ACM ByteCast - Ramesh Raskar - Episode 13

Episode Date: March 15, 2021

In this episode of ACM ByteCast, Rashmi Mohan hosts Ramesh Raskar, Associate Professor at MIT Media Lab where he directs the Camera Culture research group. He holds more than 90 patents in computer vision, computational health, sensors, and imaging, and has co-authored books on Spatial Augmented Reality, Computational Photography, and 3D Imaging. His many awards and recognitions include the prestigious 2004 TR100 (MIT Technology Review), 2016 Lemelson–MIT Prize, and 2017 ACM SIGGRAPH Award. Raskar discusses the fascinating research field dedicated to capturing and recording the world in new ways. He explains how computer vision provides a new eye and brain to help us both in seeing and processing the world and shares his recent work with extremely high-speed imaging. He also mentions his COVID-19 project: developing privacy-first contact-tracing tools to stem the spread of the outbreak. Raskar also discusses balancing entrepreneurship and research, and his REDX project to bring peer-to-peer invention to his students and advance AI for Impact. Follow Ramesh and PathCheck Foundation on Twitter. Camera Culture research group at MIT Media Lab. PathCheck Foundation (COVID-19 research & technology)

This is ACM ByteCast, a podcast series from the Association for Computing Machinery, the world's largest educational and scientific computing society. We talk to researchers, practitioners, and innovators who are at the intersection of computing research and practice. They share their experiences, the lessons they've learned, and their own visions for the future of computing. I am your host, Rashmi Mohan. What you see is what you get. And yet, what if you could see much faster, in greater detail,

and just a whole lot more than what you do today? Our next guest is a visionary, pun intended, in the field of computer vision and computational healthcare. Ramesh Raskar is an associate professor of computer science at MIT and the head of the MIT Media Lab's Camera Culture Research Group. Amongst many other accolades, he received the Lemelson-MIT Prize in 2016 and is a renowned speaker. Ramesh, welcome to ACM ByteCast. Thanks, Shalini. How are you? I'd like to lead with a simple question that I ask all my guests. If you could please introduce yourself and talk about what you currently do and give us some insight into what drew you

into this field of work. Thank you, Rashmi. You know, my passion is to make the invisible visible. It has to be magical to make it interesting. So a lot of my earlier work has been in looking at computer vision and imaging to build cameras that can see around corners or build medical devices that can see inside the bodies in unique ways. And nowadays, I'm also focusing on how to make the invisible data visible, looking at how can we do machine learning that's privacy-preserving for health. Got it. And as a young computer engineer, when you started out,

what really attracted you to this field, Ramesh? There are so many different options to pick from when you sort of first graduate from college. Yeah, you know, I mean, I got into the whole field of computer vision, computer graphics, because of the movie Jurassic Park. You know, I saw, wow, this dinosaur. In fact,. So, it's just amazing to see them on big screens. And

so, I went to UNC Chapel Hill, North Carolina, which is the number one school in computer graphics. But near the end of my graduate school, I saw South Park. And I said, hmm, South Park has pretty much no computer vision, no special effects, but it's probably as entertaining as Jurassic Park. And that was kind of a downer. And I had the kind of classic South Park versus Jurassic Park moment. And what I realized is that the functional aspect of what we do is so much more than just the pure, the visual aspect of it. It's not just the special effects, but it's also the story and what happens behind it. And when you kind of think of that in reverse, I realized the ingredients and how they come together are so much more important than how the polished presentation is at the end. And so as a

fundamental science, I think that's always an important question. Should you be driven by great ingredients and great tools, or should you be driven by the final product that you're going to deliver? And over time, I have realized that what really matters to me personally is building tools, you know, building this magical tools that creates something that's, you know, is seemingly impossible. And, you know, when somebody looks at it and says, how did you do that? And then I get the pleasure of explaining what the underlying tools are doing. And so whether it comes to computer vision or computer graphics or digital health, the underlying tools that I have been using are roughly the same, which is

