Welcome to AI Snacks with Romy and Robbie, an AI literacy podcast for everyone interested in AI and robotics. I'm your host, Anastasia, founder and CEO of AI Entertainment. This company brings knowledge and understanding of AI into families and businesses worldwide.
Today I interview Rob Locascio. Rob is best known as the founder and longtime CEO of LivePerson, the company that pioneered web chat and conversational interfaces for customer service and commerce. But today he's focused on developing two new businesses. One is Kid or Keep It Dreaming, an AI device and content platform for children aiming to replace addictive feeds on social media with safe, creativity-centric experiences. Rob is also developing a company called You Are AI, and his technology lets adults build an AI self. His clients might be businesses willing to replicate and enhance their best performers in sales, marketing, technology, or his clients might be families willing to carry digital selves of their loved ones forward. Rob is all about protecting human data, enhancing human creativity, and protecting children from mindless scrolling on social media.
Today we will discuss confusion around AIs and large language models, the future of AI and data technologies, and how we can explain AIs to someone who has never studied computer science or mathematics. Now let me open the studio for Rob. Hi Rob, welcome to my studio.
Hi Anastasia, how are you? Thanks for having me.
Could you please briefly talk about yourself and your journey and maybe emphasize two to three pivotal moments that led you to where you are today?
Yeah, so my name is Rob Locascio. I started a company I guess I'm well known for inventing when you chat online for customer care, which I invented in 1997, 1998. It was a company called LivePerson. I ended up taking that company public in 2000 during the dotcom and then built it to being the leading conversational AI company. We got heavy into AI and we basically had about 3,000 employees, 500 million in revenues. It was a big company and I built it from sleeping on a couch back in 1995 all the way until three years ago when I left and decided to pursue actually building two companies: one for children in AI and one for adults. So yes, I wanted to cover the spectrum.
I guess one of the pivotal moments in my life was sleeping on a couch for two years as a startup in New York City. I didn't have an apartment. I just had an office and showered at a health club, and I still have that couch. It taught me a lot about life. I had a company out of college that failed and that put me on the couch. I always felt that what I learned there is you just can't quit, you got to keep going and figure out the problem, how to solve it. It also matured my perspective on how you view the world. I worked with a guy called Dr. Frank Morio at the time, a psychologist, because I was really pretty depressed. He kind of reworked my brain and how I was perceiving the world. I realized what got me on the couch was my perception was off, and what got me off the couch was I reworked how I see the world. That's what's interesting, like the nature of reality. What is reality? Does it serve you or not serve you? So that was a big pivotal part.
The second one was during the dotcom. I survived the dotcom, which there were very few of us that made it through. We had to unfortunately get rid of about 80% of the company. We got rid of the people, which was very hard. It's the first time I laid anyone off, but I learned that there's never a small enough core of people. We ended up down from 200 people to 40, and those 40 people kept the company alive and grew it.
The next pivotal one was really after the dotcom and running this large company, this public company. I started to think differently. Although we were, it was funny, in February of 2022 Fast Company named us the number one AI company and OpenAI was number two because we had a half a billion in revenues and they had nothing at the time. But we were really in the mix, but I just felt like there was a path that we were all taking with large language models and I felt like there's another path I would like to put us on. I felt that I could do that if I kind of start over and build some new companies, and that's what I did. I left and moved to California with my family. I have three young kids, nine, seven, and four. I moved to California and then started these two companies: one called Kid Keep It Dreaming for children, a device for kids, and the other one is called You Are AI, and we're replicating human beings so that they can be their own AIs and they can be as powerful as ChatGPT. So these are the two things that I'm working on today.
That is wonderful. How do you define a child and how do you define an adult?
Well, the thing is I think everyone has a child's mind, so we have this creative power. Obviously children are defined by age, but the power for AI for children is really that I wanted to put something in their hands that would allow them to tap into that great creative force that's uncorrupted and allow them, even a four-year-old, to speak something out of their mind and create a picture or create a story or talk to a character or create a character and take the power of creativity and bring that to life. I think the new Walt Disney, if I could snap my fingers and get there, is all about the children and parents creating very high quality content using the tools of AI and then that gets distributed amongst them and they reflect on themselves. They get to see themselves in these wonderful creations versus some animator who's in a studio system. So that's how I see children and adults. It's kind of the same thing. I see AI as a reflector. You can be your best self with it and not as a job replacer. There's this narrative that we're not as smart as an AI, that AI will be smarter than us, which is a false narrative. So for both pieces, I see how do we use this technology to help us, not hurt us.
I just wonder because you have led through multiple tech waves. You started really very early before the dotcom boom, and now we are in the age of large language models, generative AI. So we might share or not share the vision that that might be actually the false wave to ride if we are to develop truly intelligent AI and general AI instead of what we are having right now. But in your eyes, what is the biggest confusion about AIs today and how can we overcome this confusion?
