now is that everyone on LinkedIn is like, is, is making a, using dolly to generate like an uh, themselves as an action figure and yes. Like, right. Like you're in the action figure box with your little accessories.

Speaker B: Yes.

Speaker A: So we're already doing it. The preference to, to speak to a human like avatar, is it empathy? Is it that it's better whether they prefer, they say they prefer it or not, it's better uh, performing because we are more patient with a human being than we are with a machine. And like if you walked up to, if you're at a Starbucks and you say your order and the, the person working the register is like, sorry, it's my first day, you'd be like, oh, that's okay, you know, whatever, it's a cup of coffee. Whereas if you're like typing on a ah pad and it like starts loading, you're like, start hitting the machine. You're like, come on finger, hurry up. So I wonder if like that part of it work, we're playing into human empathy a little bit. Like there's that need to feel like you're connecting.

Speaker B: Yeah, for sure. I think there's some, you know, deep psychology there about the human empathy side. And also I kind of, I kind of feel like once um, you create the avatar, you almost have some ownership, like some relatability to. Because it's yours. Right. And maybe, just maybe it will act on your behalf. Right. Because it belongs to you. Right. So it's like an extension of like it's your assistant, your personal assistant in that company. So it's a, it's an interesting. There's a lot to unpeel there. It's a little bit of Freudian stuff going on here, but I think it would be very interesting. By the way, back to the action figure. I posted something on my Twitter account the other day on. Basically, uh, there's all these companies that have come out now and put out apps where you pay like 4 99, 1099 to create this action figure. Right. And it's basically a very simple prompt that you can put on chat, GPT or anywhere and create it. But, but this also just goes to show the connection on how people are trying so hard to make AI with even maybe unknowingly more human.

Speaker A: You mean because they're gonna pay for an app that automates it rather than figure out how to do it themselves? Or just because it's the compulsion to create the avatar.

Speaker B: The compulsion to create the avatar.

Speaker A: Right.

Speaker B: Like, like what makes you go create a toy that looks like you using generative AI, Right, yeah, there's some connection there between, between the symbiotic nature of somehow creating a digital being in some shape or form around in your head. What you think it should be like, you know, there's some direct connection there.

Speaker A: Yes. Yeah, it's, it's, it's, it's like a God complex, particularly with the, the designing an avatar, even if it's not a facsimile of you. If it's for example, your, your agent, your assistant. So are you saying that, like the, the Nintendo Wii example is instructive? Because I do believe the first time you ever fired up your Wii, you plug it in and then you turn it on and you're prompted to design your avatar right away. It's the first thing. Would it be the same with Agentic? And your company pays for the software, the vendor's approval, blah, blah, you go to fire it up and then before you can start asking this thing to write code for you, what color hair should it have? You know? Yeah. Design a sim. You design a me and are participating in the creation of this agent. It's not just this mysterious thing being handed to you.

Speaker B: Yeah. And this is exactly the way we think about it. So platform. The first thing you do when you go to a, uh, care call that you haven't been to before is you basically get these three boxes like, hey, like, how would you like your care agent to look like? So it's like you've got your complete digital avatar, you've got a male or female, then you pick one and then you go, what type of tone of voice would you like it to have? Right? And you pick the tone of voice. What type of voice would you like it to have? So you're almost giving this thing life, right? So uh, you are the creator, you get the pride of authorship because you created it. And in a way you will hopefully be more empathetic, you will hopefully interact with it a little bit better. Now you could have probably got the same answer without any of this stuff in the front end, right? But at the end of the day, we're human and we like to do this stuff, right?

Speaker A: Yeah. It's like you're Dr. Frankenstein and you get to pick the tone of voice and then click next enough times and it's alive.

Speaker B: That's right.

Speaker A: This is so interesting because it's non technical. This is just like user experience, but also psychology. The AI practitioners who I interview and who listen to this, they understand, they understand the opportunity. They don't probably don't need an avatar who looks like an ex girlfriend to listen to it. Like they, they don't need that. But what they struggle with is, is the, the company buy in, is the user buy in. How do we change behavior? How do we make people want to use this and, or even just buy in in your own company, in your business units. Right. And so selling this thing internally, getting people to use it is a often, ah, an equal challenge to designing it in the first place.