imaging, computer vision, machine learning, and so on. Got it. Yeah, I mean, I can understand. I mean, Jurassic Park and South Park, polar opposites in terms of content. But what is interesting to me is I was doing a little bit of research as well. And it seems like, and maybe this is just my perception, but seems that computer vision as a field of research seems to have gained a lot of popularity in the last few, maybe decades. What do you think caused this sort of,

this spike or this interest? Was it a certain innovation or a certain need that really sort of drew people into this field of research? Yeah, if you think about human intelligence, I would say computer vision is probably the most challenging problem. If you just look at the human I.O., our seven senses, the visual sense is probably the most important out of the seven senses.

But when it comes to output, we humans are actually pretty terrible at high-bandwidth output. It'll take us forever to draw a diagram to explain to somebody or to explain verbally what this all means or to gesture to explain a scenario. It's like playing dumb charades. So humans are terrible at output and we are okay at input. And probably the best input is the visual input. So even in terms of human intelligence, computer vision plays that role of kind of seeing the world. But of course, the challenge is that we don't see with our eyes. We just record with our eyes

and see with our brain. And so computer vision is really a fusion between how our eye and how our brain works together. And so I've been fascinated with both aspects of it. How do you create a new eye? And how do you create the new brain for visual processing? So to your question about the importance of computer vision, I think it'll continue to grow. For the first 30, 40 years of research in computer vision, the emphasis was on very model-driven techniques.

And thanks to new data-driven techniques with machine learning and new ways to solve inverse problems, we have just seen just a tremendous growth in how we solve, how we think about computer vision and how we solve, you know, the big societal problems with computer vision. Yeah, that's really great to hear because obviously there are so many more problems to be solved. And, you know, new researchers coming into this field have a lot of exciting things to sort of crack open. One of the things that I want to touch upon is the fact that you were talking about data, right?

There is a proliferation of images and videos now. So tons of data available for anybody who's trying to solve problems in this space. What's your take on that? You know, is it a good thing to have that kind of data to be able to sort of feed your models and get more accuracy? Or are there financial or environmental costs associated with cleaning this data and processing it? Yeah, I mean, any AI machine learning project has three parts, capture, analyze, and act. And for the longest time, we thought the analysis, the computer vision problem itself is probably the most exciting.

And then we actually realized the first part, which is how do you capture the data and how do you act on it is also equally important. And so previously, capture was with standard RGB cameras or very small data sets. And now, as we can see, a whole industry, a whole research field has come up. And how do you even capture the data? So just creating data sets, creating new types of cameras that can record the world in new ways, you know, that has exploded significantly. And then also, how do you act on the analysis you have done? As you can see, whether it's self-driving cars or AR, VR or medical imaging, what do you do with it? How do you make it actionable? Has also become a very interesting research problem.

But yes, I mean, it has lots of downsides. When it comes to data, there are issues with privacy, issue with ethics and bias. When it comes to its analysis, there are environmental factors, like you said, there are financial costs. There are also issues of, is this a winner takes all? I know the entity that has more data and more resources will leap ahead.

And especially folks who live in low-income and middle-income countries, will they ever be able to catch up with this? And the same thing with ACT. There are ethical issues in how do you make that analysis actionable? Is that going to lead to job displacement? Is this going to make all of us very dumb? And are we going to lose a whole generation that understands the fundamental physics

and principles behind all these activities? Or are they just going to create data-driven models, kind of data in, data out, black box models? So yeah, there are a lot of underlying questions that we need to tackle as we shift away from a kind of a model-based, kind of, you know, rule-based computer vision and machine learning to something that's more data-driven. Got it. Yeah. And I think you, you know, bring up an excellent point. I think,

