CEO Runway: The Latino Who Beat OpenAI and Google in AI

In terms of quality, we are one of the strongest research teams in the area of video and images, and we have many people who don't have PhDs or master's degrees and are still perhaps the best researchers.

You can imagine a very near future where I can create a podcast in real time, I can have an AI version of you, an AI version of me.

One is being able to research and create new frontiers, build products quite quickly and usefully in very short times. Now, the important thing is that you have to be able to do that every day, every month, every year.

Yes, good question. I don't know, I'd have to ask my mom to see what she thinks. Look, I think a large part of my career and my professional ambitions have been guided more than anything by curiosity. I like this idea of understanding different disciplines or areas of knowledge and approaching them in a quite direct way without necessarily having to ask. And that's basically, for me, curiosity is like interest in something without filtering yourself. I think the barriers for me are quite, or the worlds, the logics or the structures are quite arbitrary and a bit unreal. The only laws are the laws of physics. The rest is invented by humans, and the moment you start thinking a bit more like that, you start like, well, why can't I do this? Why can't I do the other? And you realize that you can, if you dedicate time and are curious and interested. A large part of what I started doing at the beginning and before starting Runway was driven by that. I was working, as you said, as a professor at a university in Chile, doing art projects, I had a mini consultancy in video and marketing, and I programmed on the side. And maybe it was hard to classify, maybe it was a bit hard to know what it was, but that was the point. The idea is to be a bit hard to classify, because I think when you take that perspective, you are able to discover things much more interesting than simply the things that already exist.

Yes, it was basically, I imagine everyone has used AI models to create images or video or content. One of the first AI models was a bit discovered, created in 2014, 15, and it was called Deep Dream. And it's not that it was capable of generating psychedelic images, it's that simply that was what it was capable of doing. They were quite surreal images, hard to explain, and for me it's part of the process of machines learning to understand a bit more how to create better consistency. But at the beginning, they were images, of course, quite abstract, quite surreal. And as you say, I had the opportunity to come across those papers and models back then, and I pushed myself, the same thing we were discussing before. It interested me too much creatively. I was very...

Simply the idea that first, aesthetically in terms of style and composition, it was something I hadn't seen before. And I was interested in knowing why this existed, what was making it, and what was allowing it to exist. And second was this idea that you can train algorithms and systems that allow you to create images. For me, it was a sort of, there's something there, there's something sufficiently interesting in that concept of having algorithms that generate images that I need to know more. It's a kind of hunger, a bit of intellectual curiosity to know why that was possible, and there I got into a kind of hole that I still haven't come out of, where I've been researching and learning a bit more about why. And I think I'm a bit closer to learning why, but I still have a long way to go.

Yes, well, in Chile I studied economics and I learned, I had several classes in calculus, statistics, I was taking econometrics classes, which is like the most statistical branch of economics. I realized that many of the systems for training automatic systems or machine learning were basically very sophisticated versions of linear regressions or prediction models. And the truth is, as I was saying before, I don't really believe in the label. I think that today, being someone technical or less technical is basically a quite arbitrary definition. You can program, you can learn to do basically whatever you want. You have the tools today, your exposure, and for example at Runway, we have a first-class research team, we are one of the strongest research teams in the area of video and images, and we have many people who don't have PhDs or master's degrees and are still perhaps the best researchers, simply because they approach the world and problems from a quite different perspective than from a purely academic point of view. And for me, it has to do with being able to solve a problem and the intuition and a bit of the determination to learn how to do something. When I started researching and programming and getting a bit more serious about AI, I took it seriously, I worked very hard, I learned to program, I learned to train my first models, and in general, anyone who sets their mind to it can do it. That has been perhaps one of the things I learned in university, which was basically if you learn to learn, you can go very far.

Yes, I always liked art. I had the opportunity to exhibit in a couple of galleries in Chile and work with artists in Chile. I love cinema. For a long time I thought about being a curator or also, I made films, I made shorts, I won a couple of mini-festivals, and it was basically again interest. I didn't have access to a camera, I got cameras from friends, and I was very interested in this idea of being able to tell a story through a video, an image, and sound. And I think that the most interesting projects in terms of art and science come a bit in the union of those two worlds when you are able to understand how engineering and science work, and the scientific method and logic and analytics are able to work hand in hand, speak the same language as aesthetics, art, a more open sensibility of the world. For me, many of the best artists have also been scientists, have been able to research, experiment, see what works, open new possibilities, show the world what is possible, which is basically the role of a scientist. And the same, there are many scientists who I consider artists because they are able to tell a story or help see the world from a different perspective. Again, for me, perhaps articulating this many years ago wasn't possible because I simply was interested in both things. I was interested in learning about statistics, calculus, and programming, and I was also interested in seeing films at the cinema, art in downtown Santiago. And finding a label for that seemed less relevant than simply enjoying it, immersing myself in it.

Look, I came to New York with very low expectations, to be honest. I think I had lived my whole life in Chile, in Santiago. I had a very unclear idea about what was going to happen next. My idea was to come for 2 years, study at the university, learn. I took it very seriously. Basically, I said, I have 2 years to study everything and learn as much as I can. So I worked, I don't know, 18, 20 hours a day non-stop for 2 years. It was quite intense, but I absorbed everything, I took classes in everything, basically. And my idea was to return to Chile. I basically had a plan to return to Chile because I never thought I could stay, but I also didn't have many expectations, like let's see what happens. And I think that ability to, maybe it's a bit obvious, but not being able to plan very long-term allows you to open up to new opportunities that are quite difficult to predict. What you realize gradually is that if you have a quite concrete objective and you know how to plan very well at 6, 12, 24 months, you're going to close yourself off to anything you discover that was unpredictable. I started to encounter ideas, concepts, and things that seemed quite interesting, and the idea of not having a plan was perfect because it allowed me to discover and learn things that I simply didn't know I didn't know. And gradually, I started questioning what I wanted to do.

