WEBVTT - Katie Plus One Presents AI For Dummies with Vivian Schiller, Vilas Dhar, and Chris Wiggins 0:00:00.200 --> 0:00:03.640 Cancer Straight Talk is a podcast for Memorial Sloan Kettering 0:00:03.760 --> 0:00:08.360 Cancer Center. We're host doctor Diane Reedy. Lagunis has intimate 0:00:08.440 --> 0:00:13.480 conversations with patients and experts about topics like dating and sex, 0:00:14.000 --> 0:00:18.319 exercise and diet, the power of gratitude, and more. I 0:00:18.480 --> 0:00:21.439 love being her guest back in April. Listen to Cancer 0:00:21.480 --> 0:00:29.760 Straight Talk. You'll learn so much. Hi everyone, I'm Kitty 0:00:29.840 --> 0:00:36.680 Kuric and this is next question. So I have to 0:00:36.760 --> 0:00:40.760 confess friends that if I sound weird, it's because I 0:00:40.800 --> 0:00:45.919 have a terrible cold. So I apologize in advance. Luckily, 0:00:46.120 --> 0:00:50.200 my plus one isn't sitting next to me catching my germs, 0:00:50.280 --> 0:00:54.800 but she is at a remote location. Where are you, Vivian. 0:00:55.000 --> 0:00:56.560 I'm in Bethesta, Maryland, in my home. 0:00:56.880 --> 0:01:01.600 Oh nice, Well, Vivian Schiller, as you probably heard, is 0:01:01.680 --> 0:01:05.160 my plus one today. And Vivian, I thought i'd start 0:01:05.200 --> 0:01:08.920 by telling folks how we know each other? Do you 0:01:08.959 --> 0:01:09.559 want to start? 0:01:09.880 --> 0:01:13.679 Well? Actually, I think we knew each other when you 0:01:13.760 --> 0:01:16.160 were at CNN, but you would not remember me. Well, 0:01:16.319 --> 0:01:17.200 you knew my husband? 0:01:17.480 --> 0:01:18.920 Yes? Was this in Atlanta? 0:01:19.080 --> 0:01:21.600 This is in Atlanta? You worked with my husband. 0:01:21.440 --> 0:01:24.800 In the early days of CNN. Was that at Take two. 0:01:24.920 --> 0:01:26.800 In the mid eighties, So I think that's how we 0:01:26.840 --> 0:01:27.559 initially met. 0:01:27.800 --> 0:01:31.679 Vivian and I have cross paths at various times in 0:01:31.720 --> 0:01:34.760 our lives. I think Vivian is the only person I 0:01:34.800 --> 0:01:38.120 know who has worked at more news organizations than I have. 0:01:38.959 --> 0:01:41.120 Vivian give us The Rundown. 0:01:41.000 --> 0:01:48.040 CNN, New York Times, NPR, NBC, The Guardian well as 0:01:48.040 --> 0:01:51.640 a board member, and also at Twitter doing a news 0:01:51.720 --> 0:01:55.320 role there, and at Discovery running a news documentary channel. 0:01:55.480 --> 0:01:59.000 So basically, like me, Vivian cannot hold a job. So 0:01:59.360 --> 0:02:03.440 we are going to have a conversation today that actually 0:02:03.600 --> 0:02:07.400 was prompted by a conversation I heard at an Aspen 0:02:07.480 --> 0:02:11.400 Institute board meeting. Vivian and I are both very involved 0:02:11.480 --> 0:02:14.280 in the Aspen Institute. In fact, she's got a paying 0:02:14.400 --> 0:02:17.760 job there. Vivian, what exactly is your role at Aspen? 0:02:18.400 --> 0:02:21.400 I run a program at the Aspens too, called Aspen Digital, 0:02:21.400 --> 0:02:24.680 and our focus is on all things technology and media 0:02:24.720 --> 0:02:27.160 and their impact on society, so exactly all the stuff 0:02:27.160 --> 0:02:28.040 we're talking about today. 0:02:28.360 --> 0:02:30.480 And Vivian and I got to know each other even 0:02:30.520 --> 0:02:34.880 better when we both served on the Aspen Institute Commission 0:02:35.120 --> 0:02:38.680 on Disinformation, which has a more formal title, which is 0:02:38.720 --> 0:02:39.359 go Ahead. 0:02:39.200 --> 0:02:41.720 Vivian Commission on Information Disorder. 