2807: The AI Pathway: Nikola Mrkšić's Journey from Serbia to Silicon Valley

[00:00:00] Welcome back to another episode of the Tech Talks Daily Podcast. I'm your host, Neil C. Hughes, and today we've got a fantastic guest joining us. He is the CEO and co-founder of a company called Poly AI, and they're a company at the forefront of conversational AI technology.

[00:00:19] But there's also a fantastic origin story here that is truly remarkable, and his background and the pivotal role he played in shaping Apple's digital assistant. Yeah, I'm talking about Siri! But it all began by selling books on the streets of Serbia to earning his PhD in machine learning

[00:00:38] and natural language processing from the prestigious University of Cambridge. This is one of those stories that has got it all and is incredibly inspiring. You're going to love his incredible journey from humble beginnings to becoming a lead voice in AI and technology entrepreneurship.

[00:00:56] Now, before I get today's guests on, it's time for me to mention the sponsors of Tech Talks Daily. And in an era where digital security is non-negotiable, legacy managed file transfer tools, they simply don't cut it now. So that's where Kiteworks comes in, revolutionising the MFT

[00:01:15] landscape with unparalleled security credentials, including the much-coveted FedRAMP moderate authorisation. This isn't just about compliance, though. It's about offering a secure, efficient platform for today's remote workforce. So with Kiteworks, you can benefit from advanced file sharing, email security and customisable integrations, all within a platform designed to

[00:01:38] safeguard your most sensitive data. So don't let outdated technology compromise your security. Step into the future of secure managed file transfer. Get started today by going to kiteworks.com. That's kiteworks.com, where security meets sophistication. But now it's time to get today's

[00:01:57] guests on. So buckle up and hold on tight as I beam your ears all the way to London here in the UK, where today's guest is waiting to share his story.

[00:02:08] So a massive warm welcome to the show. Can you tell everyone listening a little about who you are and what you do? Hi, my name is Nikola Mirkovic. I am one of the co-founders and the CEO of PolyAI.

[00:02:20] We build incredibly human-like voice assistants for enterprise clients. So kind of like, you know, machines that pick up the phone and answer the way that your best agents would to help you have much better customer experiences. Awesome. And one of the things I always try and do

[00:02:35] on every episode is get the guests talking about their origin story. What was that moment that put them on this path? So in your case, can you share some of your pivotal moments during your time?

[00:02:45] I think you began selling books in Serbia and something happened there, propelled you towards a career in AI and technology. And it feels like there's quite a story there of your own back story.

[00:03:00] Wow. I think that's some deep research. I think the anecdotes of selling stories was when I was 12. I'm Serbian, so I grew up in Belgrade, the capital of Serbia. And when I was about 12 or 13,

[00:03:16] I wrote a bunch of kind of like Star Trek inspired sci-fi stories and I printed a bunch of them. I started going around like the cafes, Belgrade has quite an active kind of like cafe culture.

[00:03:29] So I started going around and pitching people and like, hey, like I've written some stories, would you like to buy one? Right. And I had like bundle packs and everything, which kind of tells

[00:03:36] you that I was always more of a salesman than actually a writer because I spent a lot more time selling the product that I had worked on for probably a few days. But it was really interesting

[00:03:45] because at one point I met like a few of the like really, really popular actors. I spoke to some journalists and at one point one of them was like, hey, let's make a show about this. This is like

[00:03:55] crazy because I think there was a day where I made like the equivalent of a monthly Serbian salary in like one day. And I did it for like two, three weeks until I was like, okay, like

[00:04:06] I bought a computer that I wanted to buy and I was like, okay, I guess I'll stop now. You know, school starting again. That's a good research. But yeah, other than that, I was actually a very studious kid. I went to a highly specialized maths school that,

[00:04:22] you know, kind of like prepared you for maths Olympiads and stuff like that. And then I'm going to Cambridge on a scholarship to study computer science and maths had the moment where I was rebelling against it and almost left to do investment banking instead.

