WEBVTT - How AI Could Actually Make the World Better 0:00:15.356 --> 0:00:31.836 Pushkin. I'm Jacob Goldstein. This is what's your problem? And 0:00:31.876 --> 0:00:35.276 my guest today is Josh Tieringel. He's a staff writer 0:00:35.356 --> 0:00:37.716 at The Atlantic, and he just wrote a new book 0:00:37.756 --> 0:00:41.236 called AI for Good And the subtitle of the book 0:00:41.316 --> 0:00:44.796 explains the problem that Josh is interested in. The subtitle 0:00:44.916 --> 0:00:48.716 is how real people are using artificial intelligence to fix 0:00:48.796 --> 0:00:51.956 things that matter. As you'll hear, I talked with Josh 0:00:51.996 --> 0:00:55.956 about several of these real people. Professor trying to understand 0:00:55.996 --> 0:00:59.436 her nonverbal son, a general who ran the logistics side 0:00:59.436 --> 0:01:03.156 of Operation Warp Speed, a hospital CEO trying to reduce 0:01:03.196 --> 0:01:06.196 the rate of deadly infections. And one thing just to 0:01:06.236 --> 0:01:08.356 note here is you know, and I guess this is 0:01:08.356 --> 0:01:11.076 not surprising in the state of AI. Some of these 0:01:11.116 --> 0:01:13.356 stories are not yet finished. They're kind of stories that 0:01:13.396 --> 0:01:16.516 are still in the middle. But before we got into 0:01:16.516 --> 0:01:20.276 the specific stories, I asked Josh what common threads he 0:01:20.356 --> 0:01:23.036 saw across all these people he wrote about, who are 0:01:23.076 --> 0:01:25.756 you trying to use AI to do good things? 0:01:26.436 --> 0:01:30.876 They're largely stubborn. Uh huh, They're really stubborn. And you know, 0:01:30.916 --> 0:01:33.556 one of the things that I admire. I'm very impatient 0:01:33.596 --> 0:01:36.356 and when I see a problem. Oftentimes I have what 0:01:36.436 --> 0:01:38.916 I think is a fairly normal human impulse, which is like, 0:01:39.276 --> 0:01:42.236 wipe the mismatched puzzle pieces off the table and get 0:01:42.236 --> 0:01:46.036 a different puzzle. Right, It's frustrating to dig into systems 0:01:46.436 --> 0:01:50.916 to learn technology, and these people, they don't get frustrated, 0:01:51.276 --> 0:01:53.796 not to the point of destroying the system. In fact, 0:01:54.716 --> 0:01:57.316 most of them have no experience with AI or technology. 0:01:57.716 --> 0:02:00.636 They care about medicine, they care about teaching, They care 0:02:00.676 --> 0:02:03.476 about solving a really hard problem, and so they just 0:02:03.556 --> 0:02:07.276 keep looking for solutions. And when something seems promising, their 0:02:07.316 --> 0:02:10.476 ability to concentrate and learn, earn the technology of that 0:02:10.636 --> 0:02:14.636 promise and apply it to their problem is through the roof. 0:02:15.196 --> 0:02:18.276 And so I just kept being awed by the fact 0:02:18.276 --> 0:02:22.916 that they were curious, they're patient, They were impatient at 0:02:22.956 --> 0:02:26.356 exactly the right moments in the sense that things weren't moving. 0:02:26.716 --> 0:02:29.876 They knew how to deploy either their political capital or 0:02:29.876 --> 0:02:34.036 their temper in ways to get things moving. And you know, 0:02:34.076 --> 0:02:35.996 I found time and time again that when I would 0:02:36.036 --> 0:02:38.796 talk to other people who were witnessing these phenomenon, whether 0:02:38.836 --> 0:02:41.716 it was in hospitals or in government, we would all 0:02:41.756 --> 0:02:43.996 just kind of be like, Yeah, these are the ones 0:02:44.036 --> 0:02:46.996 making it happen. They're kind of Samurai, right. They care 0:02:47.076 --> 0:02:50.396 about their problem more than other people care about standing 0:02:50.396 --> 0:02:52.916 in the way of the solution, and they get it done. 0:02:53.156 --> 0:02:54.476 Tell me about Christy Johnson. 