Episode #1: Translating AI To Make a Better World by Helping Teachers (The Merlyn Mind Story)

Media Thumbnail
  • 0.5
  • 1
  • 1.25
  • 1.5
  • 1.75
  • 2
This is a podcast episode titled, Episode #1: Translating AI To Make a Better World by Helping Teachers (The Merlyn Mind Story). The summary for this episode is: <p>How can we translate the thrilling advances at the forefront of the science of artificial intelligence into solutions that can make the world better today? In this episode, we explore the past, present, and future of AI and how translating AI into solutions to help people be more productive as they do work that is uniquely human is the future. Satya shares childhood stories about his obsession with science fiction and technology and how that led to his passion and career as a technologist. We get a front-row view to the exciting times at IBM as Watson beat Jeopardy! and the world saw the potential of AI. We also learn from the failures and difficult lessons learned in Satya's career that led to the foundational insights and inspiration for starting Merlyn Mind and building a digital assistant to help teachers. </p>
Merlyn Mind is an AI company.
00:38 MIN
It all started by secretly reading Asimov's "I, Robot" as a child
01:01 MIN
Technology has the power to change the world for the better
00:26 MIN
Head spinning progress in AI over the past decade
00:44 MIN
Watson winning Jeopardy! was a watershed moment in computing
01:17 MIN
Satya given the opportunity to lead the efforts to take Watson into education
00:11 MIN
Pursuing the grand AI challenge at IBM to get a computer to teach
01:23 MIN
AI and people are good at different things. Successful AI needs to be scoped carefully to focus on the right problems computers are great at solving.
00:47 MIN
Teachers are brilliant at the very human process of teaching and learning
00:57 MIN
Computers can't teach and motivate like human teachers, but technology can work with teachers to help them as they teach
00:43 MIN
Inspiration from "Neuromancer" and similar science fiction to create technology that is seamless for humans to interact with
02:34 MIN
Inspiration from amazing technologies like Siri, Alexa, and others and an opportunity to apply advancements in AI and digital assistants in education
01:06 MIN
We're focused on translating AI to help teachers and take the mundane tasks away from them with Merlyn so they can focus on the human work they do really well
01:31 MIN
We learned from failure. An unfortunate investor demo helped us see the need to give teachers multimodal control. Voice alone is note enough to assist teachers to interact with technology more seamlessly.
03:51 MIN
We listened to teachers and learned what they were struggling with and how AI could really help them. Teachers didn't want one more app. They asked us to help them more easily manage all the technology they are using in classrooms as they teach.
00:57 MIN
The best AI solutions are narrowly focused on specific industries and specific workflows.
00:27 MIN
Teachers deal with an incredible amount of technology as they teach today
00:33 MIN
If Merlyn can help teachers take some of the burden away of managing all the technology as they teach it will help teachers teach and students learn. This is what Merlyn does.
00:49 MIN
Education needs private and secure AI built for classrooms. We built Merlyn Mind to bring AI and digital assistants into the classroom to meet the privacy and security needs of schools.
01:42 MIN
Teachers, administrators, and students who have Merlyn in the classroom love it. And we're just getting started!
01:22 MIN
The immediate future for Merlyn is bringing AI and process automation together to continue making life easier for teachers and students as they teach and learn.
02:52 MIN
Thrilling advances at the forefront of AI science. But a gap in companies that are applying these advances into solutions to help people today.
00:58 MIN
We are translating the science and breakthroughs at the forefront of AI into solutions that can help people today.
01:14 MIN
The AI revolution is the most transformational revolution I've seen in my lifetime.
00:35 MIN
Technology makes me optimistic about the future of humanity and the world. We can use technology to solve hard problems.
01:47 MIN
Education is foundational to all of us. If we can use AI to improve education then we can have a deeply transformative impact on the entire world.
01:23 MIN

Levi: This is Supervised Learning, a podcast where the Merlyn Mind team learns from experts in artificial intelligence, technology and education. We hope you enjoy learning with us through these conversations with those who know. Time to learn. Welcome to the podcast, Satya, I'm very excited for us to have this conversation because the world needs to know what we're doing here at Merlyn Mind. So how about you start us off by telling us a little bit about who you are and what Merlyn Mind is and why it matters.

Satya: Thank you, Levi, glad to be here and glad to tell the world our story. So I am a technologist. I've been a technologist my entire career and Merlyn Mind is an AI technology company. We exist at the intersection of the science and AI, which is moving forward at a thrilling pace and the world of applications, right? So we believe firmly that we can take the science and apply to improve the lives of people everywhere. And that's what Merlyn Mind is. It's a company that exists at this intersection. We're looking at ourselves as translators and the first industry that we are going after in a very concerted way is the education industry.