you know, as we consider these new areas of research, we must be looking at not just the great advantages that we get from it, but, you know, how do we do this? How do we conduct this research in a way that is fair, that is equitable across all of the various lines that we have drawn, you know, societally speaking. But I'd like to sort of, you know, dig a little bit deeper on the point that you were bringing about capturing data, right? Many of us have seen and been completely wowed by your video of light traveling through a plastic bottle. I know I've seen that video multiple times. So your work in femto photography is super exciting. And we'd love to hear, you know, your journey through that area of work and how have you seen that evolve? Yeah, certainly. As I said, my passion is to make the invisible visible. Any research we do has to look magical. And when

it comes to this camera that can see around corners, I was very inspired by some of the work from people like Steve Sides and Kiros Koutoulakos at University of Washington and Toronto, respectively, on inverting light transport by observing light in the scene and figuring out things about the scene

or about the light. And I was very inspired by that work in 2004, 2005. And with my colleague, James Davis at UC Santa Cruz, I said, hmm, you know, I have an idea of what we can do with extremely high-speed imaging. And at that time, James had some other ideas on what you can do in terms of 3D scanning

with extremely high-speed imaging. And I said, hmm, maybe we can build a camera that can see around corners. And James said, that's maybe we can build a camera that can see around corners. And James said, that's too challenging. So I ended up talking to a whole bunch of people all over the world who know something about this. And, you know, I'm not a physics person, so it took me a long time to understand

the underlying physics, the underlying quantum theories, and so on. But eventually, with my postdoc, Andres Welten, who is a faculty now at the University of Wisconsin, you know, we were able to build a system. And Rashmi, as you said, you looked at that video so many times. And every time I looked at the video, I appreciated it, but always found some fault. It's like, this doesn't make sense.

That doesn't make sense. And I would always have an argument with Andreas, my postdoc at that point. And Andreas would say, I don't know. This is raw data we captured. So what are we talking about? It cannot be wrong. The physics cannot be wrong. The physics cannot be wrong.

And what we started realizing is that there are all these amazing space-time effects that you can observe when you start recording at, you know, this femtosecond, picosecond resolution that are just counterintuitive. And so we can take all that information and invert that and infer things about the scene that, as I said, just seem magical. So that's kind of the root of that whole project. Originally, we thought we're going to build a camera that can see around corners. But in reality, once we can image the world at femtosecond, picosecond resolution, then we can do so much more. We can create new medical instruments that can see deep inside our bodies without x-rays this is a new nsf project we have a moonshot project

to create five micrometer resolution at five millimeter depth inside the body so that's in collaboration with cmu and rice and harvard and some other universities and we can create cars that might avoid collision of what's around the bend. We might be able to create solutions for firefighters to find survivors in hazardous conditions. So from a micro scale to room you can, you know, from Doc Edgerton kind of bullet through an Apple or all these TV shows that can smash things. But then they're only going at 4,000, maybe 5,000, sometimes tens of thousands of frames per second. But when we go to roughly 1 trillion, not a million or a billion, but 1 trillion frames

per second, the world really opens up. And the exciting part is once we have this new way to capture, again, capture, analyze, act, once we have this new way to capture, we can analyze and act on this information in completely unique ways. So our dream is to create a new field of computer vision that defines what a camera means. Because we think of a camera of recording what you can see in front of you. But if you can see what's around the corner, then all the rules of computer vision get rewritten. And so we're very passionate about creating a whole field of computer vision about things that you cannot see.

It sounds like you said, magical. It sounds like a blue sky project somewhere far out in the future, especially because you're saying we'll be able to see things that currently the human eye can't. I can't see around a corner and you're going to enable me to do that. How far do you think we are from this actually being out on the streets, Ramesh? 10 years ago, I thought, hey, it's only 10 years out. But I started the project 2007, so it's been 13 years. But if you see the latest SPAD cameras that are in the latest iPad and iPhone 12, and I'm sure you'll see the same thing on Samsung and other high-end smartphones,