0:02:55.076 --> 0:03:00.316 Yeah, So, Christy Johnson is certainly one of the most 0:03:00.396 --> 0:03:05.796 inspiring interesting people I've met in my journalism career. So, I, 0:03:05.876 --> 0:03:08.436 like a lot of people, was thinking about AI and 0:03:08.516 --> 0:03:11.796 human connect and reading all the headlines about the sort 0:03:11.796 --> 0:03:16.596 of parasocial crisis in which people were dating, chat GPT, 0:03:16.916 --> 0:03:20.276 falling in love with Claude, realizing that this was mimicking 0:03:20.356 --> 0:03:23.076 a lot of the worst elements of social media, seeing 0:03:23.116 --> 0:03:26.916 the studies and getting very, very depressed about what AI 0:03:27.036 --> 0:03:29.596 would mean for human connection. So I was having that 0:03:29.676 --> 0:03:33.036 conversation with Rosin Picard, who runs the MIT Media Lab, 0:03:33.076 --> 0:03:34.876 and she basically said, look, you got to meet Christy, 0:03:35.676 --> 0:03:39.396 and so she introduced me. Christy Johnson was a physicist 0:03:40.276 --> 0:03:47.076 and a brilliant student. She met another physicist in a program. 0:03:47.516 --> 0:03:49.996 He is an astrophysicist and actually was on the team 0:03:50.636 --> 0:03:53.076 that took the first picture of a black hole. The 0:03:53.116 --> 0:03:55.236 two of them fell in the deepest form of nerd 0:03:55.276 --> 0:03:59.036 love you could possibly imagine. I mean even the two 0:03:59.036 --> 0:04:01.196 of them laugh at just how dorky their love letters 0:04:01.236 --> 0:04:03.996 are to each other, all about physics and math and 0:04:04.516 --> 0:04:07.956 the world. They get married and they have a kid, 0:04:08.036 --> 0:04:12.116 and about a couple of weeks in she's like, there's 0:04:12.116 --> 0:04:14.596 something wrong with my son. And she's kind of gas 0:04:14.716 --> 0:04:17.276 lit by the medical profession for a year just hearing no, 0:04:17.356 --> 0:04:21.636 he's fine, Stop worrying, enjoy your kid. Finally they get 0:04:21.636 --> 0:04:25.476 some testing done and he has a genetic deficiency. And 0:04:26.476 --> 0:04:29.876 the other six children like him in the world are 0:04:29.916 --> 0:04:35.556 all severely autistic, epileptic, and completely nonverbal, and so this 0:04:35.596 --> 0:04:39.876 is the kind of diagnosis that changes everybody's lives. Within 0:04:39.916 --> 0:04:44.636 the family. They begin to grieve a little bit for 0:04:44.676 --> 0:04:47.636 their expectations for their son, but they're also scientists, and 0:04:47.996 --> 0:04:51.396 Christy and Michael both are like, we should understand this, 0:04:51.516 --> 0:04:54.516 we should do something about it. A couple of years 0:04:54.516 --> 0:04:57.676 ago past and Christy is like, no longer all that 0:04:57.716 --> 0:05:01.036 interested in physics. She's very interested in her son. And 0:05:01.076 --> 0:05:03.396 so what she set out to do was solve the 0:05:03.436 --> 0:05:08.236 problem of the more than one million nonverbal Americans who 0:05:08.556 --> 0:05:11.436 are stuck to their parents and their caregivers. That's it 0:05:11.756 --> 0:05:15.556 for everybody in that universe, the universe shrinks. 0:05:16.036 --> 0:05:18.596 Just to be clear, what is her dream? What is 0:05:18.596 --> 0:05:20.156 the outcome she hopes to achieve. 0:05:20.636 --> 0:05:23.156 Her dream is that her son and other people like 0:05:23.236 --> 0:05:25.836 him can interact with people they don't know all that 0:05:25.876 --> 0:05:29.636 well and share what it is they want and need. 0:05:29.796 --> 0:05:35.236 So a machine will turn their nonverbal vocalizations into words. Yeah, 0:05:35.396 --> 0:05:38.316 verbal people can understand that other people can respond to 0:05:38.476 --> 0:05:40.876 and interact with. And so she made a bet. She 0:05:40.956 --> 0:05:42.996 basically thought, look, I see where AI. 0:05:42.836 --> 0:05:45.756 And machine learning are right now. My hunch is that 0:05:45.796 --> 0:05:48.276 they will get better and better. But what I need 0:05:48.476 --> 0:05:52.476 is data. And so she created a scientific protocol to 0:05:52.596 --> 0:05:57.316 get vocalizations from these kids, and she used every tool available. 0:05:57.396 --> 0:05:59.676 So she went on YouTube, she went on social media. 