[00:04:36] So rather than being a nerd, I was going to be a finance nerd, which is not really all that creative at the time. I felt like it was. But then I met these guys starting a company

[00:04:48] that was building voice assistants. And you know, it was a different time. It was 2014, like Siri existed. Alexa wasn't yet there properly. And it was a brave new world of these things. I got

[00:05:00] excited about what they're doing. So I joined them. We ended up acquired by Apple about a year and a half ago. I, subsequently I did a PhD with one of the co-founders, a guy called Steve Young,

[00:05:11] who is one of the greatest researchers in the field of speech recognition. And you know, it went from there. I met my Taiwanese co-founders there. And we were really lucky in that we were

[00:05:23] working on deep learning as it was really taking off for spoken dialogue systems or conversational AI before anyone else. So kind of like classical, right place, right time, right people. And I think

[00:05:39] we just became really good friends. And at the end, we started a company and do this for a while. And I think while these consumer voice assistants like Siri and the others are interesting,

[00:05:50] they're not essential. Like if we removed Siri from your iPhone, you still buy an iPhone. It's a killer product, right? Whereas like when you think about customer service, that's where there's a bit of automation out there and it's not very good. So people really hate it. And

[00:06:07] you know, if you hear an automated thing on the phone, like your first reaction is probably just to try to find a way to circumvent it and get to a human. But you know, having worked on this stuff

[00:06:16] for more than a decade, like we know how to build things that are far better. So for us, like, you know, that's the mission and goal in life. And it's kind of my life story. Love it. And obviously

[00:06:28] you mentioned Cambridge that we're both talking in the UK today. So I'm curious, how did your experience in education at Cambridge, how did that play a role in shaping your approach to machine learning and AI development? Was it the institution, the education itself? Was it the

[00:06:42] people or was it just a big melting pot of all those things? Probably that last bit, right? I mean, Cambridge is a great place. A lot of really good people work there and from all over the world.

[00:06:56] You know, I think that experience was very interesting. I mean, undergrad was one thing, uh, kind of a graduate studies completely different. Um, and I think that, um, you know, it's been a center for, for this area for awhile where like a lot of people have worked diligently

[00:07:14] to, you know, create a framework and a methodology for building these things. Um, so, you know, we're kind of like when we started, we already had a lot of the high level system design charted out.

[00:07:26] Right. And then what we brought in was this, you know, like new age deep learning that kind of like uses data to drive everything rather than like, you know, uh, handcrafted programming.

[00:07:41] Um, but that's maybe a bit too detailed for the audience, but, um, yeah, it was basically like a desire to create things that would learn from data that's out there already. So kind of like

[00:07:50] conversations that we have between us, between, between us and machines, right. To create machines that can have better conversations with us. And from there, of course, you transitioned from academia to technology, entrepreneurship, and you mentioned your company was acquired by

[00:08:06] Apple last year, but I'm curious, looking back, what were the most significant challenges that you face and how did you overcome them? Because it's almost like a 10 year overnight success story. And very often we celebrate the success at the top of that mountain, but not, not those,

[00:08:21] the school of hard knocks, so to speak. Can you remember any significant challenges from back then? Well, I mean, like it wasn't last year that company was acquired just a year after, after it started, right. That was, and, um, we started poly AI, um, a few years afterwards

[00:08:37] because we kind of like really wanted to focus on the enterprise segment and building voice assistants for other companies, right. Things that would pick up the phone for them. Um,

[00:08:45] so we'd probably, uh, we've been doing it for six years and a bit, so like six and a half almost. And yeah, I mean, that's been like full of all sorts of challenges. You know, now we work with

[00:08:55] you know, a hundred different companies around the world and the U S and the UK and Europe and in Asia and Latin America. Um, it's hard, right. And I think like, you know, it wasn't obvious that

[00:09:06] we're going to get here, I think from the very start, because I think a lot of people have good technology, but don't know how to sell software. I think for me, one thing that was definitely

[00:09:15] like a big learning curve was like, it doesn't really matter if your technology is good or not. If you don't know how to like convey it to the right people, like why this matters,

