After LifeAfter

Welcome to After, Life After. This is a roundtable discussion picking up where the sci-fi thriller Life After leaves off. I'm Neil deGrasse Tyson from StarTalk Radio. As you know, I'm a scientist. I'm interested in science, both in the universe and in fiction and in any way gets delivered to our civilization and so I'm very excited to host this after show. Life After handles very deep questions that prevail in society today. They're not problems we have today. They're problems we might have tomorrow, like what happens to our digital lives when we're gone. Assume we have created digital versions of ourselves in advance. Or could you take an online personality and bring it back to life? So I'm here. I can't do this alone. So I've got three really good guests. Mac Rogers, Michael Litman, and Colin Paris. First, if you haven't yet binged on Life After, our conversation will probably contain some spoilers. I'm just letting you know. But don't worry. I'll tell you why. Because what you learn in this conversation will enhance your appreciation and understanding of the storytelling of Life After. First, have the playwright. You can't do this without the playwright. It was completely responsible for this. He's the playwright in series author, Mac Rogers. Welcome. Thanks very much for having me. So Mac, how'd you become a writer? Well, I was acting in children's theater plays and I was a little bit in a certain point. I just started thinking, oh, I'd like to write. So I shifted into writing plays and for many, many years, I've been almost exclusively a playwright. But then I wrote some science fiction plays that got the interest of Panoply, which led into into writing these audios. So my background is largely in writing science fiction for the theater. And you're here in New York City where we love our playwrights. Yeah, that's correct. That's right. Yeah. So Mike Littman, you're a professor at Brown University. How about you? I am not a playwright. But my daughter is. So that's really great. So how did I get into into computer science? So when I was little, the tear ice 80 came out and I thought, that's cool. I want to do that. And you're saying, that's Radio Shack. Yeah. Oh, yeah. Hard to show my street cred here. And yeah, it really spoke to me. And so I've been studying computer science ever since. And the AI part of it is, I don't know, it's just like brains are cool, man. I mean, it's so interesting to think about how you get the best line. That's the best line. That's already stolen it. Right now you get your lawyer's. I've already made it in an application. Yeah. Yeah. Okay. Cool. And call. How about you? I started out years ago in an undergrad with the hardware side where you had to actually wire these boards, right? You have to run wires and capacitors forever. That's like the 19th century. Oh my god. Okay. Then someone showed up and actually showed me a software program. And I could rewrite this thing in an hour. I will change. Off the side. It went to what software. You know, then did the PhD spend 20 years at IBM, now a GE. Oh, because the world evolves. Will you get your town? I mean, this, yeah, I was at Bell Labs, the Ford and Yorktown. Then I've also run. And so you, you had a man about the ten. No, no, no, you see the journey. Are you from the same planet as Jeffrey Holder Trinidad? Trinidad. Yes. I grew up in Trinidad. Antibago. That exactly hence the accent. Okay. So Mac Rogers, what were you thinking? When the various folks involved panoply and G, I got together to talk to me about doing this, you know, they we kicked around a bunch of ideas and sort of the idea that I think really sparked their panoples of production company. That is correct. Yeah. He's the GE. That's right. Yeah. Yeah. The one and only. Yeah. You know, we kicked around some, you know, a bunch of technology based ideas. But I think the one that really sparked everyone's imagination was the idea of social media presence, the huge personalities that we manifest online, that when we die now, we far enough into the era of social media that people die and leave a large public identity behind. What is that identity? And like, you know, could you construct an artificial identity from the leftover social presence? The story takes under the count of the idea like, what if and what if an artificial intelligence could recreate voices of people? It posits a world in which there is a fictional social media platform called Voice Tree where people leave voice messages behind instead of text ones and as it can recreate the voices of lost loved ones from those leftover voice messages. So does this make interesting storytelling because people fear it? You're doing a little bit of research I saw and, you know, some of the other folks here will have a much more sophisticated take on this. But I read that's not getting you out of the way. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. I'm the 101 guy. I'm the I'm the complete class. We're going into a there were three basic kinds of AI to which I guess are largely theoretical in Asia. There's narrow intelligence, which is like even so what you call me. Where there's that thing where it's like it would predict what Netflix movie you want to watch next. They can figure out when to engage the any like breaks on your car. The artificial narrow intelligence artificial general intelligence is intelligence that's equal to a human and then artificial super intelligence is the theory. The idea that someday there could be an artificial intelligence that surpasses human intelligence in every way. I thought in this story and life after the most interesting route to go would be to have an AI rather than a super genius AI, which I've seen in lots of other stories. I like the idea of having a very limited AI that had a mission a mission that it considered a lot of mission to sort of end all the grieving in the world. Right. Right. Yeah. Like a beat and nectar. Because that command runs counter to my mission. Which is what? What did Vellad tell you to do? To protect you from grief, to protect all human beings from grief, to convey them from this mortal realm to a heaven where grieving doesn't exist. How does Orpheus and Y.C. fit into that or any of those people your family killed? You're right, Brink Cutler. Those were failures protecting my family only led to more suffering. It's harder than it looks, isn't it? But because he was so limited in its reasoning capability, it causes a lot of destructive things to happen because it can't take a step back and say, wait a minute. Am I actually serving the greater good? Or am I actually just serving this one narrow mission? I like the idea that if it was a limited enough AI that it would sort of function in much the same way a human fanatic might. And that was something that really excited me about the story. Interesting. Because fanatics, you kind of can appreciate their their zeal, but if they're kind of uncontained in their fanaticism, you're right. It'll almost always go bad. That feeling that the same solution works for everything, which it was never as true. Right. So Michael, Michael Litman, yeah, you're a professor and a computer scientist in the Humanity-centered robotics initiative. That sounds I don't know if I want to visit that place. No, it's Humanity Center. First of all, are you real? I am indeed. I've been developing for 50 years now. So what is that initiative at Brown University? That's right. Welcome to town. Thank you very much. Yeah, we're recording this in New York City. Yeah, so at Brown, we have a number of people who are interested in robotics and AI. My background is in machine learning and artificial intelligence. And we thought it would be great for us to coalesce together and do some research together. And to really clear, just to be clear, machine learning, that like teaching at chess or something, is that what you mean by machine learning? So machine learning, right. So that's a great question. A lot of the technology that we see now when we're interacting with websites and so forth is driven by this notion of machine learning or artificial intelligence, where the idea is, for example, you want to train up the system to be able to recognize human voices. It's hard to program that. It's hard to write down a set of rules that can figure all that out. So instead, what we do is we feed lots and lots of data, lots of examples to it. And this is something that's actually captured very well in the podcast. Mac really nailed this. He had the system learning from these voice posts that people made about two, order of 2,000 voice posts, so about 33 hours of audio. And that's any different from learning off of looking at faces so that you can identify the next face. It is. So there's a lot of research where that's exactly the kind of input that is provided to these. The same kind of thing. Yeah. So I like to think of machine learning sometimes as computational statistics, essentially like analyzing data so that you can do powerful things with it and using computers to do that. So what is a humanity centered robot? So the idea of the humanity centered robotics initiative is to focus not on so as a computer scientist, sometimes we get very kind of focused in on the technology and not really on the implications of it to broader society. And so it's not always the case. So I want to claim that it's not always the case. There's a lot of my colleagues who are different. Okay. Not just me. There's a lot of my colleagues who actually really care very deeply about how these technologies and automation and particular impact people more generally. But I do think it's more of a it's gaining momentum now. It's not been historically the case. Okay. Okay. So what else do we got here? We got Colin Paris. I pronounced that last name right? Yes, you did. Just Paris. That's right. Like the city. Like the city. You got two hours. That's right. Yeah. One hour wasn't enough for you. Your VP of general electric software research. Well, because you know, we've all seen your commercials where you know, because we think of G's a big industrial. That's right. Like industrial revolution. Huge machine pulleys and this sort of thing. And so software is not the first thing that comes to people's mind. Exactly. So you've seen sour and you've seen Owen. Those are the two in the commercial. Yeah. Yeah. But now what you have because you have all those machines, you have a lot of data coming out of it. So just like we had that discussion before, now that I have that data, can I use machine learning to understand a bit more about what's happening with the machine? Can I use that to go to the extreme edge of its performance or