0:05:59.796 --> 0:06:01.956 She said, this is me and my son. This is 0:06:01.956 --> 0:06:04.956 what happens when he's hungry. Listen to that vocalization, and 0:06:04.956 --> 0:06:08.836 now I'm recording it, and she began to generate a 0:06:08.876 --> 0:06:10.956 little bit of momentum. She went to the MIT Media 0:06:10.996 --> 0:06:13.716 Lab and spoke to Rosbaccard one night, kind of cold 0:06:13.796 --> 0:06:18.596 calder and Roz realized, like, this woman has no qualifications 0:06:18.636 --> 0:06:21.476 to be at the media lab, but my god, like, 0:06:21.876 --> 0:06:26.276 what an incredible human she got there, she did her PhD. 0:06:27.076 --> 0:06:29.516 And she proved that if you can gather enough of 0:06:29.556 --> 0:06:33.236 these vocalizations as audio files, you can begin to attack 0:06:33.276 --> 0:06:36.676 the problem. And so what she learned largely through using AIS. 0:06:36.716 --> 0:06:41.716 At first you have to multiply the files, so a 0:06:41.956 --> 0:06:47.116 system like chat, GPT or Claude is trained on potentially 0:06:47.156 --> 0:06:50.156 trillions of data points in order to become as fluent 0:06:50.196 --> 0:06:53.996 as it is. She had a couple thousand audio files. 0:06:54.316 --> 0:06:56.996 So the first thing she did is get her graduate 0:06:57.036 --> 0:07:00.236 students to figure out how to synthesize more data, how 0:07:00.236 --> 0:07:04.516 to take those data files, those original sounds, and grow 0:07:04.556 --> 0:07:05.916 them and multiply them. 0:07:05.996 --> 0:07:08.316 It's like what they're trying to do with self driving cars, 0:07:08.436 --> 0:07:12.196 right exactly, like drive on pretend roads essentially exactly. 0:07:12.316 --> 0:07:15.396 It's called synthetic data. And AI is really good at 0:07:15.396 --> 0:07:16.316 making synthetic data. 0:07:16.316 --> 0:07:18.316 And they're even they're working on that for text, right 0:07:18.356 --> 0:07:20.996 because now the current models basically are trained on the 0:07:20.996 --> 0:07:23.596 whole Internet, so like there's no more real text for 0:07:23.636 --> 0:07:25.636 them to read, so they need synthetic texts. 0:07:25.836 --> 0:07:28.116 They're out of stuff, right, And so if you want 0:07:28.116 --> 0:07:30.196 the training to advance, you have to synthesize a lot 0:07:30.236 --> 0:07:32.756 of this data, and she's continuing to grow the organic 0:07:32.756 --> 0:07:33.556 files as well. 0:07:34.196 --> 0:07:35.596 Tell me about ROSCO. 0:07:36.836 --> 0:07:42.716 Yeah, So ROSCO is a protocol and basically, you can't 0:07:42.996 --> 0:07:46.476 do what Christy is trying to do without science because 0:07:46.996 --> 0:07:49.396 you need some sort of rigorous idea of what you're 0:07:49.436 --> 0:07:53.676 testing for or else. The data that you're recording is meaningless. 0:07:53.676 --> 0:07:54.636 It's not really data. 0:07:55.036 --> 0:07:58.036 It's not really data, it just sounds. And so working 0:07:58.036 --> 0:08:00.436 with a grad student, what she figured out is, Okay, 0:08:00.516 --> 0:08:04.876 we don't need all the data. All the data is unprocessable. 0:08:05.316 --> 0:08:07.596 We need the right kinds of data for this task. 0:08:07.676 --> 0:08:10.596 And so what they did is create this protocol where 0:08:10.596 --> 0:08:13.956 they would run families through a certain kind of test 0:08:14.076 --> 0:08:16.476 and it was split timed so that it was standardized, 0:08:16.516 --> 0:08:20.996 and they would say, for two minutes, give your child 0:08:21.316 --> 0:08:24.076 a YouTube video because we want to hear. And you know, 0:08:24.116 --> 0:08:26.476 in the autism community, YouTube is great. It cues up 0:08:26.516 --> 0:08:29.156 the same video as many times as you want. People 0:08:29.196 --> 0:08:31.756 always have their favorite video, she said, whatever their favorite 0:08:31.796 --> 0:08:34.916 video is queued up. At the one forty five mark, 0:08:34.916 --> 0:08:39.076 they would introduce a beach ball and buffering. 0:08:38.996 --> 0:08:41.636 Beach ball meaning the spinning wheel that says like pause, 0:08:41.676 --> 0:08:42.396 it's loading. 