[00:09:26] because they'll probably believe you when you say that the technology is good. Like you are an expert, you worked on it for a long time. But like, so what, right? Like, will it be used by

[00:09:34] everyone? Do you know how to like convince others to go on their journey with you? Like that's all difficult. Right. And, uh, I think I had to learn a lot about, you know, just like selling software,

[00:09:44] implementing software, building bigger teams that would kind of like run all this. So I'm still learning. I think we're still, you know, probably at the very start of like how big, how big this is

[00:09:53] going to get. How do you see the evolution of digital assistants though? And when I say that, I mean, in terms of their ability to understand and process human emotions and context, because it's something that they've struggled with over the year, but how do you see that evolving?

[00:10:09] Yeah. Well, look, I think that, uh, we're undergoing like a massive, massive surge in the capabilities of artificial intelligence. And that includes, um, digital assistants, right? Voice assistants or, or, or chat bots. Right. Um, especially for voice, it's a very rich channel.

[00:10:27] You can tell a lot from the way that a person is speaking, like, um, especially if you know them well, like, are they angry or shy? Are they like stressed? Are they timid? Um, are they frustrated?

[00:10:39] So like we're getting better at all that. Right. And I think the depth of conversation, the depth of like, just understanding you and the context you're in is also accelerating rapidly where a

[00:10:48] system can, you know, look at everything. If it has access to look at everything you've ever done in life, everything you ever bought on Amazon, everything that you might've ever like typed out, right. Obviously that, that, that, that's not available to them, but like they have access to

[00:11:03] a lot of information, but a lot of other people where if you think about how like advertising is really good at, like, you know, if you speak to someone about something at their house, right.

[00:11:13] And you were on their wifi and they've been browsing a certain kind of product. You're probably getting an ad straight away because you were associated with them. Right. Now, if you think about a voice assistance, we can use that to greatly personalize the way that

[00:11:24] they should work for you. Right. If we figure out what kind of person you are, what your preferences are, you know, that assistant you have on your phone might finally understand how you like to get things done. And then maybe those consumer voice assistants will finally be able

[00:11:37] to do things like, Hey, like book me a haircut. And maybe it understands that I wake up early. So if there's an available spot at 8am, I'm not going to be upset. Right. Whereas another person

[00:11:47] might get really cross if the assistant has booked with them in for like a very early time. But it's a story of context. That's really difficult because you know what I've always found

[00:11:56] interesting during my time at Siri was, you know, I tell people, Hey, I, you know, I work on Siri and people from all kind of walks of life will look, maybe I just met them. They would just look

[00:12:06] me in the eye and be like, Oh, Siri, Siri is shit. And I was like, excuse me. Right. Um, and like, I'm quoting verbatim because that's mostly how it happened. It happened so often

[00:12:15] that like, it became a real like sticking point for me because I was like, most people are not even saying that because they're trying to, you know, provoke you or anything like that. It's

[00:12:26] clearly just like a reflex reaction, right? That's how they feel about it. And I was like, well, why do you feel that way? What do you want it to do? And the real answer is if we're just to write

[00:12:38] the perfect product specification for Siri, there, there isn't one, right? Like people can't tell you conclusively what they wanted to do. So saying that it doesn't work, it's kind of like saying, you know, I'm using a supercomputer and it doesn't work by trying to, you know,

[00:12:53] solve a problem in physics or crack, uh, you know, some password or I don't know, like mint Bitcoin, like we got to set like the goalposts. And I think with, with, with assistance, it's unclear what we were trying to achieve in the first instance. Okay. If you're

[00:13:10] trying to set an alarm and it doesn't work fair enough, that's like technical and that's improving and has been improving massively already year on year. Right. Like, uh, how they will be part of

[00:13:21] our lives. Cause I think like the real interesting anthropological aspect here is like, you know, we know of like Jarvis or similar, like, you know, sci-fi assistants that are like basically super intelligence available to you, you know, at the tip of the tongue. Right. And you say something