Speaker B: Yeah, yeah. And if you think of, you know, that was a fun part, now you think of the intelligence behind it. Right? So let's compare like a human agent and a avatar agent in a telecommunications provider. So the human agent probably knows nothing about you. So if you're calling in complaining about your bill being high or whatever, they need to go back and look at your last month's bill, see where it is, maybe look at the last couple interactions. Your avatar could know everything in an instant, right? Could know when you got a service, when you canceled it, that, uh, eight months ago your child moved college and went somewhere else and know all of this information. So, so you're backing your avatar that the person created and embedding all of this knowledge historically over even multiple years. Right. And just making them advance. So I think the two coming together could potentially have the chance to create some magic. And I think some of the early feedback we've gotten, uh, from actual end users was really funny. We really love It. But can you make it more empathetic? I think there's something here, there's something very deep here that people want, at the end of the day, people want to feel listened to and heard. Right. And they want to know that the person talking to them understands where they're coming from, what they've been through and can relate to it. And I feel like we're kind of pushing the boundaries here and it'll keep on evolving, but this notion of creating a agentic behavior and then adding a level of EQ based on what you've been through, I think will make it more attractive to users at the end of the day.

Speaker A: Certainly, you know, ah, at the top you spoke about how excited you are that this, this industry is moving so quickly. And what we're sharing now, or what you're sharing now is a year and a half ago this was just a chatbot, this was just text, right. And now we're talking about an, a live conversation that has audio, that maybe has a video component, that has more human like qualities, has like, we're talking about emotions, we're talking about empathy, we're talking about tone. It's not just racial expressions.

Speaker B: Right.

Speaker A: A pop up window in the bottom right of your screen anymore. It is more like asking your assistant a question, which is how it was always built. It was like, think of this chatbot like your assistant. It's like, yeah, but you know, I, my assistant is more than a box in the bottom right of my screen often.

Speaker B: Yeah, you're absolutely right. And you know, Chatbots by design was built essentially on a workflow, right. So it would go, you know, Anthony said this, so here's your response. Like kind of decision tree workflow, but generative AI is very different. Right? Like this is, you can pretty much go anywhere you want to go and it will pull the information together and deliver you the most appropriate response. I mean we are getting accuracy levels, you know, in the high 90s and I would debate that, that you know, humans probably couldn't provide that level of accuracy in terms of depth and accuracy in terms of some of the answers today, probably not.

Speaker A: But this is, it's not dissimilar from the self driving car where it's like, it cannot merely be as good as humans. It's expected to be perfect or, or much better. It changes depending on the stakes of the task, I suppose. But for your average user, it needs to be far better than an average human to be accepted. No?

Speaker B: Yep. Uh, for sure. Right. Like, I mean if we're Just recreating the benchmark. I think you kind of fall short. Right. Um, in my example before, you know, we did, we kind of did a benchmark and um, the average call center agent took 12 to 14 minutes to review your last bills, come to a conclusion and deliver the results. You can deliver the same information in a more articulate, accurate MANNER in about 45 seconds.

Speaker A: Mhm.

Speaker B: Right. So there's, there's definitely. And the person is. If you, if you measure. In the industry we measure something called nps, which is like Net Promoter Score, which basically is. Are you willing to like, are you happy with your service? Are you willing to recommend me to someone else? Right. This is basically the overall satisfaction index. And that was much 50% higher than interacting with a human agent. Why? Because the answer was accurate. It knew what it was talking about. It was clearly articulated and it was fast.

Speaker A: Yeah. And you didn't have to listen to any, uh, boring hold music.

Speaker B: You know, all of that being said, I think the roles of what humans do will evolve to different facets and different aspects. Right. So I think I always ask myself just in my daily life, when I have a task, I asked myself, you know, like, what is my highest and best use? Like, what is. Could I be doing something better with my time? Right. In terms of my roi? And then it's, uh, a very simple, like a very simple question. Right. And I feel like how we need to think about technology is. In that way. So there's a bunch of stuff here that can be done. So what else can, you know, can you maybe drive sales? You know, more customer facing time? So, so you move humans to the next level. Because I always think there is a need for it. It's not a, not necessarily a replacement per se.

Speaker A: Yeah, yeah, certainly you, you move yourself up the skill tree a little bit and focus on only those more complex acts. Certainly. I'm glad you mentioned that, Anthony, because look, we have to stop goofing around here. We've got to start the show. All right.

Speaker B: We haven't started yet.

Speaker A: No, no, no.

Speaker B: Having fun. Yeah.