the underlying technology now is available at consumer form factors. Not at the same resolution, not at the same sensitivity, but they are here. And even if you think about LiDARs on self-driving cars or autonomous vehicles, they're reducing the cost from tens of thousands of dollars to below $10,000 now. So the underlying pieces of the puzzle are actually already here. And the algorithms from my team and my former group members who are now faculty at some of the top universities, and of course, a large DARPA project that was sponsored based on our research, a program called Reveal, which is a $30 million DARPA program that was launched, partially inspired by our work. Those $30 million have

really taken the field to the next level. So, Rashmi, I mean, we are, I feel really close. I won't say 10 years anymore. Maybe we are two or three years out to see robust, at least benchtop solutions to see around corners for industrial applications, medical applications, you know, in the beginning, and maybe in five years, you know, consumer applications. That's incredible. You know, Ramesh, I was doing a little bit of research about your work in preparation for this conversation.

And one of the things that struck me is how much of your work is relevant and something that is, you know, impacting society in a way that is tangible and is something that is close and meaningful to us. I know that you've also been doing work in these times, in this pandemic times, helping with COVID-19 contact tracing. And I'd love to hear more about that as well. I know it may be a little bit of orthogonal to the traditional work that you're doing, or maybe not. I know it may be a little bit of orthogonal to the traditional work that you're doing, or maybe not. I'd love to hear more. No, not at all. As I said, my whole life and my whole research direction is all about making the invisible visible, whether it's invisible object

or invisible data. And so the problem that we have been tackling for the last seven or eight years is this notion of invisible data. And the idea that if you magically have a God's eye view or a bird's eye view, without making it too religious, of the world, let's take the example of health, let's take diabetes. If we know every person who's diabetic, what their genomics, what their environmental, behavioral,

socioeconomic factors are, what treatments are working, what lifestyle changes are helping, and so on. If you have this bird's eye view of every person who is diabetic, one could argue that we can create the ways for health. It's like when you're using this traffic navigation app, Waze, from Google, it shows all the other people who have gone from Boston to New York, what route they took, and whether it worked out for them or not. More importantly, if I start right now from Boston

to head to New York, what are some challenges along the way? So Waze makes that invisible data visible to me by sharing with me in real time what has happened to all the people who are ahead of me from going Boston to New York. And a diabetic person should also get this dashboard of what would happen if they took every action. And that could motivate them, that could give them the right advice at the right time, and so on. So one could argue that many, many, many health problems can be solved overnight if we have complete visibility on everything that's going on in the health

ecosystem. But the challenge is folks don't want to share their data because of privacy or regulation or trade secrets, sometimes even national security. And because of that, the data remains invisible. So we have this classic trade-off between privacy and utility. When it comes to Google Maps and Waze Maps, we have very little privacy. We willingly give away our location data and we get great utility, which is traffic conditions. But when it comes to health data, we have the opposite,

which is we are not willing to share our data because of privacy and we get very little utility out of it. We just have to believe what our hospitals are saying and what our doctors are saying or what the research that's conducted on 50 or 100 patients is saying. So we have this kind of a dichotomy between privacy and utility of the data, but it's a false dichotomy. And when it comes to COVID-19, when the cases were going up in China and Korea in February, I said, why don't we do this in the U.S.?

Why don't we create this bird's eye view into everything that's going on about people's interactions, but at the same time, do this in a privacy preserving way so we can get both privacy and utility. So we launched the nation's first contact tracing app back in March. We also published the first algorithm, a decentralized privacy preserving algorithm in mid-March. And it has been great since then. The pandemic is just awful, but it's also wonderful to see communities come together to work on this very important problem of privacy preserving solutions for public health.