Levi: Okay. And we're going to talk a lot more today about AI and education and what we're doing and why it matters, but let's rewind and talk about your background a little bit. What's your story and what motivates and inspires you?

Satya: Sure. Yeah. Well, I mean, ever since I was a child, I was interested in science and science fiction and there is this funny story which I tell my friends, which is, how did I get interested in science fiction in the first place. So I had a brother who was four years older than me. And at that time I was 11, 12, and one dwell-knownay he came home with a book by Asimov, I think it was I, Robot and other stories and of course, he was reading the book. So I was forbidden to read the book at that time. So he had to finish it and then maybe I can get the scraps afterwards. So, but I just remember looking at the cover and the back page and the back cover and being so entranced by it that I couldn't wait for him to finish the book. So I used to wake up in the middle of the night and sneak into his room, take the book and read it under the light of the naked bulb that was hanging outside our bathroom. And that's how it got into it. This future world that Asimov was writing about robots, right? Mankind spreading to the stars. It was so entrancing that I felt this is something that I want to actually grow up and be a part of. This actually leads to a very well-known phenomenon. So I'm by no means unique in this, right? There's something called the Star Trek effect, which is, young people watch shows like Star Trek or read books by people like Asimov and Clark back in the 70s and 80s or these days, I guess Neil Gaiman, et cetera. And they want to grow up and build that world. They want to grow up and live in that world and build that world. So really a lot of my motivation was inspired by great works of art, of fiction literature, et cetera. So that's how the whole thing started, how the interest in technology and science started for me.

Levi: How fun. So you are now living your childhood dream! You're using technology to try to create a better world. That's a fun path to be on.

Satya: Very much so. Yeah. I mean, there's lots of stories we can ... Lots of little rattles we can go down here but this ... As you grow up and as you learn more about the world and everything else, you basically realize that technology in fact has an enormous potential to change the world, right? And so at its core, at the core for people ... For all of us, I guess, right? This is enduring belief in the power of technology to change the world and to create a better world and to create a slightly more utopian situation for all of us. And that's kind of why Merlyn Mind exists. We saw an opportunity. We saw the evolution of AI. The last decade has been head spinning in the evolution of AI from the Hinton publishing about deep learning, deep belief nets, Watson learning Jeopardy! Deep learning as a field really coming of age, advancing language understanding, speech recognition, image recognition, and these days helping generate language. All these things have incredible potential, computers helping take decisions and so what we are really entranced by is the idea that, how do we take these fundamental advances and turn them into products that really make a difference to people?

Levi: Yeah. So let's dig into a few of those topics specifically. Before we do, maybe you can tell us a little bit about your journey through technology in your career. I mean, I guess you have come into technology at an interesting time because you started before, we had computers in our pockets that could do almost anything, but you were really part of the foundational technology efforts to make that possible with advancing Moore's law while you're at IBM. Just tell us about your trajectory and what you've seen in your career.

Satya: Sure. Yeah. No, it's a ... I mean, for me, I've had this abiding level of computing, right? So we talked a little bit about my childhood and the history of science fiction and shaping me, right? But I finished my PhD in the late 1980, in the late 1990s, I'm sorry. And I joined IBM Watson labs about a year after. And I was ... So I was at IBM Watson labs at a very interesting time. No, IBM Watson labs is one of the places, arguably one of the few places on the planet that had a significant role to play in the evolution of computing itself. We had ... So down the hall from where my office was, Bob Dennard had an office. Bob Dennard, for those who don't know, is the guy who invented the DRAM, who invented Dennard scaling, which is the underpinning by which Moore's law actually progressed. Moore's law was an observation but Dennard scaling made it to the reality. And so Bob Dennard was a legend in the industry. And so he's still there by the way. And so growing up around people like Bob Dennard, Mandelbrot who obviously discovered Fractal Geometry, Charlie Bennett, who advanced quantum information theory. It was a heavy place to be. And my early years, the first part of my career, I guess, my first 10, 11, 12 years was really spent in the company of a number of giants who were helping advance computing. And it was a thrilling time to be there working on very esoteric problems that were important to solve for the next generation of chips to be made. But along the way, I also have got ... I've been interested in AI for a very, very long time and IBM research and IBM Watson labs was one of the places where some of the early advances in AI also happen. Deep Blue beat Kasparov, the early demonstrations of speech recognition, computer vision, using an older technology at this point, feature engineering. They all happened there and then Watson on Jeopardy! Right? And so I happened to be ... I got very fortunate and that I happened to have ... I happened to be around staff around the time Watson on Jeopardy and there was a job I had, I think in 2010, where IBM was celebrating their Centennial and the head of IBM research came to ... OK, he put together a small team of four people. I was one of them and he basically said to us, "We want to write this very visionary story of the future of computing, right? So your job is to help write the story, go figure out how computing will evolve and how it can transform the world." And it was perfect for somebody like me. I spent all my life dreaming about it and I'm a dreamer and a builder and this was perfect. And so we realized one of the things we realized very early on is look the farther back we see the history of computing, how it evolved the farther forward we can project. But eventually trying to forecast, he wanted this 100-year vision of the future of computing, and trying to predict that was just going to be really, really difficult. And so we said ... We went back and said, "How about we talk to you about the next 25 years of computing?" And he said, "Deal." And this is ... At the time, David inaudible and his team were preparing for the Jeopardy! tournament with Watson, right?