Speaker A: Uh, look, the, it was not an accident that you were on stage at Nvidia, because Amdox has been partnering with, uh, Nvidia for a good long while here. And the thrift of the partnership, as I understand it, was uniquely exciting to me because you're focusing on the layer above hardware. And I had a recent episode with a, uh, fellow, Jay Diwani is his name, um, founder of Lemurian Labs. And he was telling me all about how the need for compute is going to so far outstrip the possible production capabilities of hardware. If we're going to win, it has to be software. Software has got to be more efficient. And this is before all this stuff about the tariffs and like, how are we going to, you know, import more and more semiconductors and blah, blah, blah. So I think he was probably right before and more so now with all the uncertainty on import export. So I would love to hear you reflect a little bit on that partnership and if you agree or disagree with Jay's assessment that whoever is going to win is going to have to really figure out software because it cannot be a matter of fabrication.

Speaker B: Yeah, I don't necessarily think they're in contradiction. So maybe that's where. May I have a slightly different viewpoint, you know, ah, there's a paradigm, um, called Javon's paradigm, right. Which basically says if something gets more efficient, more of it will be used or something. Something along those lines. For sure, software will get better and better, more efficient. I saw, um, Microsoft put out something today about running their models more efficiently. We have all the deep SEQ stuff coming out. I don't necessarily think just because software becomes better, you will need less hardware. You will just find more uses for it. You will just use it in more ways. Right. So I think the two go in parallel. So I think hardware will be better, cheaper, uh, faster, uh, low energy, and software will continue to kind of go up the food chain. Because with software, yeah, now we've now gone from, you know, let's say two years ago, everyone was so excited that ChatGPT could just, you know, you could say, hey, write a letter to my friend and could write a letter to your friend. Right now you're like, well, how do I automate this? And can I orchestrate it? And you want like some real output from it. Right. And so I think the next level of software is really going to make a change in terms of the practical side. So there's a lot of physical AI that is happening, meaning, you know, that the best example, obviously, are the robots that you see advancing every day. You know, think of like the roombas of the world, you know, a house, right. Are going to do more and more and get more tasks associated with it. And, you know, but I think this incorporation between the physical world, the software world and the hardware world are going to come together and all three are needing to move in parallel. Because at the end of the day, we're going to measure productivity from a human perspective. And from a human perspective, all three need to work in parallel. So this is why I don't think, forget it. Hardware is good. We'll just leave it on the side and just work on the software. Forget it. Software is good. Let's work on the physical side. Like all three are going to need to advance in parallel.

Speaker A: Certainly. And when you speak about physical integration, are you talking about obviously, Iot the example of Roomba? Are you speaking about like wearables or what. What is the. What is the disruption that we will see, you know, when we're afk?

Speaker B: Yeah, look, I think, I think, um, it's a spectrum of things, right? So obviously you spoke about cars and you know, anyone who's gone in like a wemo or something like that, the first time you get in a wemo, when you're getting it, it's like, whoa, you know, like the first 60 seconds, you know, is like, it's a freak out moment, right? And then you forget about it. Like within two minutes you're like, you're gone. You've forgotten about it. Right. You know, we have Roomba in our house and it goes and vacuums the whole house and goes back to its spot and you don't even think about it. Right?

Speaker A: Yeah.

Speaker B: And I think this seamless integration into our day to day lives will happen more and more. But I think the magic is going to come when tasks can be stuck together. Now it's like an individual kind of task. Like I can get something to do a. Like my. So one of my neighbors actually has this electric lawnmower that runs around, you know.

Speaker A: Fantastic.

Speaker B: Yeah, it's kind of pretty cool, right?

Speaker A: But like, imagine like a Roomba for your lawn.