No, it wasn't even a master's in artificial intelligence. It was a master's, a quite multidisciplinary program from the NYU Tisch School of the Arts that combines science and technology, a quite independent program. Basically, it's people who are on average between 25 and 35 years old and who have worked or done something professionally, and they are either engineers or artists, and if they are engineers they want to study art or learn more about art, and if they are artists they want to be engineers or learn more about engineering. And the program is basically like you discover or investigate or figure out what you want to do. And my obsession, I was reading my application from a bit of time ago, my university application, was basically I want to learn about, again, this is 2016, I want to learn about artificial intelligence, I want to learn about how this artificial intelligence or this idea of deep learning models, as it was called back then, can be applied to the world of art and technology and creativity. And that was my obsession. There wasn't a specific program about this. Again, 2016 is like let's talk about making cinema in 1850, it was quite early. There wasn't an infrastructure, there wasn't a community, there wasn't necessarily the knowledge and the advances we've seen today. And I built my own program, I built basically my own path, and the consequence or result of that path was Runway.

I think I gradually realized that this idea I had, this obsession and curiosity, was quite more interesting or quite broader than a personal interest. I realized that many people approached me, asked for access to what I was building. I remember I posted, I didn't have social media, but I don't know why I made a Twitter account, and I wrote, I posted what I was doing, which again, you can search for it, it's on Twitter, quite simple. I'm doing this idea of Runway, I want to build these models, and it went viral. An engineer from NASA wrote to me, artists in South America wrote to me, designers in California wrote to me. It was a blog post, a repository, right? Simply a 2-minute demo that I posted on Twitter. I had, I don't know, 10 followers, nobody back then. And somehow, I don't know, I think it touched some interesting chords in some people, and it started expanding and went semi-viral in a very specific community. And there I gradually realized that maybe it wasn't simply my own interest, but there was something deeper and more interesting. And so I kept turning it over and over, and I started talking about it with my classmates at the time, with Alejandro and Anastasis, and there we started thinking about this idea of what happens if we take this more seriously. And the three of us had very similar interests. So it was a bit like a quite obvious decision once we said, 'Hey, why don't we start a company and do this in real time?'

I think the insight in general was that the way we create content and media and stories is fundamentally going to change. And we are still perhaps in that story, a bit like the analogy of the history of cinema. We are in 1900 still, and I think the impact in terms of art, for us we have always thought of what we are doing as the beginning of a new form and a new format of communication and basically a new medium of art. And we want to be the company or the people capable of thinking about that differently or contributing to that change in a unique way, or training these models, as you say, differently. Today we are being used by companies, studios in Hollywood, by designers, by brands, by marketing teams, by creators who have always looked at musicians. And I think what we do is put a bit of humanity and creativity and art to hard science, and I think that is reflected in our product.

Fundamentally different. Think of it more than anything as a kind of, the difference between a horse and a car, between a horse and a bicycle, between a camera and a paintbrush, between a camera and an hour of theater. They are basically different mediums. Telling a story with artificial intelligence is a different medium, it's a different way of thinking creatively. It requires tools and people and a form and a process quite different from the one we've worked with in recent years. The most important thing about artificial intelligence, and this is fundamentally quite important to understand, is that a large part of what we are living is not a linear transformation, it's not simply a better way of what we were doing before, it's a totally different way of doing things. And when you realize that, you realize from the potential and the capacity of what it can become. And that has always been, again, our way of seeing the world. So we don't think very much about, well, Runway is a better way to do, I don't know, editing tools or image creation tools. I think that's a quite linear way of approaching the problem. And if you realize and project the capacity and progress a bit, you realize that these models are simply totally different creative mediums, they are totally different ways of understanding the world. Being able to work with an intelligence, with an autonomous system, intelligent, that is quite more intelligent than all of us, simply requires another way of understanding the world, and that's a bit of what we want to do.

One of the most important things when you do a creative project, for example in films, is planning, right? If you think about a film, a large part of the budget or time is dedicated to the specific planning of each of the scenes, of each of the projects, of each of the shots, of the dialogues, and the way to plan to do that is basically so that when you get to the set and you have to shoot, you don't have any improvisation because improvising is quite expensive. You can't be on a set and say, 'Hey, let's see what to do now' because you have to pay the whole team, right? Exactly. You realize that the barriers of planning, of coordination are basically cost barriers. If I told you, 'Hey, forget about the costs, forget about nothing,' simply you can get to the set and do whatever you want and you have the set 24/7 and nobody will question you and you can go to sleep and come back and the people will be there. Probably your project would be quite different. You'd say, 'Okay, let's try two different scenes. Let's talk first. Let's put the camera here. Let's do all these crazy things and let's see what we like about what we're doing.' Well, that's exactly what happens today. Today you can have a creative studio, a production, you can have an entire creative team at your disposal 24/7, which allows you to ideate and create at the same time. The separation between ideation, planning, and waiting all that time to finally create that is compressed a bit, and we are living in a world where I can think of something and execute it immediately. Today I can think of a software and I can create it literally in less than a minute. Today I can think of a scene and I can create it literally in less than a minute. If I have that speed of creation and that cost of creation so fast and so cheap, I can create 10,000 versions of that and I can choose, right? The one that seems most interesting to me. And from a creative point of view, that's incredible because I can explore ideas I never thought of exploring. Today, a large part of the best ideas, of the best creatives, of the best directors, of the best art directors or photographers of any discipline, are limited because there's no budget, they're limited because there's no time. If I remove the time and the budget, I think the world becomes much more creative. And that's a bit of what we want to achieve.