[00:13:38] and it gets done and you know, are there other films about her X-Machina everything, right? It's like, okay. Like, are we going to have that next year? Probably not next year. Are we going to

[00:13:51] have something very similar to that in five years? Yes. So, you know, we probably can't exactly place in which ways it will exceed our expectations and in which ways it will still be a bit

[00:14:02] disappointing. But like, I think what we've already seen with, uh, chat GPT is that like, you know, it's, we, we've reached a level of reasoning that even those who worked on it weren't really, there were a few that were confident that we're going to get here,

[00:14:18] but for the most part, even those that worked on it, didn't think it would have got this fast. And of course, whenever we talk about AI, it's almost the law now that we must talk about the

[00:14:27] ethical side of that. So as a, someone that is a leading voice in AI, are there any ethical considerations that you believe that anyone working with AI should be considering, especially when you're at the forefront of AI development and deployment? It must be a question you get

[00:14:44] a lot at the moment, right? Yeah. I think like there's many ways to feel that one, right? Yeah. Yeah. Like broadly, you know, all industry leaders working on this should

[00:14:58] think a lot about like what they're doing to the world, right? Are we doing good thing or a bad thing? And I think, you know, the, the, I think the most frequent debate is around the economic

[00:15:07] impact, right? I mean, there are really two that catch a lot of public eye right now. One is, um, around the economic impact. The second one is existential, like, well, will AI, you know,

[00:15:19] rise and turn us all into batteries or something worse than that. And, um, I am firmly in the camp of no, that will not happen because when you really think about what we are in this gets

[00:15:30] like philosophical, right? Are you the same person when you wake up that you were when you fell asleep like that interruption of consciousness, like what is consciousness? What is it an individual,

[00:15:40] right? And like, in many ways, right. We already are Androids, right? Like we know how we feel when there's no internet and like someone leaves you in the wilderness with no signal for two days.

[00:15:53] Look, you probably feel quite alone. You feel like you've been like, you know, detached from a certain body part, right? Um, and really we are already as humanity, an interconnected hive mind. And I think with technology, we're going to continue to evolve, right? We have evolved very,

[00:16:11] very rapidly. A human with a smartphone is not the same species as a human without a smartphone, right? Our capabilities are vastly different. Like if I had access to internet and you transported me to Roman times, I'd be the equivalent of a God. I would know everything, right?

[00:16:28] Like, and that's just like, you know, it's interesting, right? Cause it changes the perspective of like, have we already evolved? We have, right. And then, you know, when you think about the ethical considerations, well, I think we're about to undergo at first it's going slowly.

[00:16:44] I'm actually surprised with how slowly, how little of an impact has had over the last year to three on human life as a whole. Right. But I think that it will go much more quickly than

[00:16:56] people think. And we will, you know, fundamentally find that, you know, maybe, you know, our kids will go to school and use technology to go about their work completely differently. Entire job categories that we needed millions for today will not exist. Right. And then

[00:17:16] the technology doing that will spawn off a different field of human activity where we will find like our use, right? Cause you know, the number of neurons you have in your brain is still

[00:17:27] higher than the number of parameters in GPT-4. Right now we're not very good at using them, but the hardware we're wielding is pretty impressive. So we can tap further into that

[00:17:36] and maybe technology can help us do it. Then, you know, we unlock the next phase of human development. Oh, when it comes to AI, there's a lot of business leaders sat on the sidelines at the moment. They

[00:17:46] want that AI story. They want to get involved, but a little bit nervous of how to do it, where to start and all those things. And this feels like a good opportunity to introduce everyone listening to

[00:17:55] Poly AI. Are you on this mission to empower enterprises to be the best version of themselves for their customers? And for anyone listening or hearing about Poly AI for the very first time,

[00:18:06] can you just tell them the kind of people that you work with and the kind of problems that you solve and business value that you deliver through this technology? Absolutely. So we build voice assistants that sound perfectly human and engage with customers when they call you on the

[00:18:20] phone or when you call them on the phone, right? They help power that interaction. And you know, we work with the likes of FedEx, Marriott, Unicredit, Caesars, PG&E. So verticals ranging from like delivery, retail, hospitality, logistics, utilities, financial services.