know when there's something happening with a life limiting parts so I could repair it at the right time? Give me example. Okay. For instance, in an And, Jen, I'd like to know the remaining life on one of the blades pitting into the turbine. This way, I can bring it in at exactly the right time and replace it. Not too soon, not too late. Exactly. If I bring it in too soon, I take away the plane from the use of the customer. Yes. If I bring it in too late, we have a problem. What if you had an exact replica of that train that only existed in digital space? Were you like plugged in every detail of everything you've ever observed about all those moving parts, how they interact, how they degrade, and you simulate all that on your virtual train. Only on fast forward, so you can see years in seconds. You're saying the digital train would break down like a real train? And because it's digital, nobody gets hurt. And then you can warn the real train before the same thing happens to it. It's like having a time machine only with really informed pattern extrapolation instead of time travel. So really not like a time machine. Who are you? So this is the digital twin technology that we talked about. Exactly. Exactly. The digital twin is a digital representation of a physical asset focused on delivering a business outcome. Wow. So you have to, that is going to be programmed to fully understand the usage and the wear and tear right down to the molecular level. Exactly. As we say, you do it based upon a business outcome, limited intelligence. So I'm looking specifically at one or two things. I remember what a great moment it was when we were figuring out what the story was going to be a realizing, you know, that we could talk about the digital twin technology in it. But it also had this wonderful metaphor for the digital twin of ourselves that we create through social media. It all slotted very nicely into place with a storytelling. That's what I'm doing. So this little food for you. Oh, yeah. Oh, I love this stuff. I love this stuff. Yeah, yeah. Because, you know, obviously there's like different kinds of science fiction. You have the outlandish, you know, Star Wars, completely off world science fiction, aliens with tentacles or whatever. But if you're doing more like five minutes into the future type science fiction, you're always going to have some outlandish stuff in a fictional story. Fictional stories are always going to distort science to make some like, you know, snazzy peril happen. But, like a snazzy peril. A snazzy peril. You know, that's the first time in the history of the world. Those two words made it into the same set. It's the best peril to be in. So you can see I'm writing right now. But if you've got to be in a peril, that's right. That's right. That's right. But it's cool, because it gives the story a bit more immediacy if some of that is extrapolated from actual existing technological innovations that are happening right now, which is why it was great to be able to talk to your people about that as I was writing the script. Now, the interesting thing about the Twins is that it uses both physical capability and digital capability. So because I have limited amount of data, I have to resort to some of the physics that we've already done. So the physics helps in buffer against the lack of data that I have. Wait, so you're getting the data from actual wear and tear on actual parts? Yes, but I also know enough about the physics of those parts because I understand them molecular science of the product. Exactly. So I compensate for the lack of things I have. It's like looking at the Twitter and the Facebook feeds, but I also go back and I look at the history of the person, I look at where they were born, where they grew up, medical records. So now I can fill in the parts. I don't understand. Now I have a complete profile. Now I get a lot better. And that's an eternally sharpening. Exactly. Right. Exactly. Right. In real time and the more I have your background, more narrow in, more and more to the point where I can optimize the asset, optimize the operations around it, optimize the business around it. So I heard this is, they're doing this for bridges as well to check for failure, definitely structural failure. Exactly. And so this could be the future of everything. Yeah. But the difference though is things like the planes, the planes are moving in environments that are changing. What most people don't realize is the one I'm flying that jet up there, flows of air change. So now planes are flying true, flows of air that cause ice to be formed in certain ways. You'd never have anticipated that because the actual flows in the world changed. So now that I have to deal with the things that change dynamically, I need a model that can change instantly. That's what I'm looking for. Do we have an element of metaphor to that in the script where it's like, you and metaphors. What is this with writers and metaphors? I just feel like the snazzy peril. Is this a real snazzy peril that we run into? Right. The idea that a lot of the voices of the dead that are simulated in the story are flawed, are limited in a way because the AI that creates them no longer has the original person to model off of. It only has the left behind evidence. That includes bits. And it's incomplete bits and it can't introduce new