So that's kind of our effort. And we have just today, we launched our app in Minnesota, and we are in five US statesS. states and three nations. You'll hear about them very soon. So on one hand, the pandemic continues to create challenges. On the other hand, we can create these smartphone-based computing solutions that are privacy-preserving, but make the invisible data visible. I mean, it's the absolute need of the hour, Ramesh. I mean, obviously, like you mentioned, this has been an absolutely terrible sequence of events

that has happened, impacted the entire world. And anything that we can do to sort of ease the pain and sort of maybe at least stem this rapid spread of the infection would be tremendous. And I think data is definitely the solution to that. But what I also find interesting is that in your journey, in your career, there's been a lot of intersection of your computing work with healthcare. You've always been interested in it. How and why did that occur? Yeah, yeah. I think when it comes to

kind of the fusion of the physical and the digital, people who think about the atoms in terms of visual computing, it's displays and cameras and so on. They have this feeling that as long as they do really good devices, all the problems will be solved. And you can kind of call them kind of, you know, photon chauvinists. And on the other hand, there are people on the computing axis who think, you know, as long as you do great machine learning and computer science and signal processing and build great processors and so on, we can solve any problem. And let's

call them computational chauvinists. But I think the real action is at the intersection of the physical and the digital. And so it always fascinates me to think of that as a joint optimization, joint or co-design of the physical and the digital. So yes, a lot of my work thinks about, you know, how can we build medical devices that fuse clever sensing mechanisms with computational mechanisms. And that's the device you mentioned, iNetra, which is a device to get prescription for your eyeglasses by exploiting extremely high resolution displays on smartphones. Two, you know, as I mentioned, cameras that can see around corners. I mean, there are many teams that are working on lasers

and many teams work in computer vision, but maybe our team was the first one to think about them in a joint manner. And the same thing here with COVID-19, which is many folks working in digital health and many folks working in privacy, computational privacy, but I think our team was able to bring them together

and deploy them at scale as well. At the same time, there are many open research problems in this field of how are we going to create what I like to call participatory epidemiology, which is kind of this crowdsourced notion of public health, not relying on what CDC is telling you or the city or county is telling you, but the data is coming from the users in real time, again, like ways for public health. So there's something to be, there's something

magical when you try to think about when you do co-design of the physical and digital systems. Yeah, no, absolutely. You know, and the interesting thing that sort of comes through in your conversation is also how, you know, you've made this journey from being a researcher to sort of working in the practitioner's world so seamlessly. I mean, I looked at your LinkedIn profile, it shows you as a co-founder for so many initiatives. How do you straddle both those worlds as being an entrepreneur as well as, you know, sort of pushing forward on your research work? I believe in a very fortunate world where when I was a graduate student, the best thing you could do was write a paper, do your thesis, and maybe somebody will

pick it up. Maybe somebody will build something and maybe 10 years later, that'll become a real product. I think the beauty for the students today and researchers today is that they can go from an idea to impact, first of all, completely on their own and also in a matter of weeks or a matter of months. And that's just a blessing for kind of computational researchers nowadays. I think the mistake a lot of us do is trying to stay siloed, which is just do research or just do entrepreneurship or just do social impact. And I feel we can seamlessly move from them because to me, these are all learning experiences. You know, I'm the youngest of four, four siblings. And I thought that the more I talked to my siblings, I learned so much. So I had this benefit of being the youngest.

And I feel like as long as we move around in these areas, we learn from them. And as long as we don't have, you know, preconceived notions of what these particular concepts should be. So, you know, I started the health innovation team at Facebook. I've done several startups. I've launched a few nonprofits. But, you know, what I like the most is doing research and being a professor. And this is really a unique world we live in. And it really changes the way I think about research problems that impact how disparate ideas connect with each

other, also getting stimulated by a lot of young talent that I meet in different contexts. So I think being a professor is the best job in the world. And by dabbling in the real world scenarios informs, you know, how to be a great researcher. Yeah, I think you hit upon an incredibly important point, which is really the ability to leverage this community that we're surrounded by. Ramesh, I first heard you speak at the NASSCOM product conclave in India in 2014. And I know you were talking about solving problems at scale and you were excited about your work at the 2015 Kumbh Mela. And that's where I heard about your project RedX. I'd love to hear more about what were you thinking when you started that initiative and where is it today? Yeah, so Red X is an idea that, you know,