[00:18:42] And we help them pick up every call instantly. And then we triage, right? If our AI can talk to you the way that a human would, it does that. And we help every company recreate exactly the kind of

[00:18:53] voice that they want to use to embody their perfect service representative. Some of our clients even have several different voices for their AI. And if you want a Geordie accent, if you want a Brahmi accent, we can do that, right? If you want a Serbian accent in English,

[00:19:09] we could even do that. No one's ever asked, right? But if you want a Texan one, if you want a Californian one, those are all things that we do, right? And you can decide exactly the tone, the pitch, the vernacular, the vocabulary, the speed,

[00:19:26] everything, the style, level of formality. And when you do that, then you get people to really engage with these systems and you never leave anyone on hold, right? And then the one thing

[00:19:38] we do really, really well is we can tell if the user is not having a good time interacting with maybe because they don't want to, or because we're not able to provide the level of interaction that

[00:19:50] we should. If we detect that's the case, we hand off to the human and fast track them so that either you've spoken with an AI that really helps you, maybe even better than a human,

[00:20:00] because something it can do better than a human, or we'll hand you off to a human really, really quickly. So however you turn it, the company we work with has provided a much better

[00:20:12] customer experience. And that means that service levels, like how fast you pick up calls really improve. It means that you never miss a revenue generating call, like an appointment or customer intake. And it's also a lot more cost-effective than trying to find enough people to do those

[00:20:32] jobs. Because what's actually happening in contact centers right now, and that removes the ethical consideration for us, is many of our clients have employee churn north of 200% in the contact center. These aren't particularly well-paid jobs. And most of the entry level

[00:20:51] roles, people join, see that they don't like it and then leave. So to us, we're not automating the entirety of a contact center, but we are turning those managers into managers of AI agents rather than necessarily managers of people who come in and out for two months.

[00:21:09] So that then really helps with the P&L of that company as well. I love the idea there where if it's just a simple question, it gets asked 200, 300 times a day that the AI can just handle that and seamlessly give me that information where equally,

[00:21:24] if I'm unhappy, it will pick up sentiments, I would imagine, and then direct me to the human person straight away. But is AI involved with that as well, knowing that I'm a little unhappy at that time? Absolutely. I mean, AI is used for absolutely every single thing related

[00:21:41] to the understanding, to the reasoning that the system does, and to the response generation when the system speaks back. And what kind of feedback have you had both from your customers and

[00:21:52] the customers of the people that you're serving there? What kind of feedback do you get from this? Yeah. I mean, look, in terms of the enterprises that we work with, this is a lightsaber. There are

[00:22:02] a number of companies who told me that they would not have survived, say, high season around Christmas with the number of agents they had. They tried to hire extra people, but so do all

[00:22:10] their competitors and they're unable to find that stuff. And then they miss a lot of calls, their NPS scores lower, and that has a negative impact on their business. So they love it.

[00:22:25] In terms of the consumers, we have a lot of people who sometimes don't even realize they're not speaking to a human. Then when they realize, they go, oh my God, I can't believe this is not

[00:22:33] a human. And then they laugh, they ask questions, they try to ask system out on dates and stuff like that. It's really funny. It's exactly what you would expect from humans interacting with an AI superior to anything that they've ever seen up until that point.