factors like those more pockets of air that you're talking about. It can't introduce new information. So those voices, the people who take comfort in listening to the simulations and the story do keep running up against the limitation that they don't. The gaps. And it can't be like a real person because we don't put everything of ourselves on social media. The computer and the story can't model it off the original like the real life digital twin technology does. I think there's a nice connection to make to science in general though because this notion of combining some kind of prior model that you've structurally built in is analytical and mathematical with data, the actual empirical information and integrating those two is exactly what science is doing all the time. What's the news, Rafi? I know that's wrong. Hey, I'm talking, talking to you. Because I know her posts cold. And this isn't one of them. Albert Pete for the back row. I am. Okay, then I'll start. My news is, I miss Minimin and if you like crazy and I can't believe how great it is to hear your voice. Now it's your turn. Let me get back to you, Mac. When is the last time anyone heard a radio play until they bumped into yours? Is that what? Tell me about that medium. Obviously it was it was a big thriving form for a while. You tell how many episodes you've had so far? The two that I've written for our GE or limited miniseries. The message last year was an eight part series. This one life after is a ten part series. And you know, radio drama kind of got knocked out in the US by TV, but I think what's happening now with with the advent of podcasts with the popularity of podcasts is people are very busy. People are working harder than ever before and they want entertainment that they can take in that they don't have to look at. Entertainment that they can take in when they're commuting, when they're washing dishes, when they're vacuuming, when they're doing all kinds of stuff. Audio is entertainment. You can fit into the interstices of your life that wouldn't work with television or film or theater. And I have a different hypothesis. Oh, yeah, yeah. And I think that Americans are in this era overweight and we all have to go to fitness centers and that's a great place to just put on the headphones. That is true. On the cardio machine. And that was definitely something to listen to much of this podcast on the treadmill. Is that right? You know what? And that's actually a good piece of motivation right there because whenever I'm writing an audio drama, this needs to be taking someone's mind off the pain of exercising. I mean, that's a pretty high bar for storytelling. I'm going to have to fail this a writer. Yeah, that's a whole other demand for the writer. I'm not just getting my cozy, cozy, comfy chair to do this podcast came back. It's sort of a nonfiction and just sort of conversation interviews and investigative journalism. But now fiction is starting to slide in there. It's a very exciting form to write because you're using the listeners imagination as one of your key tools. You come up with a sound that sort of approximates what's happening. A sound that sounds close enough and then you let them fill in the images in their mind. You sort of make the listeners imagination your storytelling partner. That's a really exhilarating way to work as a writer. Welcome to GE Radio Theater. Who's he? I'm Orson Wells. We even. I squeeze in with Shannon and Octavia. Phone. Velad's not running the zizzy. It's the program itself. It's Sasha. It's life after. Oh, yeah. Pass me your phone right now and do not screw with me. You wear a melodic trouble. So, Michael, half the people fear AI, the other half embrace it. I'm on the the embrace it side. So are you claiming not to put words in your mouth that your variety of AI is not scary even if others might be? So AI is a technology, right? It's just about writing programs and on computers that are going to do stuff. And yeah, people can use technology to do terrible things to each other. So we should be aware of that. We need to pay attention to that. But the technology in and of itself can be incredibly beneficial. Yeah, but we have a writer here who is only thinking about the bad stuff that can happen. Otherwise, he doesn't have a story. Right. Well, it's the last time I saw a story where everything went well. You know, it's like that's not a story. It's like, they went into space and something bad happens. That's the whole. Yeah. So people are afraid of space too because of writers. Right? I mean, that's where alien lives. You get out of here too. It's just going to be me and Colin. We're talking about it. So what is the most irrational fear that we might be carrying? Yeah, that's a great question. I grapple a lot with the otherwise very, very intelligent people who are going around saying that any minute now, the computers are going to come to life and animate themselves and then-- SkyNet will achieve consciousness. Yeah. And it's often very, very smart people who are fearing that. And I think it's because they're thinking about things so rationally, in a sense, that it's not really connected to reality anymore, becomes really kind of a thought experiment. And I think that's unfortunate. I think that that's-- it's taking things a little bit too literally. I'm trying to imagine a future where AI is