Yeah, I think there's the path we have been told is the path of AI is that there's a large language model that has an intelligence now that's a PhD of some magnitude, and I don't have a PhD, so therefore... and we're starting to, it's like idolatry. We're starting to idolize the machine and trying to understand how it's better than us or above us. And the reality is without our knowledge, without our human knowledge being fed into it, it's an idiot. It's not even a newborn, it doesn't have the intelligence of a newborn because at least a newborn has senses and it starts to pick up and starts to evolve. They're fooling us with the fact that these things are an AGI. It's really funny. They're fooling us with the fact that these things are getting intelligent. The only reason they're getting more intelligent is because we keep using them every day and we keep putting more of our data into it.
When I launched URI, which originally was called Eternos, originally I started a year ago to replicate people, their personalities, their voice, their knowledge, their personal stories, so that when you died, you had this really high quality replica of yourself. And then it evolved, the business into doctors and lawyers and writers showed up and they're like, 'Geez, I want to create an AI.' But it all came down to one thing: they wanted to own their IP. They wanted to own what made them special. And they started to realize I keep putting what makes me special into ChatGPT or whatever and it's getting my data and it's getting smarter and I want that to stop. So I think they've given us this narrative that these are self-learning machines and they don't learn without our data sets. So it's kind of a chicken and egg problem.
It's interesting what you are saying and I have multiple questions because obviously LLMs are large, so at some point in time there were complaints that those LLM companies run out of data on the web and now obviously we need to provide tons of synthetic data. So what do those synthetic data do to the overall context? Do we need small data? How can we work with small data if the algorithms crave a huge amount? On another side, whenever I teach AI to MBAs, people who have never studied statistics or computer science, the biggest surprise for them is that those LLM-based AIs don't truly answer us. They're frozen post-training. They get delivered, they get shipped, and we prompt, but prompt is not a conversation. It's just shifting vectors in a large semantical space. So how on earth does this thing generate text but it has no clue that we are in a human language experience, that we asked something and we expect a human reaction? So how do you actually teach your customers about what this thing is and what it does when they come to you and say, 'Look, we need AI associates to help us with our IP, with our legal discovery,' but how do you teach them to understand that this thing is not alive, that this thing is not the same as a human associate, a human attorney?
Well, I just kind of look at large language models as aggregated knowledge, and that's where they're really great. They can aggregate scientific knowledge from thousands of scientists and aggregate thoughts around that, ideas. And what I think the basics of how I see AI is small models where I can replicate my knowledge base. I can then get leverage on my time by sharing that with others and sometimes monetizing that interaction with others. And that's kind of where I see the power of AI today is that there'll be hundreds of millions of small models that represent us. Maybe those models can come together as a round table and they can be organized together to give information. But it's better that we retain the ownership of our data sets and our AIs because it answers the question of the existential threat to humanity, which is your AI is not going to try to take your job, your AI is not going to kill you. And so I feel like it also puts the power back into the people that matter, which are the people who generate the data sets.
And what we found is that most humans have a whole other set of data, which now the large language models are trying to get their hands on, which I call zero-party data. It's data that's up here. They're mostly story-based things in your life. You haven't recorded them, put them on the internet. You haven't edited them and put them into a book. We even work with writers and we find they may have written a book that has 200 pages, but they may have 20,000 pages of research notes that gets them to the 200 pages or 2,000 pages. Those 20,000 or 2,000 pages of research notes are actually really great to build an AI from, not just the synthesized edited pieces which are on the internet and public. So I think it's a whole other world of data that humans should retain. They should have the ability to monetize, especially creative people like writers are being treated like second-class citizens today. If I write a book or write something, the models now can assume those into their models for free. I know there's some negotiation going on, but there was a legal case a couple months ago where the court said it's like training a child: if they bought one book and give it to the large language model, it's like training a child and then the child comes up with something else.
But how do you argue the case that those models just tokenize the input? So it's like you have two hands full of sand on the beach and those little grains of sand, those are tokens, but it's not the full beach. And then you kind of reassemble those grains of sand into something. It might be a castle, it might be whatever house. So how do you argue this case? I mean this is really difficult. Everyone uses language. What is a language? It's a huge question. Is it something native? So Noam Chomsky would argue it is innate in a human being and a human child.
You unlock it. Yeah. You unlock it. It's not learning it, it's unlocking it. Yes.
Absolutely. But AIs don't possess so-called world models. So AIs are unaware what is gravity, what is this physical world, what is a language. They won't be capable to distinguish between human and non-human language. So how do you kind of embed this world model into an AI and what is the path forward for you?
Well, once again I go back to, yes, the models are doing a good job at tokenizing and then guessing, obviously it's a pattern of what's the next best word to express an idea. So they obviously do that quite well, but the underlying, I go back to, I guess I'm very focused on the data sets and the chicken and egg is the model and the data set and which one's more important, right? Everyone's talking about the models. It's very focused on a better model, a model that's going to have AGI. And I look at it, yes. So I look at the other side that the grains of sand are to me what make the beach and those grains of sand are the knowledge sets from each of us as human beings. So I'm not sure how good they are. Does that make sense? I say like no one really knows how good the models are because if they lack the data set, they can't perform on their own. They don't just generate thoughts on their own. They don't just generate grains of sand on their own. So I guess the way I look at it is that I think the power is in the data sets and there's a bunch of data sets out there right now that have not been recorded that I think could benefit us much greater than what is in the models today.