Speaker B: Yeah, it's like a Roomba for your lawn. It's like I was looking at like I drove in the other day and I'm like, what is that? That's pretty cool. You know, it's like, ah, it maps the thing and it goes and cuts the lawn and does all of the stuff. But now you get to the stage where you start sticking like chaining events together, right? Things that you need to do, things that you need to get done. Um, uh, I'll give you a very simple example. My daughter came to me the other day and she's like, hey, Dad, I need to. Here's a drawing that I need. I need to print this in. I can't remember what the amount was. It was huge amount, 15 by 26. And obviously our paper was, you know, much smaller. So she needed to print it on multiple sheets of paper and then stick it Together. And she's like, what's the best way to do this? Like, which application? Like, she'll use Photoshop, cut it up. Like, uh, it's like going all over the place, trying to think about it, right? I thought about it for about a minute and I'm like, maybe, you know, I'll go and ask, uh, ChatGPT on how to do it. And so ChatGPT was amazing because it told me, hey, like, do this, do this, this, this in Photoshop. Then I would not use Photoshop after that. I would go into Adobe Acrobat because this is much better to create banners. And here are the changes I would make. I mean, it worked beautifully. Okay? Like, it probably would have taken me an hour and a half to research. And I got the answer in two minutes and it was printed out and done. But then I thought to myself, like, what if I had an execution button and it did all of this stuff? It went into Adobe, did everything it said, went into Adobe Acrobat, pulled it up, found my printer, printed it up, and it was there. That's the magic that's happening now, right?

Speaker A: That's Agentix, right? That's what we, that's what we expect.

Speaker B: That's right. That's the orchestration that's happening now on stitching all of these, like, what we used to think as disparate events coming together. And there are, you know, there's anthropic and there's a whole bunch of companies that are looking at stuff like that. Like, how do you do all of these tasks and take this intelligence and embed it and create it? I think that's the, that's the next level we're going to see in the next year also come together with physical AI. Like, you know, my, my house, going back to your geek comment, you know, is fully automated. I have, you know, hundreds of IP addresses controlling a whole bunch of stuff in my house. But still, you know, I have a app for this, I have an app for that, I have an automation app, I have this kit, I have the home kit. At some stage, you need some cohesiveness or you need some glue to bring all of this stuff together and then add to that intelligence. And that's kind of where the magic happens.

Speaker A: Is what you're describing just a really, really advanced if then chain?

Speaker B: No, great question. Because an if then chain assumes that you know what the if is and you know what the then is. What if? You don't? Okay, so if you think of the algorithms of how generative AI works, Right. It is not given an if and a then necessarily. Right. So it is using, like, probabilistic determination to figure out what the next action should be. Right? So this is why generative AI is so amazing. Because as a program, if I'm writing, you know, going back to like, some code, right? If I'm writing a bit of code that says if, then it says, if Anthony turns up to his house, then turn the light on, right? Very specific, right? But I have to know what those conditions are, and I have to tell it what to do. But what if it can learn? What if it can determine by your behavior? What if there is a completely different way to approach it? And this is, I think, what generative AI brings to the table. And this is, I think, what excites people.

Speaker A: So it's the reasoning part is why it's not if, then it's not fair

Speaker B: to compare it to kind of human thinking, right? We are obviously far off from that. But all of a sudden, when you get these inputs coming in today, if you think of a large language model, pretty much 90% of the world are using large language models by giving it, what, text input, right? A prompt, some, maybe an elongated prompt or whatever. But now think of getting vision, getting sensory data, getting smell translated to, you know, there's a smell of smoke, like, what do I need to do? Right? All of these types of information as input, then you start to create something really that. That's kind of amazing. And I think this is already happening today. It's not like, you know, this stuff is not happening. And you see, especially when you look at the advancement of robotics, I think we are not that far away from having a robot made in your house, right?

Speaker A: We're really not that far away, Rosie, from the Jetsons.

Speaker B: That's right. The.

Speaker A: I'm glad you. You called out human cognition because that was what I was kind of getting at. There is not just like, hey, explain how generative works to me. But the point is like, okay, what are we approximating with generative and agentic? Isn't it human cognition? Isn't that what AI is all about? Aren't we approximating human cognition, but the way a human brain works and the way we train a machine? Are we limited by our understanding of how learning works? How, like, we as humans learn and how we make decisions? It is, isn't it just endlessly complicated? If then every action I take, every time I exercise my free will is taking in a bunch of factors. If I see X, Y, Z, then abc, right? And that Feels like we are couched in that we're stuck in this human learning version and we're training machines that way. Is that where we are right now? And is that a problem?