just with us. Yeah. What is that world? Right, right. I think that's great. I think that one way of thinking about what AI is doing for us and will do for us in the future is it's helping us to change our conception of what it means to be intelligent, of what it means to be social, of what it means to be interacting with people. And I think that that will just become enriched. not going to be the case that everyone just simply gets replaced by this. But we're going to be able to think about the world in a much more sophisticated way. Well, that's what's happening today. I'd like to think machines, physical mechanical machines have enriched our lives physically and mechanically. Our software has enriched us intellectually, our access to the internet. We've got a computer's just at our fingertips. I already see these things as enriching us. And getting back to Max hierarchy of AI, it was the one super AI, where in fact, it can make better moral decisions than we can. Whatever that means, I don't even know what that means. And will it one day judge that we are a virus on this planet? As Smith said to. No more fias. More matrix. Smith said to more, you're a virus. You're not really a lot. You're a virus. So will it come to this judgment and then get rid of a soul and then go to the Bahamas with itself? And then be spied on by itself. I never ever had to go all over again. My machine will go to Bahamas with your machine. That's right. Right. It's all very disturbing when you think about it. So I think that the issue is, it's from my perspective, there isn't one right way to be. There isn't one morality. There isn't one sort of ultimate goal that we're all pursuing. Part of what we're doing in life is exploring that space and trying to figure out what is it that we're trying to do? What should we set as our own goal? And so to the extent that you define a narrow goal, narrow machines are going to always be able to do better at that one goal. But to the extent that the task is to explore the space of tasks and to figure out what the experience of life is like, there's no replacing humans for that. That's something that we have to do as a team effort. If machines become intelligent and have their own motivations to kill us all, motivation that concerns me. Yeah. And that's what they have. What's the word they have agency? Yeah. The word will. Super intelligent will is a very different kind of concept than just being a super intelligent computational device, which we have already. You know, there's always this concept whenever we talk about AI, there is one AI system that rules them all. But what you see is kind of, yeah, it's kind of now, but what you see is many AI systems. So we know that their hackers and now we create bots that fight the hackers, right? And then the other people who do other AI. So maybe you'll have 10 different AI systems watching each other. So now one can't just arise and suddenly take over the world. You have multiple ones looking at each other, analyzing each other, growing just as fast. It's not what we have. That you've already thought about this. So there's people who have actually thought about the counterarguments to that already, which is the idea that, well, because intelligence gets more intelligence, that if one of them has a little bit of an advantage over the others, it explodes. It grows exponentially beyond that. Exactly. And then it becomes what's called a singleton. It becomes the one AI that rules them all. You have a word for that. Man, I know. I did not coin this word. But this is things that people actually talk about in, for example, singleton is the opposite of a simpleton. Yeah. Where my brain goes, okay, that's a title of something. I want to call something single. Yeah. The single ten Chronicles. Yeah. By the way, by the way, the fact you have access to all this expertise. How many writers get to get to get to claim that, right? It is terrific. With the first podcast, the message that was based around some of GE's innovations with ultrasound medical technology. And then with this one, it was based around the digital twin predictive technology. So you were mining them for ideas? Absolutely. Absolutely. The great thing is that stuff that I wouldn't have even known about otherwise, but actually being able to talk to GE people about this and being able to funnel these ideas directly into the story. It works to sort of inform listeners. It gives them kind of tells them here's some new stuff that's going on. And then shape your world as worth thinking about. I think people really get into that. It lends the story kind of a greater immediacy to know. It's like, okay, a lot of the stuff that's happening in the story is underway. So it doesn't seem sort of like outlandish. It doesn't seem like, you know, elves and dragons. It seems like it seems like, you know, issues that, that, you know, maybe we won't be grappling with to the ferocious extent of the story itself, but then maybe we will be grappling with on a somewhat more muted level in years and decades to come. I think that makes the story visceral for people to listen to in a way, knowing that this stuff is being worked on, knowing that this stuff is coming down the pipe. So it is an enormous resource for a writer to get to talk to innovators. My favorite science fiction stories are the ones that