0:08:42.596 --> 0:08:45.316 Yeah, they would pause the video because what they wanted 0:08:45.356 --> 0:08:50.076 to do was get a vocalization of dissatisfaction and frustration, 0:08:50.316 --> 0:08:52.036 and so that was a way to do it. And 0:08:52.116 --> 0:08:55.156 they would say, okay, this is a five minute snack period. 0:08:55.516 --> 0:08:57.156 We want you to make sure that you are giving 0:08:57.196 --> 0:09:00.196 them their favorite sack, and we want to hear what 0:09:00.196 --> 0:09:02.476 that is like. And so they would do these things 0:09:02.516 --> 0:09:05.156 that are really relevant to her ultimate goal, which is 0:09:05.156 --> 0:09:08.076 making sure that these kids could be out there in 0:09:08.116 --> 0:09:10.796 the world talking to other people and having needs met. 0:09:11.596 --> 0:09:14.916 And so it was a very scientific protocol. And then 0:09:14.956 --> 0:09:18.356 she would send it to families along with a phone 0:09:19.316 --> 0:09:22.156 and a camera and basically say, record it from these 0:09:22.236 --> 0:09:24.596 various angles, send it back to us, so that the 0:09:24.716 --> 0:09:28.276 data is standardized. It's not just random vocalizations that they're 0:09:28.276 --> 0:09:32.316 figuring out. And because the population is so specific and 0:09:32.316 --> 0:09:35.196 because the keepers of the knowledge are so specific to 0:09:35.276 --> 0:09:38.396 parents and caregivers, it took her a while to figure out, 0:09:38.476 --> 0:09:40.076 that's how you actually want to get the data. 0:09:40.636 --> 0:09:44.916 Huh. Yeah, it sounds like a lot of work. It 0:09:44.956 --> 0:09:47.476 sounds hard to scale. 0:09:47.636 --> 0:09:50.036 Yeah, And it would be hard to scale if not 0:09:50.076 --> 0:09:52.716 for synthetic data, and if not for all of the 0:09:52.756 --> 0:09:54.556 advances on the other side with translation. 0:09:54.756 --> 0:09:57.476 Yeah, so you don't have to scale that much, is 0:09:57.516 --> 0:09:58.276 what those things. 0:09:58.196 --> 0:10:00.076 Well, you don't have to scale it nearly as much, right, 0:10:00.116 --> 0:10:02.676 And what would have been a project that would probably 0:10:02.756 --> 0:10:07.476 have happened over three scientists lifetimes yea, is now maybe 0:10:07.596 --> 0:10:10.756 only over half of her career. And that's a real 0:10:10.876 --> 0:10:13.756 form of progress. And so even though Christy doesn't have 0:10:13.836 --> 0:10:16.916 the kinds of funding that Google or open Ai have, 0:10:17.516 --> 0:10:19.716 she's able to kind of surf in their wake, and 0:10:20.116 --> 0:10:23.476 they've made these progressions on things like zero shot translation, 0:10:23.596 --> 0:10:26.076 with some people may have heard of. You know, historically 0:10:26.116 --> 0:10:29.476 we would translate things by taking one set of language, 0:10:30.516 --> 0:10:33.836 decoding each word and rewriting it in a different language, 0:10:34.036 --> 0:10:37.276 and up until about ten years ago, that's what machine 0:10:37.276 --> 0:10:41.396 translation was. Zero shot translation is actually a translation at 0:10:41.436 --> 0:10:45.796 such a deep level that you're not translating language to language, 0:10:45.836 --> 0:10:50.796 you're basically translating every language to every language simultaneously. And 0:10:50.836 --> 0:10:52.916 then we got to a certain point recently where we 0:10:52.956 --> 0:10:56.356 can translate based on audio waves. And so if you 0:10:56.356 --> 0:11:01.756 can translate audio waves, it becomes much easier to translate 0:11:01.796 --> 0:11:05.756 vocalizations from people who don't use words at all, and 0:11:05.796 --> 0:11:08.556 so she you know, you mentioned a lot of these 0:11:08.596 --> 0:11:11.476 stories aren't done. This is the least done story, right, 0:11:11.756 --> 0:11:16.716 And you know what's interesting is that technology has already 0:11:16.796 --> 0:11:19.996 changed the lives of lots of autistic people, because you know, 0:11:20.076 --> 0:11:22.956 she showed me her phone. 