Speaker B: It's evolving, right? I mean, I think what you kind of outline is the fundamentals of an algorithm at the end of the day, right? But as you train these, right, we have this term that we use called emergent behavior. Emergent behavior is almost not a trained behavior. And you see if you Google it, you see all these weird random examples of some behavior that popped up and you went, huh, huh. That wasn't something that was ever taught, right? But this is the part where you start approaching those boundaries of going from the if, because the emergent behavior is not part of the if then combination that you thought about, right? And you see, you see different models start to have various emergent behavior. We have seen large language models solve some problems where we know that this was never a trained behavior on how it got to. Got to the conclusion. Now, obviously, you know, very early on, everyone is very concerned about hallucinations and that aspect of it, right? Because if you are now orchestrating an action based on some generative AI output, you have to be very careful because there's a physical impact, right? It's one thing to get something textually that's wrong. It's another thing to act on it and do something physically. You know, I was thinking the other day, we have these ceiling sprinklers in our house, right? And so if one of those goes off, the amount of damage it causes in the house is like. It's. It's crazy, it's ridiculous. So I, um, was trying to, like, do some analysis in my head. I'm like, when. What's the ROI on this? Like, is it better?

Speaker A: How much fire damage would it have to.

Speaker B: Yeah, like how. Exactly, right, like, exactly. Like, how much of a fire does it have to be? And if there is a fire and the sprinkler comes on, well, it's, it's all gone anyway.

Speaker A: It's worse. Yeah, exactly, right?

Speaker B: Like, so what am I really stopping here, huh? Because we had a, we had a freeze in, in Dallas, believe it or not. You know, several years ago, we had this crazy freeze. And, and the sprinkler pipes aren't, um, built for it, right? So one of these sprinkler pipes burst. I'm telling you, the amount of water that poured in was crazy, right? Because these are high powered. So I'm trying to think of, you know, again, going back to the geek stuff.

Speaker A: Right.

Speaker B: Like, I'm going, hey, what are the conditions that you can input this? So you can have some smart algorithm on when it can detect when it should go off or it shouldn't go off? Right. So, for example, if my family or if there are people in the house and if there is a fire, maybe it should go off because there's people there. Right. If you only smell smoke, well, maybe you should call the fire brigade instead. Right?

Speaker A: Yeah.

Speaker B: So there's all of these, like, intelligent thought processes. And so this is why I think, again, kind of tying all of this back together. These are some of the things I think that the future will start to solve for us. We haven't even, you know, touched on quantum computing. And potentially what that will bring to the table, that's like a whole another podcast, I think. But all of these are, uh, all of these are coming in together very, very fast. And there's so much money, obviously, I keep tabs on kind of what's going on in the industry, and there's so much money being invested in this space right now that I think what we deliver in the next year will be close to what we delivered in the next last five years. I think that time frame is just shrinking.

Speaker A: Is emergent behavior an edge case? Is it a hallucination? How is it different than those two things?

Speaker B: I don't think necessarily it is a hallucination. From that perspective, hallucinations, in a way, is coming to a conclusion of something where the conclusion is potentially catastrophic or wrong. This is the generally how we associate hallucinations, emergent behavior. I think more of two data points being put together and saying, hey, I never trained it on this, but based on this data, here is a way I would deal with it. Right. This is not something necessarily the model has been trained on. So I don't think necessarily the two are, uh, uh, categorically the same. And maybe technically, maybe there is some overlap in there somewhere, but emergent behavior

Speaker A: just has a positive value judgment. Is it the same, exact same result? It's just, uh, this is a plus instead of a minus. Yeah, but then is that what we're indexing on? We're indexing on emergent behavior because that is like true reasoning, decision making that does not have to be babysit.

Speaker B: Yeah. And this is also going away from that if. Then use case.

Speaker A: Right.

Speaker B: Because if we were always depending on this. You're only as great as what your programmer can think of. Right. You know, back in the day, when you think of like, uh, writing, uh, Error code management kind of software. Like you need to think of all of the type of errors and you need to then write, well, if it's this error code, then say this right? Where else today you can say, well, if this is running and this is done, here's potentially what the problem is. So even in a simple example like that, working from a technology perspective on a completely new system, using generative AI on how to manage errors around a system, because you're not just looking at a single error code and saying, well, if it was 0076, this is what it meant, for example, because today you can look at, well, what was my CPU usage when it happened? What was my memory utilization? Was there traffic coming in? What was this doing? And based on all of this, you can say what the most likely scenario is. And so I think those are, uh, some of the things that people are really going to tap into going forward.

Speaker A: If emergent behavior is a sweet spot of reasoning and approximating cognition, does it pose a problem for explainable AI?