Levi: Can you actually tell us a little bit more about ... I mean, for those listening, why was Watson beating Jeopardy! such a big deal? What happened then? What was different with computing? Why was that such a watershed moment that changed the way humans and computers interact?

Satya: Yeah. Good question. If you look at the history of AI, the field has advanced. I mean, there's obviously lots of background technology that was being built, but the field seemed to make leaps through these kinds of demonstrations, big demonstrations, Kasparov beating ... Being beaten by Deep Blue, which is the first demonstration of a chess-playing machine or computer beating a human. And-

Levi: Not just any human, right? The best human.

Satya: The deep impact on the greatest place in the history of-

Levi: Right.

Satya: The game. Right? And so Watson winning Jeopardy! was an interesting advance along the same vein in that Jeopardy! is a very complex game, right? It's not just a test of your factual knowledge of facts around the world but it's also an understanding of language. The language there is very complex. There are puns and there are illusions and it's ,... And you have to unpack what it is, what is the focus of the question and what the answer is, and all of that in a very short period of time. And it's not simple, the queries are quite complex. And so people who play it well, get really good at passing the language, unpacking the illusions and the puns, and trying to figure out what is it that they're trying to answer and chiming in all within the matter of two or three seconds to get to the right answer. So, Watson was a demonstration of the ability of computing to actually do this, to understand language, a complex language and to retrieve important information very rapidly. And it was a demonstration of a machine getting so good at it that it could be the two most winning champions in the history of Jeopardy!, right? It was a pretty watershed moment in computing. Of course, subsequently we have AlphaGo beating world's best Go player from DeepMind but I was there, I happened to be at IBM Research at a time when this happened, it was just completely mind blowing and very inspiring. And the subsequent interest that we saw from all corners of the world, people from different walks of life, whether it's healthcare professionals, educators, the legal profession, finance, all industries were interested in, what can I do with this ability? And so I happened to be there at IBM at that time. So that got me. So that was a transition from ... For me, when that happened, I said, actually, I've been doubling around in AI and a fairly obscured field called neuromorphic computing before that. And so the demonstration happened and I realized I need to actually be a part of this thing.

Levi: Okay.

Satya: And so ... And I jumped onto the bandwagon and I was given this incredible opportunity to take Watson, lead the chart with Watson in education. And that's kind of when my whole journey began in this field.

Levi: Okay. So let's talk a little bit about what you did with the Watson lab and all the research at IBM for education. What did you try to solve? What problems were you looking at? What types of things did you learn? What were the successes? What were the failures? Just tell me, what did you gather from that experience?

Satya: Yeah. I mean, that's a great question. In fact, I look at Merlyn Mind, this company and what we're doing here now as an end product of all the lessons we learned there.

Levi: Yeah.

Satya: So back in 2012, 2013, when I was given the chapter, it was this incredible opportunity and I was sitting at the IBM Watson labs ... history and so one of the things I went back and talked to the leadership about there was, I said, "Look, I don't feel that we should just be building something incremental here, we could be doing something very grand." And it made sense. You're an R&D lab, you have thousands of PhDs, you have decent budget and I thought it was an opportunity to go after one of the grand challenges in AI, along with, of course, doing a bunch of other very useful things with Watson and education. And so one of the grand challenges in AI, if you trace back to the history of the field in the 1950s, mid to late 1950s, the founders of the field used to say, why would ... They used to be asked all the time, they envision intelligent computing machines Alan Turing and then subsequently Simon, Newell, Marvin Minsky, all these people, they used to talk very passionately about intelligent machines, 1940s, 1950s post-war. And people just ask them, "Why would you need an intelligent machine?" And one of the use cases they would advance was, "We would like a computer to teach." Right?

Levi: Interesting.