0:02:41.919 --> 0:02:44.240 Thank you, Yes, And we got to know each other 0:02:44.360 --> 0:02:47.400 well during that time, but we've known each other for 0:02:47.440 --> 0:02:51.120 a long time, so we are excited to have this 0:02:51.200 --> 0:02:54.400 conversation for all of you. I learned a great deal 0:02:54.720 --> 0:02:58.160 and we have two incredible experts who are coming on 0:02:58.440 --> 0:03:03.000 to talk about not all what AI is, but obviously 0:03:03.280 --> 0:03:07.000 the promise and the perils of this new technology. And 0:03:07.040 --> 0:03:10.600 of course we're going to touch on Sam Altman's auster 0:03:10.960 --> 0:03:15.160 and then reinstatement by the board at Open AI, which 0:03:15.280 --> 0:03:19.000 is full of all sorts of intrigue. And this is 0:03:19.040 --> 0:03:22.920 hopefully a podcast that is AI for dummies, but I 0:03:22.960 --> 0:03:26.200 fear the only dummy in the conversation is yours truly. 0:03:26.560 --> 0:03:30.560 So without further ado, let's invite in our guests, Chris 0:03:30.600 --> 0:03:34.480 Wiggins and the Last Star. Welcome to the podcast. Thank 0:03:34.520 --> 0:03:37.040 you so much for being here. I should note that 0:03:37.280 --> 0:03:42.840 November thirtieth today is the one year anniversary of chat GPT, 0:03:43.160 --> 0:03:46.520 so we actually have a newspeg for this podcast. And 0:03:46.600 --> 0:03:49.840 before we dig in, I thought i'd ask you briefly 0:03:50.120 --> 0:03:53.400 about what you all do and why you're qualified to 0:03:53.480 --> 0:03:55.960 have this conversation Chris, Why don't we start with you? 0:03:56.480 --> 0:03:56.760 Sure? 0:03:56.840 --> 0:03:59.480 Fair question. So for the last twenty two years have 0:03:59.520 --> 0:04:02.480 been on the fact ficulty. At Columbia, I teach applied mathematics. 0:04:02.480 --> 0:04:05.120 My research is in machine learning, mostly apply to biology. 0:04:05.480 --> 0:04:07.320 For the last ten years, I've also been the Chief 0:04:07.400 --> 0:04:09.760 Data Scientist to The New York Times, which means I 0:04:09.840 --> 0:04:12.360 lead a team that develops and deployees machine learning. 0:04:12.240 --> 0:04:14.560 And I'm a lost star. Twenty odd years ago, I 0:04:14.600 --> 0:04:17.479 started my career as a computer science researcher, working on 0:04:17.520 --> 0:04:20.800 what we called artificial intuligence back then. Since then, I've 0:04:20.800 --> 0:04:24.279 spent my career building private and public organizations that focus 0:04:24.320 --> 0:04:26.760 on using tech as a way to advance justice and equity, 0:04:27.080 --> 0:04:29.919 and now lead probably one of the largest film propic 0:04:30.040 --> 0:04:32.800 organizations focused on funding AI that makes the world a 0:04:32.800 --> 0:04:33.360 better place. 0:04:33.760 --> 0:04:36.279 Well, I'm very excited to have you both, as well 0:04:36.320 --> 0:04:39.760 as my friend Vivian, who you both know well. And 0:04:40.000 --> 0:04:43.560 I thought i'd start with a very basic question, what 0:04:43.720 --> 0:04:46.800 is AI? Who wants to take a shot at that? 0:04:47.320 --> 0:04:49.520 I could try a historical view. AI is one of 0:04:49.520 --> 0:04:52.640 my favorite drifting targets meeting It's a term that means 0:04:52.680 --> 0:04:56.400 different things to different people in different decades, in different communities. 