Speaker B: Look, I don't necessarily think so because, for example, when we are doing a lot of things, we have an audit trail. Because I think it becomes more and more important to have an audit trail. Right? Um, even for digital content, sometimes I think you need to have. You know, I think the world is very fastly evolving that today it's very difficult to tell if a video is real or not, right? It's super, super difficult. I saw a video floating around the other day. I think it was Ronald Reagan saying something and I was like, did he really say that? But those are, those are real things. And I think when it comes to AI, when people are building platforms and systems, in our case, we need to have an audit trail. We need to know what data you look like, what data you tapped into, how you came to the reasoning. Um, that whole reason chain needs to be understood and needs to be transparent. And I think lots of companies are building these types of technology into their platform. It's, it's not just. And by the way, sometimes like in our platform, we go to multiple large language models. So we may take an output from one and feed it to the other and correlate it with something else. So there are multiple checkpoints. You know, it's almost like one checking the other in a way. Uh, one of the guys that works for me was saying, hey, like I created some code in this and then I got this large language model to go and verify if this code was correct and how it should improve it. So I think there are a lot of gates that need to be put into place and guardrails, I would say. Right. Trusted AI, I think, is going to take front and center stage in many, many ways going forward, certainly.

Speaker A: So emergent behavior does not necessarily mean it's like a black box. There are still guardrails you can put up. You can still dig in and show the work a little bit. Okay, good. Well, Anthony, uh, before I let you go, I'm going to do something very non technological, very non artificial intelligence.

Speaker B: I love it. Let's go.

Speaker A: Because you have lived this tech life and you have such, uh, a. When you were explaining to me your home life with your, um, your, your, you know, your sensors and your, your robots and Iots around your house. I want to know how you touch grass. I want to know, do you do yoga? Do you meditate? How do you. How do you unplug?

Speaker B: Yeah, so I unplug in, I would say in a few different ways. It's a. It's a great question. Um, obviously family is a very, very important thing, and I think having kids always grounds you. Um, so if I'm really into tech and gen AI and things like that, um, you know, like, both my kids, um, you know, they're 18 and 22, and both of them are super skeptical around generative AI and everything that it does. The stuff that came out the other day around the, you know, people were posting all those kind of the studio Ghibli images, right? And you know, like, my daughter just rolled her eyes. I'm like. She's like, you can't do that. Like, that is someone's artwork. That is someone you cannot.

Speaker A: That is.

Speaker B: That is not right. So I think that next generation also has a very clear perspective on when it's right and when it's not right. And that type of stuff forces me to stay grounded because I'm always debating, you know, uh, but I, I love tennis, you know, really like, into it. So I think it's always important to just find something that allows you to kind of unplug a little bit and stay away from tech. I try not to wear an Apple watch on purpose so I don't get buzzed and, you know, not. Not exactly. Exactly. I love. I love the old analog watch. And, uh, everything is electronic. And people always ask me, like, why don't you have an Apple watch? I'm like, it's intentional.

Speaker A: It's not because I haven't heard of it.

Speaker B: I do have one that I wear, you know, when I exercise or something like that. But that's for, again, that's for a purpose. And other than that, I don't need to be connected every single second of every day. And I think what you raise is a super important point, right. I think every one of us needs to take a step back and go, just because you love tech and just because you're a geek, you shouldn't lose touch on the human side of life and evolution and people at the end of the day, right? And I know we spoke about empathy, and I always tell myself, one day I'm going to write a book, you know, and. Because I think empathy is one of the greatest human superpowers that ever existed. Right. I'm not sure that generative AI, how much we try, can come close to being truly empathetic. Maybe it can. Maybe that's the next frontier. But, um, you know, I think it's a great question and, and you have to force yourself sometimes.

Speaker A: You, you every time. Because. Because it's, it's so ingratiated in our lives and it's such a great reminder, especially coming from someone like you in your position, who would probably benefit from people never unplugging. But, uh, you know, there are more important things as you call out and uh, yeah, to, to the reminder to, especially for this crowd who is so tech forward and so tech savvy and conscious, to touch grass and to unplug is important. And now that we're at the end of the podcast episode, I can tell people that they're officially allowed to unplug and disconnect not a minute sooner. But Anthony, man, this has been a really fun episode. Uh, this is my favorite kind of of conversation where we just kind of go in all these different directions. So at this point I would just say thank you so much for being a part of the show and for your candor. I've loved this conversation or is great being here. How AI Happens is brought to you by Sama. Sama's Agile data labeling and model evaluation solutions help enterprise companies maximize the return on investment for generative AI, LLM and computer vision models across retail, finance, automotive and many other industries. For more information, head to sama.com.