it's peer-to-peer invention, especially focused on AI and data. And as you probably saw when I was talking back, you know, five, six years ago, many folks, many talented folks in mid-income or low-income countries have this artificial barrier where they say, hmm, to do great work, I must go to MIT. And as much as I love being at MIT, I think nowadays the opportunities are everywhere. And anybody with access to the internet can probably make amazing things happen, especially in the world of AI and data. The challenge seems to be that we continue to teach our students and even our researchers

the way we used to teach them 20 years ago or 30 years ago. And when I teach at MIT nowadays, I realized that, you know, the students actually are very, very well-versed. You know, they're watching videos, they're talking to each other. And for me to walk into a classroom and teach from, you know, a classic textbook has a very limited value nowadays because most of those videos are available online. And what really matters is how to stimulate the peer-to-peer kind of invention and exploration. And that's why we launched this RedX platform, Rethinking Engineering Design Execution, of how do we stimulate and how do we get the youth to think about the societal problems

and apply computing research and computing ideas to those societal impactful problems. And I think that has really worked because I see the Red X. Unfortunately, when I received the Lemonson Award, which is half a million dollars that just show up in my bank account, I transferred all of that to start this foundation for Red X. And so we support lots of clubs all over the world, college clubs. These are college clubs for AI for Impact. And it's an ongoing program.

It doesn't scale as much as I would like to see, but it's ongoing in many parts of the world. But the main output from Red X is just the philosophy and the formula. So of course we have made it open source. So when I was at Facebook on my sabbatical leave, I launched two different labs based on the Red X principles, and we continue to do more work in that space. It's incredible. I think, you know, fostering that kind of, you know, freedom to innovate and

actually make an impact for societal good is something that, you know, I think as computer scientists, we have the power to be able to do that and definitely seems like a worthwhile investment of our time and energy. One of the other things that I also heard a lot about while I was looking up the work that you've done, Ramesh, is, you know, you speak about innovation, so growing your career and strategies to innovate better. Do you believe that innovation is something that is a learned skill and can be honed? Definitely, definitely. I think there is definitely, in the beginning, it feels like it's this chaotic process. But, you know, how to invent, how to innovate, how to work in teams, and how to stay focused on your goal, all can be learned and taught. In fact, I have a whole lecture, a whole document on that called Idea Hexagon.

And when I came to MIT as a professor in 2007, 2008, I was working in the industry for several years. And when I met my students in the first year, I realized, hmm, their expectation is that they'll spend a year or two just taking classes. And maybe in the third year, they'll start doing research and start inventing new things. And that didn't make sense to me. And I said, why can't you just start on day one, thinking big and thinking about crazy ideas, and collectively, we can guide you. And so I said, okay, I'm just going to put all my skills, everything that I have learned about

how to invent and how to research into a manageable kind of bite-sized format and created something called an idea hexagon, which is if you have an idea X, it tells you what could be next. And basically six, it's like an algebra on the idea. And by doing that, you can do that and have used that in many courses and many innovation workshops I have run all over the world. So that's just one example of having this training the muscle in your brain that's constantly thinking about if you know X, what is next?

Got it. Yeah, I did see, I have seen the lecture about idea hexagon, and I think it's fascinating. Has that framework changed at all or altered in light of the fact that, you know, we have a new way of working now where you don't have the kind of bantering of ideas that happen either in a lab or in an office or in a classroom environment or a whiteboard to brainstorm. What do you think is going to change now that we're all sort of remote? What do we need to do to foster that, the collaboration? That's true. That's true, Rashmi. I mean, we live in a whole new world and there's so many theories of what could happen even after the pandemic is over. But I'm fascinated. I don't have great answers, but the fundamentals will not change. I think I like to tell my students that when they wake up

in the morning, they should just follow five letters and I call them STAMP. Space, time, action, money, and people. So you wake up in the morning and say, do I need to deal with any of these five issues today? And research is all about stamp, space, time, and money, and people. So if you can do that, and the last one of that is P, which is people, I think as we go forward, we're going to work on problems.