Satya: And so when I was given the charter, I'm like," That's going to be a hard challenge, but it's worth going after it," because of course, you're in a research lab and one of the more storied places on the planet. And I realized if I had to do it right, I also had to understand a little bit about intelligence, how the brain works, neuroscience, et cetera. And about 2, 3, 4 years into it and so we said, "Okay, maybe we need to build intelligent tutoring systems," and 2, 3, 4 years into it, the magnitude of the challenge really dawned on us. So we went after it in a bit of a foolhardy way, I would say at this point and the magnitude of the challenge really dawned on us. We are like, okay, we need to start ... Take a step back and realize that there are ... The progress in AI is all about scoping the problem well and building it in a very domain specific, very scoped fashion and there are things that people do and things that people do really well and things that computers do really well. And getting a computer to do something people do really well is not only is it a bit of a waste of time, but it's also exceedingly hard because the technology is still very limited. And so one of the big lessons we learned was scope the problem well, the technology is still very limited but it's within the limitation so the technology that you can do something really interesting and something that can impact people immediately rather than in the next 30, 40 years.

Levi: Yeah.

Satya: So these were all lessons that led to the founding of this company.

Levi: Okay. So let's go next there after the one question. You mentioned that humans are really good at some things that teachers aren't or that the computers aren't and that when you looked at getting a computer to teach, it was very difficult. So what are teachers so good at? What was it about teaching and learning? What's so complex? Why was it so easy for humans to be great teachers and not easy for a teacher to be a great ... For a computer to be a great teacher?

Satya: Well, I mean, that's a great question, a bit of a doozy but let me attempt to answer this.

Levi: Yeah.

Satya: So why do teachers do well? So first of all, teachers motivate. More than anything else, I have the suspicion that people learn from people better than they learn from anything else. Right? And teachers are brilliant at not just motivation or motivating people but the ability to communicate ideas using references, using the shared background knowledge of the world, understanding the cognitive, I guess the cognitive space of what they're trying to teach, as well as how people are receiving the knowledge, right? Are they motivated, are they bored, are they excited? What are they interested in? This ability to personalize and this ability to figure out the cognitive state of the learner and to deliver information in a really impactful way is like a deeply human process. They're brilliant at it.

Levi: Yeah.

Satya: Machines are not going to get there for a very, very long time, if ever. And that's what we realized that this is actually a problem where people are brilliant. This is a deeply human process. Computers are terrible at this. It's a waste of time trying to do this, but computers are great at other things, right? And the ... While the technology is still limited in what it can do, working within the limitations exploiting what they're very good at and helping improve things like productivity for people and the symbiosis between getting humans to do what they're very good at and getting computers to do what they're very good at is really how this whole field will advance over the next 20, 30, 40 years, in my opinion.

Levi: Okay. So then let's go to what Merlyn Mind is and how you kind of scoped the problem. You mentioned that you learned all these things about education, about AI, about where it works, where it doesn't work and then as a result of all those learnings, you had to start a new company. Why? Why did you have to create Merlyn Mind and how did you focus? What was it that you thought you could do with AI to make humans and teachers and computers work better together?

Satya: Sure. Yeah. So, I mean, so some of the early inspirations for this company were digital assistants like Siri, like Alexa but even earlier, again, I'll answer the story, but I'd love ... I'll answer the question, but I love framing it within the broader context of computing-

Levi: Awesome.

Satya: And stories.

Levi: Yeah. Do it.

Satya: Yeah. So one of ... So as you know and as I mentioned earlier, I'm a huge fan of science fiction and back in grad school, in the early 90s, I read this book by William Gibson called Neuromancer. Neuromancer is arguably one of the most important science fiction novels of the last 30, 40 years. And so neuromancer is a book that spawned off this entire sub genre of science fiction called cyber fiction. And it's the story behind how William Gibson came to write Neuromancer that's very interesting, but the book itself dealt with grand themes of humans being immersed in computing in this virtual world and so the whole field of we are the matrix movies, they all inspiring to this novel in some fashion.

Levi: Okay.