[00:22:50] Absolutely brilliant. And if we dare to look forward into the future and look into a virtual crystal ball, are there any emerging technologies or trends in AI that excite you at the moment? Anything that stands out to you or something that makes you jump out of bed in

[00:23:03] the morning? Absolutely. I mean, look, I think that what OpenAI did with large language models is astounding. I think that the vision and the beliefs that creating these bigger models will lead to a disproportionately higher intelligence of these systems was a really spectacular bet that

[00:23:23] paid off. And I think with those things, for us, we operate at a slightly higher level than those foundational models. We operate at a conversational layer, which means that we really just have a lot

[00:23:33] of IP around how we build these conversational systems. But with those LLMs, we've been able to push, especially the reasoning ability of our assistants much higher. So then now we can create much more human-like things that can do our more complicated tasks. And I've not been this excited

[00:23:52] about my work for a long time. And for anyone listening that's inspired by your story and looking up to you right now, is there any advice that you'd give to a young entrepreneur aspiring

[00:24:02] to make a mark in the field of AI and technology themselves? That's a good one. I don't know. There's a lot to learn. I think most things, most advice I got, I didn't really understand

[00:24:15] at the time. So yeah, I think like, I don't think there's much that's AI specific about entrepreneurship. I think it's like, if you're going to build a tech startup, it's pretty well

[00:24:26] understood and it's bloody damn hard. So, you know, like you start, I think like maybe one good word of advice and something that served me really well is pick the right co-founders, people that

[00:24:36] you really love spending time with. Because if things go well, you're going to spend a long time building that company. So you're kind of like, you're in bed with these people for a long

[00:24:45] time. Make sure that these are people you want to work with and that you really deeply respect. And if you do, then you'll have fun along the way. And that's like, that makes it worthwhile.

[00:24:54] Fantastic advice. And as we come full circle now, if you look back throughout your career, of course, none of us are able to achieve any degree of success without a little help along the way. And there's usually a few people we encounter on that journey. Maybe they see

[00:25:09] something in you, they give you a little bit of time, a little bit of priceless information and support. If I was to ask this to you, is there a particular person that you're grateful towards

[00:25:19] who played that part in your journey and helped you get where you are today that we could give a little thank you and shout out to them? Absolutely. I think that would for sure. I mean,

[00:25:27] two people come to mind. One is my mother who really kind of like, through turbulent times and Serbian history and a tough time to raise a kid, really gave me her all to help me get to where I

[00:25:42] am. And the second one would be my PhD supervisor, Steve Young, who convinced me to not join an investment bank and instead go and work on voice assistance with him and a few other people. Awesome. And for anybody listening, just wanting to find out more information

[00:26:03] about Poly.ai, how it might work for them in their world or their business, where would you like to point everyone listening? There's so many different domain names out there. Well, poly.ai of course is our website. And if you want to reach out to me, it's Nikola

[00:26:19] at poly.ai, Nikola with a K. So yeah, would love to hear from people who was saying that our technology can help them be a better business. Well, you really have had a remarkable journey from selling books on the streets of Serbia to

[00:26:34] becoming this leading voice in AI and technology entrepreneurship. I'm sure there is so much more to come in your story. So I'd love to stay in touch with you, maybe get you back on next year, see how this world is continuing to evolve. But more than anything, just

[00:26:48] thank you for sharing your story with me today, Nikola. Oh, thanks for having me. So a huge thank you to Nikola there for sharing his incredible journey, his insights into the world of conversational AI. And it's clear that poly.ai is making a significant impact in the

[00:27:03] field, not only by enhancing customer experiences, but also by revolutionizing contact center operations. And as we look ahead, the rapid advancements in AI capabilities, particularly in understanding human emotions and context, that is a bit that holds immense promise for me.

[00:27:21] And it all washed down, of course, with that incredibly inspiring story that took him from selling books in Serbia to working on Siri to the great work that he's doing now. But here's a question for each one of you listening out there. How do you

[00:27:37] envision the future of AI in enhancing customer interactions and support? Do you believe that AI, as today's guest mentioned, won't turn against humanity, but complement our interconnected relationship with technology? I'd love to hear your thoughts and insights on this. Please join the conversation by

[00:27:55] sharing your comments, questions, experiences, requests to come on the show simply by emailing me at techblogwriteratwork.com, Twitter, LinkedIn, Instagram, and Neil C. Hughes. Nice and easy to get a hold of. Other than that, I'll be back bright and early

[00:28:12] tomorrow morning. Hopefully you will join me again there. But just a huge thank you for listening today. And until next time, don't be a stranger.