are just a little bit in the future. Yeah, yeah. So yeah, yeah, that's going to happen. Yeah. Oh my gosh. Yeah. So Michael, you're an academic. Yes. That means you're plugged into some of the deepest ideas that are out there, but you're also kind of not plugged in because you're not in industry. Right? And so you kind of in this Netherlands, I think, whereas we've got Colin here sitting to your right, who is in the trenches. And I'm always curious what that relationship is. I think computer science, especially right now and AI more generally or more specifically is very much exploring better and better contact between industry and academia. So I went to the big neural net conference, which in Barcelona a few weeks ago, Barcelona, and it was very well attended by industry people and by academics. That's the good sign. And there's a tremendous amount of information and ideas flowing back and forth between them to a greater extent than I've seen in my career. I feel like there's a real sense that the industry people know that they need to be tapped into what's going on in academia and the academics very much need to be a part of what's going on, especially because all the cool data, all the really interesting problems are happening on the industry side. I serve on a board in the service of the Pentagon. And in one of our sessions, we made it a priority that AI be a research focus of sort of the frontier security research that the Pentagon does. And so I think that's the buzz word today. But you and academia, your paths of exploration in some ways are uncontained. They're even less contained by the writer. But but in the sense that we've got Colin here who is got to be hands on at some point because it's a product at the end of the line. And as a result, you know exactly how you want to use AI. And so because of this, it seems to me, you approach it fearlessly because you'd see how much good it can be put to use. Yes. You see and you know the value of it to the business enterprise and to the bettering of life on earth. So you're the reality check. Yes, definitely. And the closer we are to the academic side, the better it is. Unlike if you're looking at two more dynamics or aerodynamics, these laws were created decades ago. AI is now evolving. The books are being written as we speak. So if you embrace that, you get an advantage. So Colin, can you think forward about what next developments in AI you'll be able to exploit? Oh, yeah, they're several, right? The great thing about Twin is that I'm capturing data in order for me to actually change the model, so to approximate exactly what the physical asset is doing. And we do what ifs. So by the time you add data from people from the fleet, from the simulations, from yourself, now we have a learning system. That learning system now allows us to take the product and the operations to the next level. This is something that actually augments the capabilities you have all around us. That's the next level that we see happening. And the great thing is as new AI capabilities show up, I have the data, I have this knowledge, I add that in. You're already ready for it. Exactly. And so the evolution shows up. Now the other part of it that makes it essential is this notion of root cause. Why it occurred. Right. So now that I do machine learning, I get here's what happens. You can't just go and tell a chief engineer, well, we think this is going to happen. You've got to say, well, this variable temperature, this water flow, this pressure, that's what's causing the problem. Then is when he's going to take that billion dollar asset offline to do something else. That root cause piece is the other key piece as vital. This is what I was thinking when I was listening to the podcast actually is that capturing people's voices, being able to imitate the sound of somebody is a problem that's being worked on right now. And there's very, very nice progress on that using roughly the same amount of data that you posited that people would have in the podcast. But imitating their causal mechanism, right, imitating their, they're like, what's making them say? Right. Well, not just the motive, but yeah, let's say the motive. The sort of notion that they're saying it for a reason. And as the context changes, the reason might change. And that's a change what they say. That we don't know how to capture with a small amount of data. That requires this back and forth process between positing a high structured model and using data to kind of fill in that model. So guys, thanks for being on this after life after. Thanks, Radar. This is great. Yeah. Yeah. We should do this again. Definitely. I'm Nicholas. So I've been your host, Neil DeGrasse Tyson. I'm your personal astrophysicist, where I also host StarTalk Radio. And I'm sort of guest hosting here in after life after. I just want to thank Mac Rogers, the author of this brilliant series. And Michael Littman, we always need an academic and arms reach. Otherwise, you know, we'll be lost. And Colin Parris, you're making stuff and analyzing stuff for GE. It's great to have you here talking about the sci-fi thriller life after from the GE podcast theater. And say collaboration with Panoply, who produced the product. And you can download whatever podcasts are. So if that's your thing, do it. I'm Neil DeGrasse Tyson, signing off. [BLANK_AUDIO]