Satya: And so William Gibson, when he was ... When he wrote Neuromancer, just before he wrote Neuromancer, he was a young man. He was basically unknown. He had only written two short stories ever in his life. One of which was Johnny Mnemonic which has been made into a movie starring Keanu Reeves, right? And so he was given ... Suddenly he was given an advance and he had ... He was maybe 19, 20, 21, at that time, very young fellow and given an advance. There were all these people expecting him to be the next big thing and he had a bit of a writer's block. And so the story goes that one day he was walking down the streets in Toronto, in the 1980s, at that time, he had never seen a personal computer or interacted with one, right? The PC, I think the IBM PC was 1981 or something like that and he hadn't really interacted with one. And so he was walking down the street and he happened to look into a storefront and it was a video game market. So in the 1980s, there was Space Invader, Pac-Man, these were the early video games. And what he saw was the people who were playing the game were so entranced by this interaction with computing, by this interaction with this virtual world that it spawned this imagery for him of people disappearing into the computer and that's how he came to write Neuromancer. So for me, books like that were a deep inspiration, the Star Trek effect of it, captain Picard is talking to a computer, this data, this ability to interact with computing in this very seamless, very intuitive fashion has been an inspiration throughout my career. And so when Siri came out, it was really interesting, of course, before Siri there were assistants like Jabberwocky you could basically type to it, it could type back, Eliza, the early chat bots. All these are things that we all played with and Siri came out, it was very compelling and then Alexa came out in 2015 and it was the first significant concerted effort to bring voice computing into the world, right? 2014, 2015. It's a brilliant inspirational product, right? So when we looked at all of this, we said, "Look, these guys are going after a very large open domain problem. They're going after assistance, helping people within the flow of our lives and that's a hard challenge, but of course, these are big, huge companies with thousands of brilliant engineers." And we felt that there was still an opportunity. In fact, the way the field with advance is lots of people, lots of companies like us taking cuts at making advances in specific areas. And we felt there's still an opportunity to take these digital assistants and bring them into fields like education. And we all also realized back from my ... At Watson intelligent tutoring experiments that scoping the problem is really important. So we said, "Okay, we don't have to be Alexa or Siri, we can be something different. Right? We don't have to try to be solving the open domain problem, we can solve something very specific." And so kind of a hypothesis behind this company, one of the many, but one of the more important ones is that if you scope the problem, if you say, "I am going to help a teacher or a student and further, I'm going to help them do something very specific, a teacher trying to teach in a classroom or prepare lessons at home or a student trying to do homework. If you scope the problem quite well and then you say, "And now it's all about trying to build all the ... Bring all the advances in AI that are happening through the service of this problem, we can actually get people to interact with assistants in a very deeply intuitive fashion." And so that's what we are about in this company. So the hypothesis here is we do this well, you get real adoption, you make a real difference to people, you understand what machines can do really well, which is take the burden of doing mundane and things away from them and improve their productivity, leave them to doing what they do really well, what humans do really well, which is teach, impart knowledge, impart wisdom, motivate their students, et cetera. So that's the idea behind this company.

Levi: Okay. So you bring some of your former team, you attract a lot of other really amazing people to this company and you go to say, "We're going to help teachers with AI, we're going to build a digital assistant, we're going to help them with their work not replace them." Were there any surprises along the way? Has it been smooth sailing or did you learn ... Have we learned some things in the last?-

Satya: Well, I mean, yeah, so even here, lots of surprises, right? So maybe again, a story here would illustrate the advances we made here. So when we first started this company, as I said, we were pretty inspired by Alexa and we took the inspiration literally, which is Alexa, is this voice assistant, at that time, it didn't have a display. It was this Echo Dot. And we said, "We need to build a voice assistant for education but we thought, okay, we'll also power a display." I think the first Alexa display product had just come out at that time but we'd already started and said, "We have to be about voice being there to help orchestrate a teacher's workflow but it has to manifest itself on a ... The assistant has to manifest itself on the front of room, classroom display." And then in the early days that we attracted a very important investor from the valley, one of the biggest VC firms and they were very interested in what we were doing and they came down and they said, "We want to see a demo of what you guys are building." And we've been tooling on for about a year and a half at the time, made some advances. We realized we had to scope the problem even further to a teacher in the classroom and then the demo got inaudible high stakes demo and we failed. We had an HVAC system go off, the voice translation wasn't perfect and so it was an underwhelming demo, we never landed the investment and it made me sit back and think very deeply about what we were trying to solve. And so, I mean, this is just in the user interface. I'll talk a little bit more about the jobs to be done or the specific problem in a second. But what it made me realize is first of all, voice as a modality is a bit of a difficult modality to bring to life and to get users to adopt it. Among other things, what voice computing is trying to do is replace both the keyboard and the mouse simultaneously. We use keyboard and mice to interact with computing and you're taking them both away and saying, "Talk to a machine instead." And we also know that you can't talk to a machine like you and I can talk to each other.

Levi: Right.

Satya: So they start putting some real barriers to how it'll work, where it'll work. And then further, there's all kinds of situations. Classrooms have unique noise and education, the language is very different. You're saying graph X minus three and most of the speed systems think you're trying to say graphics or something. So there's all this notion that look, I need to build something in a very domain specific fashion. I'm trying to replace the keyboard and the mouse with just one modality and that's a bit too far. So we stepped back, thought about it and said, "We need to bring in a second modality or more." It has to be ... It's not about voice computing as much as computing and a much more native interaction with computing coming to the rescue of, I wouldn't say rescue of teachers, which sounds very grand, but coming to the assistant of teachers. And so we said, "Okay, let's start bringing in other modalities like an intelligent remote which has ... Which allows you to interact with clicks and with an inaudible interact with computing at a distance." So these we're some foundational insights that it's about multimodality, it's about bringing in two or three modalities and stitching them together. So it's very natural for them to use voice where it makes sense to search and use this intelligent remote to click on things and to advance their presentations. So these were important foundational lessons.