0:04:56.880 --> 0:04:59.880 So when the term was coined in nineteen fifty five 0:05:00.000 --> 0:05:03.640 by John McCarthy, it was a proposal that the idea 0:05:03.760 --> 0:05:06.880 that any feature of intelligence can be, in principle, be 0:05:07.040 --> 0:05:09.279 so precisely described that a machine can be made to 0:05:09.320 --> 0:05:12.640 simulate it. So the conception of what artificial intelligence meant 0:05:12.640 --> 0:05:15.400 in nineteen fifty five is so different than what it's 0:05:15.440 --> 0:05:18.680 come to me now even in twenty twenty one, let 0:05:18.720 --> 0:05:21.960 alone a year ago today when chat TPT was launched, 0:05:21.960 --> 0:05:25.520 And now everybody when they think of AI, they're thinking 0:05:25.560 --> 0:05:28.200 of a chatbot, which is really chatbot is a small 0:05:28.480 --> 0:05:31.960 example of machine learning, which is a small example of 0:05:32.080 --> 0:05:35.240 artificial intelligence. So the term has come to mean different 0:05:35.240 --> 0:05:37.960 things in different times, which is why the term never 0:05:38.040 --> 0:05:40.159 feels like you're standing on solid ground when you're saying it, 0:05:40.200 --> 0:05:42.560 because different audiences can mean very different things. When you 0:05:42.560 --> 0:05:44.200 say those two letters. 0:05:44.080 --> 0:05:45.880 You know, I'll agree with that. I agree with everything 0:05:45.960 --> 0:05:48.080 Chris said. I mean, on one side, AI is a 0:05:48.200 --> 0:05:51.280 technology conversation. It's a new set of tools that let 0:05:51.279 --> 0:05:54.560 computers do what people have tradisally thought only we could do. 0:05:54.960 --> 0:05:58.200 But it's also something much bigger. It's a social phenomenon. 0:05:58.520 --> 0:06:00.839 It's a moment now where we get to test and 0:06:00.960 --> 0:06:04.440 examine some pretty basic assumptions about what it means to 0:06:04.480 --> 0:06:07.840 have an economy, a political society, about what it means 0:06:07.920 --> 0:06:10.839 to be human. And that's why we're seeing this amazing 0:06:10.880 --> 0:06:12.760 grounds full of interest in what AI is. 0:06:13.320 --> 0:06:17.839 When you think about AI, can you explain in very 0:06:18.560 --> 0:06:22.080 sort of eighth grade terms, how it works, how these 0:06:22.320 --> 0:06:28.920 large language models are assembled, and how machine learning enables 0:06:29.400 --> 0:06:33.640 technology to spew out things that make sense. Chris, can 0:06:33.680 --> 0:06:34.440 you help me with that? 0:06:34.880 --> 0:06:37.720 Sure? I think again. History is really useful here. One 0:06:37.760 --> 0:06:41.760 example of how you might build a chatbot statistically was 0:06:41.960 --> 0:06:46.919 Claude Shannon in probably nineteen forty four was thinking about 0:06:47.040 --> 0:06:49.680 this model where you generate words at random. Imagine that 0:06:49.720 --> 0:06:52.159 you're reading a book and you find some word, and 0:06:52.200 --> 0:06:53.839 then you keep reading in that book until you find 0:06:53.880 --> 0:06:55.440 that word again, and then write down the word that 0:06:55.480 --> 0:06:58.320 follows it. Then keep reading the book, wait until you 0:06:58.320 --> 0:07:01.160 find that word again and write down the word that 0:07:01.160 --> 0:07:04.200 that's the basic nexus of a small language model. So 0:07:04.240 --> 0:07:06.559 you're predicting the next word based on the previous word 0:07:07.240 --> 0:07:09.359 you can think about what's being done today as a 0:07:09.520 --> 0:07:12.760 very large version of that same small language model. It's 0:07:12.800 --> 0:07:16.360 a statistical prediction model. And an important part there is 0:07:16.360 --> 0:07:18.760 that it really matters what book you're training it on, 0:07:19.080 --> 0:07:22.120 and so you need a very large corpus of training data. 0:07:22.200 --> 0:07:24.160 In this case. One of the things that makes large 0:07:24.200 --> 0:07:27.280 language models possible is the vast amount of information that's 0:07:27.320 --> 0:07:31.880 available online, and so computer programs automatically ingest all of 0:07:31.920 --> 0:07:35.160 the text on the web could be from Reddit, newsgroups 0:07:35.240 --> 0:07:37.320 or Wikipedia, or hey, they. 0:07:37.280 --> 0:07:39.920 Use my book the best advice I ever got for 0:07:40.040 --> 0:07:43.040 chat GPT, and nobody asked my permission. By the way, 0:07:43.360 --> 0:07:43.800 that's right. 