0:11:23.436 --> 0:11:23.636 Right. 0:11:23.716 --> 0:11:26.996 So we were having dinner on our porch, and her son, Felix, 0:11:27.036 --> 0:11:31.916 who's lovely sweet in many ways, a normal American teenager, 0:11:31.996 --> 0:11:34.236 just picked up the phone and went off to go 0:11:34.276 --> 0:11:36.276 do something. And then he came back and he dropped 0:11:36.276 --> 0:11:38.876 off the phone, and she and her husband showed me 0:11:38.956 --> 0:11:42.636 photos that he had taken of what he's fascinated by. 0:11:43.316 --> 0:11:45.916 And so some of them were hairbrush bristles that were 0:11:45.916 --> 0:11:49.636 actually these these kind of beautiful plastic forests. He'd used 0:11:49.636 --> 0:11:52.996 a filter and he'd taken them and hundreds of these photos. 0:11:53.436 --> 0:11:56.396 And before that, he'd taken photos of windows and was 0:11:56.476 --> 0:11:59.196 really interested in the way windows around their neighborhood worked, 0:11:59.796 --> 0:12:02.836 or castles or sketches of castles. And she said, like, 0:12:02.956 --> 0:12:07.356 before the iPhone, when he's staring at something, I would 0:12:07.396 --> 0:12:10.196 have no idea what he would have been daring. But 0:12:10.276 --> 0:12:13.356 here's this really rudimentary piece of technology, at least now 0:12:13.996 --> 0:12:16.996 and through that I'm able to understand what he's looking at. 0:12:17.156 --> 0:12:21.116 I don't necessarily understand why he's fascinated by it, but 0:12:21.276 --> 0:12:22.916 I know that he's interested. And I can make his 0:12:22.996 --> 0:12:25.836 life richer by exposing him to more of it, and 0:12:26.036 --> 0:12:28.196 I can frustrate him less by making sure there's always 0:12:28.236 --> 0:12:32.196 a hairbrush around. And so those kinds of things I 0:12:32.236 --> 0:12:35.516 have to say, like they make me more optimistic that 0:12:35.596 --> 0:12:38.996 if we can wrestle this technology, like you know, tech 0:12:39.036 --> 0:12:41.676 doesn't always have to be bad, Like we're not faded 0:12:41.756 --> 0:12:44.996 to the world sucking because technology is coming out, but 0:12:45.076 --> 0:12:46.716 we do have to figure out how we want it 0:12:46.756 --> 0:12:50.676 to work. And so I left Christy feeling like she's 0:12:50.716 --> 0:12:53.876 going to get there, and I know that that's not guaranteed, 0:12:54.436 --> 0:12:58.036 but she is so stubborn, she's so motivated by love 0:12:58.356 --> 0:13:01.196 and by problem solving, and the tech is moving so 0:13:01.396 --> 0:13:04.516 fast that she's been at this ten years, right, and 0:13:04.596 --> 0:13:07.236 ten years she's only gotten so far. But the next 0:13:07.236 --> 0:13:09.996 ten years are going to move so much faster, and 0:13:10.036 --> 0:13:11.916 more people are going to come and attack the problem. 0:13:11.956 --> 0:13:15.156 And it wouldn't surprise me if ten years from now 0:13:15.476 --> 0:13:16.956 she's successful. 0:13:18.236 --> 0:13:22.396 So another story you tell in your book is about 0:13:22.436 --> 0:13:26.556 Operation Warp Speed, and in particular it's about the logistics 0:13:26.596 --> 0:13:29.836 of getting vaccines out to the public. And you know, 0:13:29.956 --> 0:13:33.556 I knew a moderate amount about the vaccine development side, 0:13:33.916 --> 0:13:36.076 but I really didn't know anything about the getting the 0:13:36.156 --> 0:13:39.036 vaccine out to the world side. And I certainly didn't 0:13:39.076 --> 0:13:42.516 know that it was an AI story. So tell me 0:13:42.556 --> 0:13:44.956 about Operation Warp Speed, and in particular about this guy 0:13:44.996 --> 0:13:48.356 you write about, this general named Gus Purna. 