Levi: Right.

Satya: But the most important lesson of all was about what do you do with these things? And you had a very big role to play, right? In fact, maybe I need to turn the lesson back ... The question back to you, maybe you want to tell us a little bit about jobs to be done and then I can tell you how we are applying it here.

Levi: I won't take the mic for long, but basically at any innovation, you have to start with a problem, an opportunity, what are you going to solve, right? We were going to bring voice and AI into the classroom and you could do almost anything with that. So where do you start? What is actually needed? And we went into classrooms and we sat in classrooms and we talked to teachers and we listened to them and we said, "What are you currently doing? What's hard? What would you like to do that you're struggling to succeed at?" And what we heard was, "I don't need another app. I have so many apps. Don't give me one more app. Don't give me one more content resource, help me use all this amazing technology I have. Help me manage it so that I can keep my eyes on my students and I can engage with them without having to constantly be going back and messing with the technology." And so we came back together and I talked to you and we talked to the team, we said, "Look what we're hearing," and then we saw a starting point. It wasn't the end point, but it was, wow AI could be very, very good at helping teachers orchestrate all the tools they're using. So yeah, you keep going then so what would that lead to?

Satya: Yeah, no absolutely. So that's the thing. So we talked about scoping a problem. We talked briefly about domain specificity, about building computers that understand the language of the domain. When you say, "Open my presentation on Google slides," you're not saying Google flights, it's slides. Right?

Levi: Right.

Satya: So there's ... All these things are important, but my most important of all is ... And we talked a little bit about, look, what are computers good at? What are humans good at? And so the most important lesson of all here is it's about deeply understanding human workflows, the teachers workflow. I mean, a successful AI application, in fact, all the successful AI applications on the planet, have had these features or these attributes. They're domain specific, they're solving well scope problems and they're deeply aware of the workflow of the specific task they're doing. So in our case, we had to study teacher workflows very deeply realize that ... In fact, because of ... Even before COVID, but especially because of COVID, there's so many apps in the classroom, so many devices, each student has a laptop, teachers have their own computer. There's a big screen, smart TV or a smart panel, the document cameras. And they're working with 30, 40, 50 different applications during the course of a teaching week. And they're spending a lot of time orchestrating across all this technology and so in fact, you led the effort, but what we discovered is if we take some of the burden away, make it more intuitive, enter them from the desk so they're more with their students as opposed to navigating all the applications and these devices, they're likely to want to use this more, they're likely to spend more time with their students, they're likely to be less burdened cognitively by having to switch between them. So the final piece of the puzzle, but the most important one is, build an application that solves a problem for them that's within the flow of their work. So if you put it all together then where about domain specificity, scope problems, deeply workflow aware and multimodal. These are the foundational elements for what we're doing with the Merlyn digital assistant for education.

Levi: Right. And what we've seen so far from educators is they're pretty thrilled that someone is bringing this technology just for them. Right? I think if you look back in the history of education technology, we're not always getting the most cutting edge tech into classrooms first, in fact, lots of times, I don't know, for whatever reason, AI hasn't really benefited education too much yet. And that may be because people didn't scope the problem right or because they didn't understand the technology and what it was capable of. So there were a few things that stood as barriers to bringing AI into education. Can you talk a little bit about the privacy and security piece of this and the trade offs you saw happening on the consumer front and how that directed you into starting this company and the way you did?

Satya: Yeah, no, that's a great question, another foundational pillar for what we do. So again, when I look at the consumer digital assistants, while they're being deeply inspirational, one of the things that we saw and one of the barriers that we saw was that this whole idea that there's this choice between this, you have to give up your data for me to do something useful for you, we felt that was a bit of a false choice. We felt that, look, why can't we simply get you the technology without actually taking your data and selling it to advertisers or doing something with it? And in fact, in fields like education, healthcare, a bunch of public sector fields, very sensitive fields where data is actually sacrosanct and you really cannot be irresponsible with how you handle it, you don't really have a choice. You have to take care of the data very, very carefully. So we built this company with privacy at it's core. We do not ever sell people's data, whether it's voice data or any other form of data. In fact, we delete it. We don't even use people's voice data to train and improve our own models and we are in the business is selling you a product that does something useful for you and you are not the product and your data's not the product. We sell you a product and your data is yours and we delete it and we don't use anything with it. We don't use it to do anything with that. So privacy is a very foundational principle here and this false choice of giving up your data so you can have access to computing, is something we reject, we reject the premise and then we are here to try to bring you the latest advances in AI to improve your own life at work.