If we think about pandemic, it's not just a biology problem, it's not a technology problem, it's really a social problem. It's about people, right? And so I can see tremendous progress in computing that's people-centric and not necessarily centered on physical, chemical, or purely computing principles. Absolutely. I think the biggest sort of problems we will have

to solve are around how people are feeling, how do we work and interact with each other, how do we build social connections? I mean, I have daughters in high school, and for them to be able to go through this entire experience of high school without meeting their friends or teachers in person has definitely been a challenge. And so I think innovation in those spaces will be welcome to parents like myself, and certainly the younger generation of students that are going through this experience. Your information, I mean, on a more lighthearted sort of note, you know, what do you do outside of work?

Like, what are your interests or pet projects, hobbies? We'd love to know. I'm sure our listeners would love to hear more as well. I'm a pretty geeky guy when it comes to that. I like to pick up, you know, one skill at a time. On my sabbatical leave when I was at Facebook, I said, I'm going to pick up rock climbing and skateboarding. So I picked up a bit of that.

And again, for me, anything that looks magical, something that seems impossible for humans to do, whether it's in terms of research or invention or just personal activities, it's what really attracts. So I love magic tricks. I love to learn how to do magic tricks and show them to my family members. So, you know, you'll see me, you know, even if I'm kind of on an airplane, I'm probably

looking at YouTube videos of magic tricks. And that's what really keeps me going. That's fascinating. We look forward to your next talk and the magic tricks that accompany it. So what is it that you're most excited about? Or what do you think is magical in the field of computer vision over the next, say, five years? I would go back to say, you know, we have an ability to now create a God's eye view. And again, not to be religious about it, but really kind of being omnipresent,

omniscient. And if we have that ability, we can solve many problems in society, whether it's health, transportation, climate, or even democracy with that. And we can do that only if we can solve the problems of privacy-preserving computation, privacy-preserving machine learning, privacy-preserving computer vision. So I think five years from now, we're going to look back and say, hmm, we had this untapped potential, and we kept on hitting against the wall of siloed data. And that's what was preventing this revolution in machine intelligence to really scale. And we have unblocked that with privacy-preserving computation. that this magical ability is to make the invisible visible, which is creating new computer vision on objects that you cannot even see, and creating new computer vision on data that's siloed

away is probably going to be something very important. And five or 10 years from now, it'll be like the green padlock on the internet browser. I mean, the mid-90s, if you had to type a credit card number in a browser, you would do that with trepidation because you didn't really know what's going to happen to that credit card number because those of who understood TCP IP protocols, those packets are going in the clear. So anybody could tap and look at your credit card number.

But of course, then we had HTTPS that creates an encrypted channel between the browser and the server. And all of us look at that green padlock and say, hey, this perfectly fine to use our credit card. And I think five or 10 years us look at that green padlock and say, hey, this perfectly finds a user credit card. And I think five or ten years from now we will look back and say, hmm, actually, I don't really care about sharing

everything about myself. My diabetic treatment, how the treatment is going, what my social graph is. All the cameras in my house can look at everything that there is. I can send the data in an encrypted form to some service where that service

cannot even see ever raw data, get me results from it. So we will live in this kind of information rich world that has very, very low friction in terms of the data exchange. And that will be kind of just a golden era that unleashes the value of computer

vision in terms of physics or in terms of the physical or the digital aspect of computer vision to bring us great benefits in personal lives, but also in the society. I think that sounds liberating. It sounds empowering. And I'm really looking forward to the vision that you just painted. You know, Ramesh, this has been, without being facetious, an eye-opening conversation. Thank you so much for taking the time to speak with us at ACM ByteCast. Thank you. ACM ByteCast is a production of the Association for Computing Machinery's Practitioners Board.

To learn more about ACM and its activities, visit acm.org. For more information about this and other episodes, please visit our website at learning.acm.org. That's learning.acm.org.