0:13:49.516 --> 0:13:53.276 Look, I was writing, and I was hungry for understanding 0:13:53.516 --> 0:13:56.316 and trying to figure out how exactly machine learning and 0:13:56.436 --> 0:14:00.116 artificial intelligence integrate into society and ways are helpful, and 0:14:00.156 --> 0:14:03.796 so I kind of tripped across this video of Gus Perna, 0:14:03.956 --> 0:14:07.996 who oversaw Operation Warp Speed, who is not a technologist 0:14:08.076 --> 0:14:11.956 by any means, He's a logistician. He oversees, generally, you know, 0:14:12.516 --> 0:14:15.636 or did before he retired, the delivery of munitions and 0:14:15.716 --> 0:14:18.556 uniforms and food so that we can fight wars. And 0:14:18.596 --> 0:14:21.036 he was placed in charge of Operation Warp Speed and 0:14:21.076 --> 0:14:23.436 the delivery of it so that you know, hopefully Big 0:14:23.476 --> 0:14:27.916 Pharma can can execute on all these vaccines. But then 0:14:27.956 --> 0:14:30.556 what how do they get into plastic vials and how 0:14:30.556 --> 0:14:33.236 do those vials get delivered to everything from a CBS 0:14:33.756 --> 0:14:36.756 to a doctor's office to in rural areas, sometimes a 0:14:36.796 --> 0:14:41.716 general store that has no digital footprint. And so he 0:14:41.756 --> 0:14:46.236 shows up in Washington and discovers that there's no plan, 0:14:46.356 --> 0:14:50.556 like he is the plan. And so in Washington, like 0:14:50.596 --> 0:14:54.596 it or not, everything starts with consultants everything, and so 0:14:54.676 --> 0:14:56.876 people come in and they pitch him on things, and again, 0:14:57.036 --> 0:15:00.156 not a technical guy, but he keeps hearing these crazy 0:15:00.196 --> 0:15:03.036 pitches over the course of an hour. Somebody's like, we're 0:15:03.036 --> 0:15:06.076 going to put a medical blockchain together and in a 0:15:06.076 --> 0:15:09.476 couple of years, and he's like, no. Somebody else pitched 0:15:09.516 --> 0:15:12.036 like hardware, We'll just put a hardware device in every 0:15:12.076 --> 0:15:15.356 doctor's and he knows enough to say no. And finally 0:15:15.356 --> 0:15:18.756 he meets with these two reasonably seeing people who are like, look, 0:15:19.036 --> 0:15:21.556 you have a data problem. This can all be solved 0:15:21.676 --> 0:15:24.676 if we get the right number of data streams flowing 0:15:24.676 --> 0:15:27.596 into the right interface so that you can actually see 0:15:27.956 --> 0:15:33.196 everything from plastic vials, metal needles, refrigerated trucks because remember, 0:15:33.796 --> 0:15:37.636 half of the vaccines needed to be really cold refrigerated 0:15:37.676 --> 0:15:41.396 for delivery. To track which states might want Johnson and 0:15:41.476 --> 0:15:44.556 Johnson versus like you have to build a whole civilization 0:15:44.636 --> 0:15:46.036 to do this stuff. And they said, look, you can 0:15:46.036 --> 0:15:50.956 build a civilization in data, probably won't cost that much, 0:15:51.276 --> 0:15:54.436 but we will need you to solve the human problem, 0:15:54.756 --> 0:15:57.076 which is getting these various agencies to play ball, so 0:15:57.116 --> 0:15:59.356 we can actually get to the data, clean it up, 0:15:59.476 --> 0:16:01.596 and feed it to you. And so that's a machine 0:16:01.636 --> 0:16:04.476 learning story. And when I went down the rabbit hole 0:16:04.516 --> 0:16:07.276 of how it actually happened, it was really instructive because 0:16:07.356 --> 0:16:11.556 obviously the end state was a huge success, underrated. 0:16:11.916 --> 0:16:15.076 Every chance I get, I want to scream about how underrated, 0:16:15.076 --> 0:16:18.916 worse feed is weirdly politically orphaned, right, Like the Republicans 0:16:19.316 --> 0:16:21.876 decided they don't like vaccines even though they did it. 0:16:21.876 --> 0:16:24.076 It was like, as far as I'm concerned, the great 0:16:24.076 --> 0:16:27.116 success of Trump's first administration, and then because it was Trump, 0:16:27.156 --> 0:16:28.956 the Democrats can't say it was great. 