Levi: So let's talk about, is it working, we've now been at this for about four years. You founded Merlyn Mind almost four years ago, right? And what's going on? What's the status of classrooms and teachers using Merlyn in the classroom and what does it look like when they get assistance from Merlyn?

Satya: Oh boy. Levi, the response has been overwhelming. We have so many fans and I'm filled with how well we've been received by the industry not just by teachers but by administrators. I'm also thrilled with this mission attracting incredible talent to this company, people like you joining us. It's a mission that everybody can get behind. Using AI to improve education, who doesn't want to be a part of that?

Levi: Right.

Satya: So, yeah. I mean, the response has been amazing. So we've been in pilots, we've been in over a couple of different classrooms. Anecdotally, what we know is teachers are using Merlyn from the data, teachers are using Merlyn every six minutes and that's just the beginning. We are still working out exactly what to do with Merlyn in a way that deeply improves productivity. We're kind of scratching the surface, we know it's about creating shortcuts for them, taking away as much of the mundane clicks and keystrokes and interactions with computing as we can and automating them with shortcuts but even where we are, even given the few things Merlyn does today, it's been a ... The response has been tremendous, people love it, they love what it does today and they love the vision for where we're taking this next.

Levi: So what is that next thing? So let's just say like fast forward 5, 10, 15, 20 years, if Merlyn Mind is successful, what changes in classrooms around the globe? What looks different with how teachers and students interact with computing as part of the learning process?

Satya: Yeah. So I can give you again, look 20 year horizons, very hard to predict but I can give you a five year view of where you think are going. I mean, there are many different ways to look at Merlyn Mind the company and the Merlyn assistant, but one view that I have that I'd like to ... One way I look at this thing is, we're kind of at the intersection of process automation and digital assistants. So with bringing AI and AI assistants and marrying them with automation techniques to bring you advances in productivity. So what this means is in the near future, in say 3, 4, 5 years time, we would've built all kinds of automation across all kinds of EdTech apps. So we are not bringing yet another app, we're an infrastructure layer that allows teachers to do something that would've taken 5, 6, 7 steps and a bit of cognitive word associated with it. That we're taking all these things, but we can take that and collapse it all into one step, one voice command or one click. An example being, share this link with all my students. So normally a teacher would have to open Google Classroom, find the student roster, copy the link, put it into an email, send it out and we can automate the whole thing with one command. So the near future, the 3, 4, 5 year future of where we are going is we'll have all kinds of automations across the most popular EdTech apps. In the classroom, teachers will be controlling student computers along with their computer, which they control today with Merlyn. So for instance, let's say you're a student in the classroom show device, laptop up on the main screen because you solve the problem that you want to use as a teaching moment for everybody else. So orchestrating across all the devices, having Merlyn suggest some interesting new pieces of content. So the classroom experience becomes much more fluid with all the technology that you're interacting with in a seamless fashion. But Merlyn also follows you home, helps you prepare for class as a teacher, helps students find their homework, answers a few questions for them or finds them the best human tutor. Again, we talked about tutoring being a deeply human process.

Levi: Right.

Satya: Beyond answering a few questions here and there or beyond things like finding your homework or finding information that's relevant to you as a student, we don't envision computing will actually significantly solve the problem of teaching people by itself. But Merlyn's here to do something very useful by making them more productive. So that's the future that we envision, Merlyn is in student computers, smartphones, in the front of room display, in teachers computers, it's orchestrating across all the things that teachers interact with. So that's the immediate future for Merlyn.

Levi: And that's really our focus where we're looking now, but how do you see that fitting into the larger picture of how technology and human interaction evolves? What's going to happen with digital assistants and computing and humans as we look forward?

Satya: Yeah. That's a big question. So if you step back a little bit, what I am thrilled by all the advances that are happening in AI at the forefront of the science, at the frontiers of the science. DeepMind is a deeply inspiring company. DeepMind, OpenAI, these companies are pushing the science forward at a thrilling ... In a thrilling fashion and it's evolving very rapidly. But the science by itself, while what they're showing the ability of a computer to generate language, generate code images, co-create with you, it's thrilling, how we take that and apply it to a specific domain like education or healthcare is where all the action will be. It's the action, I mean, of course there's this action at the forefront of the science but there's a huge gulf between what's happening there and what we could do with it to improve the work. And that's where we play ... Where people like us will play. I look at this company as ... We are one of the translators of the science to products that make a difference. So the near future will be about taking these advances and applying them to domains like education, healthcare, heavy industry, bringing deep productivity gains, getting computing to be much more inaudible to use, voice and multimodal, gestures and getting computing to be ... You're not even thinking about it. The best technology fades into the background in a very seamless fashion. You're just interacting with life in many ways and trying to accomplish tasks. So the next 30 years will be about building a lot of these applications. It's still an art, it's not a science. To do this well, if you recap what we talked about earlier, the work you did about understanding workflows very deeply, the importance of multimodality, the importance of specifying the problem and scoping it, the importance of building something that's very domain adapted and right specific to the domain. All these foundational pillars are going to be ... They're critical to taking the science and turning them into products that really change the world. And that's really how I see AI, the AI revolution rolling out. This is the most profound transformational revolution that I've seen in my lifetime, for sure. But it'll make ... It'll basically roll out industry by industry, application by application, carefully build, carefully engineered and eventually it'll have deeply transformative effects. The idea that you can even talk to a computer, leave alone, create with a computer is just completely mind blowing to me and I'm looking forward to this future materializing and being a part of it.