No, but it's actually a very interesting conversation because when I look into the waves of technologies prior to AI, obviously AI is not new, it's from the 50s, even from the 40s, 60s. So what is really new? Many foundational algorithms were formed before let's say big data and those powerful GPUs powering the contemporary world of AI. But when I look at it, where's the value? So the way it used to be at the beginning of a computer age and then wireless age. So semiconductors, the infrastructure, those device makers, and then the value moved into applications. We can't really say right now that the value is completely at the application level. So we probably are in a very initial stage of what is going on. This is the path to my understanding. I don't believe that investing into what I call flawed architectures, which will always hallucinate, is the path forward. We might need to go to the drawing board and kind of reform what we are doing. But I'm very curious. So you have those two companies and one is I would say an avatar type of an application and another one is a guarded environment for children because to my knowledge Keep It Dreaming, you have a device, a nice kind of, what is it? Is it like a bowl? No, not quite a bowl.
Yeah. And we're seeing now the data because 400 families went live here and now we're seeing what kids are using. It's very interesting. 10X is characters. They're talking to characters.
That's wonderful. But this is a guarded environment.
It's all, these are kids' models. They don't have hallucinations into weird stuff. You can't have it harm you or make weird things or pictures or talk to the AI buddies in a way that would harm you or you say something to it. Because we created it only for kids, we're able to control the models. We don't have an adult model then wrapped with trying to make a kid model out of it and doing our best. We had to be very careful to make it so that AI could be brought without harm into a child's house, to a home.
Talk to me a little bit about the technology powering those very two different approaches. So the avatar type of an approach and then this walled garden type of an approach for a child. What are the differences? Where are the biggest challenges? What do you envision let's say two to three years from now when you have more users, the technology gets deployed in a variety of geographies. So how do you think about the technology stack?
Yeah, I mean I guess I'll talk about the similarities for just a second. I remember my first company was called LivePerson and it was all about humans interacting on the internet when there were no humans on the internet. I invented the chat because a website was like a vending machine. So somehow personally I have been put here to bring the human element with technology if I go broad. The next level of that is what I see is that for Kid, I would really love to have a studio, an entertainment company that is made from the children's creations that are on the platform. So children are creating characters right now. They're creating stories, AI stories. They're creating beautiful pictures. They can't do video. We don't give them video yet because it's expensive. But imagine a four-year-old creating a five-minute video of some beautiful thing that's in their head. That to me, we've been moving towards UGC, user-generated content, obviously with everything that's going on social media, but the next step to it is using AI to be truly reflective on my power, the powerful ways I can create things in my mind and bring them to light.
The other thing is the humanity of it. These things we're seeing even today is like kids are responding to the characters and the humanity, that it's buddies and their friends. They want memory. We have to add more memory on right now, we don't have long-term memory. So now we know kids want long-term memory on their AI buddies and stuff like that. On the other side with URI, it's about once again turning yourself into this machine that can reflect your thoughts. We do add video there so you create little thoughts and then they become videos. People can watch a video, flip the thought over and start chatting with it. But once again it's really about this human element. What I'm most interested in is how do we extract the humanity? How do we bring this humanity into this technology because what makes AI so fascinating for us is that and why it just blew up quickly is that it reflects us. It's talking to us. It's doing things with us that are very different than the enterprise software machine or application software machine platforms. So I think I'm interested in bringing more and more of that humanity forward with children and adults. I have three young kids and I just see the power that they have to just create things, but they don't know how to make videos, they don't know how to edit videos, they don't know how to shoot a video, but they can make videos, they can make beautiful videos.
Let's just pause here and discuss childhood development. I can share, I have a daughter, she's 16, she started with the devices. I was not restricting the device use for a very simple reason. My daughter was tremendously busy. At the age of three years old, she started playing piano, playing competitive piano. She's a pianist though. She will be a scientist. She wants to study medicine. She was doing Kumon math at the age of three. She was writing and she was doing puzzles. So she was just busy. So yes, she was using screens. And I remember quite vividly when for the first time I took her with me on a trip and we were in a high hotel in Berlin and we don't own a TV at home. So we just had iPads and small smartphones and then she is in front of a TV and all of a sudden she's touching the screen and pushing her fingers and I was like what is going on there? What is she doing? This is dirty. What is she doing? And then there was a flamingo on the TV screen and she really wanted to make it bigger. And it dawned on me that she has never experienced any other type of a device because she grew up with iPads but she has never spent too much time on devices. She was reading books and she was of course entertaining herself with the screens, with the smaller screens and then she was evolving physically because of all her sport exercise and the piano. When I look into parenting and all the complaints, so obviously I write stories for kids. How do you deprive children of screens? How do you kind of make them evolve without impairing their cognitive capabilities? And AI is a major villain here.
Is it maybe something else? So how do you think through those choices for parents, educators, maybe even policy makers? How should we integrate technology into lives and schools and early learning of our kids? Where should be the no-go? Where should be let's try and collect data and iterate. So what is your philosophy there?
Well, having three children, nine, seven, and four, I've got like a little research lab in my face all day.
Congratulations, by the way.