0:16:28.756 --> 0:16:31.796 Right exactly, And so it's wandering out there, this great 0:16:31.836 --> 0:16:34.916 American success story that just got buried by the stupidity 0:16:34.916 --> 0:16:38.356 of our politics. But we did it, and we wouldn't 0:16:38.356 --> 0:16:41.236 have been able to do it without really successful machine 0:16:41.316 --> 0:16:41.996 learning and AI. 0:16:42.716 --> 0:16:45.116 So how did it work? Like why was that an 0:16:45.116 --> 0:16:46.556 AI problem? And what did they actually do. 0:16:46.916 --> 0:16:49.396 So it was a data problem, but also an actionable 0:16:49.476 --> 0:16:53.956 data problem. And so there's a company called Pounteer, very 0:16:54.116 --> 0:16:57.156 very provocative company. A lot of people respond poorly to 0:16:57.236 --> 0:17:01.516 its name, and it has this kind of mystical aura 0:17:01.556 --> 0:17:03.916 to it, in part because it was founded by Peter 0:17:03.996 --> 0:17:07.516 Chiel and co funded by the CIA, and in part 0:17:07.516 --> 0:17:11.556 because it's named for the stones in the Tolkien trilogy. 0:17:12.036 --> 0:17:13.956 But really all they do is plumbing. 0:17:14.356 --> 0:17:18.516 I am not reflexively anti tech or anti company. I 0:17:18.556 --> 0:17:21.676 have some amount of wariness toward Palenteer, I guess because 0:17:21.716 --> 0:17:26.156 I associate them with domestic surveillance at some level, like 0:17:26.196 --> 0:17:28.396 that piece of it I mean, but I don't know 0:17:28.436 --> 0:17:30.316 it that well. I've never interviewed somebody from Palenter. I 0:17:30.316 --> 0:17:32.716 haven't read a book about them. So, like, tell me 0:17:32.716 --> 0:17:36.596 about Palenteer as a company. I guess, first of all, 0:17:36.596 --> 0:17:39.916 like the side that makes people weary, and then I'm 0:17:39.916 --> 0:17:42.716 curious more generally how they operate, what differentiates them, what 0:17:42.756 --> 0:17:44.236 they actually do as a tech company. 0:17:44.556 --> 0:17:48.156 Yeah. So as a company, they're created in two thousand 0:17:48.156 --> 0:17:50.796 and three by two friends who have no business being friends. 0:17:50.796 --> 0:17:53.156 One is Peter Thiel, who I think most of your 0:17:53.196 --> 0:17:57.876 listeners really know about. He's a sort of largely libertarian Republican, 0:17:58.156 --> 0:18:02.516 big Trump donor, has very controversial stances on pretty much 0:18:02.556 --> 0:18:03.836 every possible. 0:18:03.436 --> 0:18:07.236 Issue, including the Antichrist, weirdly including the anti Christ. 0:18:07.316 --> 0:18:10.036 He really finds a way. His friend is a guy 0:18:10.116 --> 0:18:12.876 named Alex Karp, who is kind of his polar opposite. 0:18:12.916 --> 0:18:17.956 He's New York born, he's half black, half Jewish, largely 0:18:17.956 --> 0:18:21.716 identifies as a socialist. A big donor to Hillary Clinton 0:18:21.756 --> 0:18:25.876 and Kamala Harris. They met in law school. They argued 0:18:25.876 --> 0:18:28.836 all the time. They agreed about the need for this 0:18:28.996 --> 0:18:31.476 kind of technology, and the only other thing they agree 0:18:31.476 --> 0:18:34.436 on is that in general, the world is better when 0:18:34.476 --> 0:18:38.316 America is a strong actor and American values like freedom 0:18:38.596 --> 0:18:40.876 are or propagated throughout the world. So that's what they 0:18:40.916 --> 0:18:41.316 agree on. 0:18:41.636 --> 0:18:45.436 Do you think that the liberal antagonism toward Palenteer is 0:18:45.476 --> 0:18:47.476 misplaced or do you think it's valid in some ways? 0:18:48.436 --> 0:18:52.316 It's incredibly complicated because it has now grown to include 0:18:52.356 --> 0:18:55.756 not just surveillance, but Alex Karp's position on Israel, which 0:18:55.796 --> 0:18:58.636 is that Israel should be supported by Palanteer products, just 0:18:58.676 --> 0:19:02.036 as Ukraine should be. And so there's lots of different 0:19:02.276 --> 0:19:05.516 valances where people can find something they don't like about Palanteer. 0:19:05.756 --> 0:19:07.716 And by the way, I respect all of them. 0:19:08.036 --> 0:19:12.396 Yeah, and so just as a tech company, what is 0:19:12.436 --> 0:19:15.796 Palanteer's thing, What do they do? And in particular, what 0:19:15.876 --> 0:19:18.796 did they do in Operation Warp Speed to get vaccines 0:19:18.836 --> 0:19:19.556 out to the public. 