Levi: Well, it's very inspiring to all of us who've joined Merlyn Mind. We have an incredible company with people to every different discipline and area of expertise that are committed to this vision of helping to create a better future with technology. You, as our leader, have inspired us. We're excited to be part of this but I'm curious as we look at the world and you see all the challenges and the problems and current events, how do you find such optimism about the future? I know that AI plays such a big role in it. What is it that inspires you about why the world can be better because of technology?

Satya: That's a great question. I mean, so I'll go back to DeepMind and it's from the inaudible who I'm a huge fan of, I'm ashamed to say I'm a bit of a groupie. For those of you who haven't read up about him or heard him, I would advice you to go listen to his interviews is a brilliant mind, very, very brilliant fellow making a huge impact in the world. And he said what a lot of us feel. Now the world has a lot of problems, we all know that, right? We're living through a pandemic, there's wars going on, there's global warming, huge problems. And so how do you solve them? You can either say, "I'm going to have a ... I'm going to impose or I'm going to hope for a transformational change in human behavior," or you can say, "I can take technology and its transformative potential and build solutions with it that could actually transform the world." And so I'm a deep believer in the ... And trying to change human behavior is very hard, we all know that but I'm a deep believer in the enduring power of technology to transform the world, to change the world. Already, we are living in incredible times, thanks to all the great technology, science and technology that our forefathers and our ancestors basically created. You and I are talking over a computer, you are a few thousand miles from me. I come from India and flying across in this several ton machine over the oceans and we have great progress in medicine that improve lifespans. So technology has had an incredible impact on the world already, but it'll continue to have a deeply transformative impact on the world. So I'm huge believer in it. I'm an optimist, I'm a technologist and I like to believe that throwing yourself at hard problems and just chipping away at them will lead to us creating this future vision that people in Star Trek and in these incredible science fiction always imagine. So that's my inspiration.

Levi: I love it. Okay. So last question for you then, why education and why is that the problem that needs to be solved? And if you are ... If you do have this impact on education with AI, why does that make the world a better place for all of us?

Satya: Well, I mean, education is so foundational, isn't it? It touches everybody on the planet from the remotest corners of India and Africa to places like New York city. It touches everybody. And so what we believe is if you make a change in ... If you help improve education, you'll create a world where people are more informed, they make better choices, they have much better life outcomes for themselves. We have more people trying to solve these very difficult problems we're all trying to face and they're trying to solve it in more informed fashion. So education's one of those fields that has a deeply transformative impact on the world and improving learning outcomes, helping teachers by freeing their time up to improve learning outcomes, they're the best levers to improve learning outcomes. I can't think of a better problem to bring AI to, especially a field like education where we've been slow to digitize it's, I wouldn't say it's over the last industries, but it certainly lagged behind other industries a little bit in terms of embracing digital technologies. So there's huge untapped potential here and we can make progress within the next 4, 5, 6 years with AI and technology and education that's why it made sense for us to start here.

Levi: Well, Satya, thank you so much for joining us for the conversation today with this podcast, it's called Supervised Learning because we go and talk to experts like yourself and understand what we can learn and how it can guide us in our work. So we're excited to bring others on the show and I'm sure we'll have Satya back here again and again for more conversations. So thank you.

Satya: Thank you, Levi. It's a pleasure.

Levi: Thank you for joining us for this episode of Supervised Learning until next time, keep learning.


How can we translate the thrilling advances at the forefront of the science of artificial intelligence into solutions that can make the world better today? In this episode, we explore the past, present, and future of AI and how translating AI into solutions to help people be more productive as they do work that is uniquely human is the future. Satya shares childhood stories about his obsession with science fiction and technology and how that led to his passion and career as a technologist. We get a front-row view to the exciting times at IBM as Watson beat Jeopardy! and the world saw the potential of AI. We also learn from the failures and difficult lessons learned in Satya's career that led to the foundational insights and inspiration for starting Merlyn Mind and building a digital assistant to help teachers.