0:19:20.076 --> 0:19:22.156 I spoke to their chief architect and he's like, look, 0:19:22.236 --> 0:19:25.436 we're the mole people of Silicon Valley. We take these 0:19:25.516 --> 0:19:28.956 data streams, we lay them out, we clean them, We 0:19:29.116 --> 0:19:32.756 integrate the data from various different places, so like, for instance, 0:19:32.756 --> 0:19:35.716 a CVS and a right aid and a mom and 0:19:35.756 --> 0:19:38.916 pop place and a trucking company, and we put them 0:19:38.916 --> 0:19:42.196 together so that on an iPad or a laptop someone 0:19:42.276 --> 0:19:44.396 could play the pandemic like a video game. 0:19:44.596 --> 0:19:47.236 Right, play it like a video game, meaning there's like 0:19:48.076 --> 0:19:50.956 a crazy dashboard, so you can see here's where the 0:19:50.956 --> 0:19:53.596 people are, and here's where the vials are, and here's 0:19:53.596 --> 0:19:55.436 where the vaccine is, and here's where the trucks are, 0:19:55.436 --> 0:19:57.356 and we need to get these trucks to this vaccine 0:19:57.396 --> 0:19:59.476 to these people. That's the exact video game you're playing. 0:19:59.716 --> 0:20:03.716 Yeah, and constantly up to date and constantly accurate, and 0:20:04.076 --> 0:20:06.676 literally an administrator can sit there and just move things 0:20:06.716 --> 0:20:08.916 around and know that they're actually going to the right place. 0:20:09.196 --> 0:20:12.756 Yes, a wildly hard problem. Like it's not sexy, like 0:20:12.956 --> 0:20:16.636 developing a vaccine, and we've heard a lot about mRNA vaccines, 0:20:16.636 --> 0:20:19.796 but like figuring out where the actual vaccine is is 0:20:20.676 --> 0:20:21.956 totally non trivial health. 0:20:22.156 --> 0:20:24.756 Yeah, it's a huge deal. And so there's two problems 0:20:24.796 --> 0:20:27.476 you're solving right. One is the technical problem, and that 0:20:27.556 --> 0:20:31.196 involves a lot of really dull, really unglamorous work, which 0:20:31.236 --> 0:20:34.436 is like trying to figure out why this pipe you 0:20:34.476 --> 0:20:36.956 know it's because it is pipe coming from a distribution 0:20:37.156 --> 0:20:40.156 center is broken, and so AI can attack that. 0:20:40.436 --> 0:20:42.316 When you say pipe, do you mean the pipe through 0:20:42.316 --> 0:20:44.276 which data is flowing? What do you mean pipe in 0:20:44.316 --> 0:20:44.956 this instance? 0:20:45.236 --> 0:20:49.156 So, to build a data pipeline is basically a work 0:20:49.196 --> 0:20:52.356 of code. But everybody's got lots of different code, and 0:20:52.436 --> 0:20:54.476 so when your data pipelines are built in many many 0:20:54.516 --> 0:20:58.876 different codes to many different standards, they have to be standardized. 0:20:59.036 --> 0:21:01.396 You have to be able to see, oh, okay, this 0:21:01.516 --> 0:21:04.476 data coming from CBS matches up to the data coming 0:21:04.516 --> 0:21:08.876 from the trucking company. It's clean, it's modern. Most of 0:21:08.916 --> 0:21:11.596 the that they were dealing with in fact was not 0:21:11.716 --> 0:21:15.156 modern in any way. It was created a decade ago 0:21:16.116 --> 0:21:20.036 using old code and often had bad information, and so 0:21:20.156 --> 0:21:22.476 you need to standardize it, you need to clean it. 0:21:22.796 --> 0:21:25.116 And then where it really gets fun is you can 0:21:25.276 --> 0:21:30.236 use those pieces of information to recommend actions to note 0:21:30.276 --> 0:21:33.556 that like, oh, you know you're overly distributing in one area, 0:21:33.636 --> 0:21:37.316 you could be moving here. In fact, this rural area, 0:21:37.476 --> 0:21:41.316 for instance, might really want the one shot doses, whereas 0:21:42.116 --> 0:21:43.556 an urban area might want too. 0:21:43.876 --> 0:21:47.876 And those suggestions might be coming from the machine learning model, 0:21:47.876 --> 0:21:48.916 from the AI basic. 0:21:48.836 --> 0:21:52.236 Correct, so that while you know, the very overtaxed humans 0:21:52.236 --> 0:21:56.196 dealing with political problems and interagency problems are on the phone, 0:21:56.756 --> 0:21:58.876