0:07:43.920 --> 0:07:45.080 That's a whole other issue. 0:07:45.200 --> 0:07:47.680 That's exactly right. That's all other issues is how this 0:07:47.720 --> 0:07:50.720 relate to the rights of the authors. But the statistical 0:07:50.720 --> 0:07:54.200 problem is one of training from data. So the data 0:07:54.240 --> 0:07:57.760 are central and it's counterintuitive, I think to many people 0:07:57.760 --> 0:08:00.880 who think that computers are about writing down rules, and 0:08:00.920 --> 0:08:02.720 when you write down the rules about how we think 0:08:02.760 --> 0:08:05.520 we think, then you'll get something that acts like how 0:08:05.560 --> 0:08:07.840 we think we think. And in fact, for most of 0:08:07.920 --> 0:08:11.120 artificial intelligence as a field, for the last seventy years, 0:08:11.320 --> 0:08:13.840 that's how people thought we were going to achieve artificial 0:08:13.840 --> 0:08:16.360 intelligence was by understanding how we think we think, and 0:08:16.360 --> 0:08:18.360 then you would just simulate it or just program it. 0:08:18.840 --> 0:08:20.640 And the truth is, it's been a realization in the 0:08:20.720 --> 0:08:23.000 last two decades that the way that we are able 0:08:23.040 --> 0:08:26.800 to achieve such exciting results is from taking really large 0:08:26.880 --> 0:08:29.520 data sets and learning from the data how to build 0:08:29.520 --> 0:08:33.400 a computer that emulates, really imitates what we sound like 0:08:33.440 --> 0:08:34.559 when we are intelligent. 0:08:34.640 --> 0:08:38.880 In words, sometimes when I'm writing emails, these words like 0:08:39.040 --> 0:08:41.920 so much. I must use that all the time, thank 0:08:41.920 --> 0:08:46.120 you so much. It shows up in my email if 0:08:46.160 --> 0:08:48.560 I want to just kind of press a button and 0:08:48.600 --> 0:08:52.679 not write anymore. Is that an example a rudimentary example 0:08:52.880 --> 0:08:54.240 of AI. 0:08:53.920 --> 0:08:54.640 Yes, very much. 0:08:54.640 --> 0:08:54.720 So. 0:08:54.760 --> 0:08:57.200 There's the math behind it, which is how are you 0:08:57.240 --> 0:08:59.360 going to predict the next word? But the other thing 0:08:59.400 --> 0:09:02.000 about it is the product and sort of the user interface. 0:09:02.320 --> 0:09:04.880 People like to talk about how in the late fifties 0:09:05.080 --> 0:09:08.720 night at Stanford there was John McCarthy who was working 0:09:08.720 --> 0:09:11.079 on the mathematics of AI, and then there were people 0:09:11.080 --> 0:09:13.839 like Doug Engelbart who were working on the product of AI. 0:09:14.160 --> 0:09:16.000 How are we going to make an interface that allows 0:09:16.000 --> 0:09:18.440 people to interact with the computer. Well, so when you 0:09:18.520 --> 0:09:20.920 just saw it, there was a good example of good math. 0:09:21.240 --> 0:09:23.040 And the math could be as simple as counting the 0:09:23.120 --> 0:09:25.160 number of times that the word what follows the word, 0:09:25.240 --> 0:09:27.720 So it's very simple math. But as well as the 0:09:27.720 --> 0:09:30.160 product idea, which is, how do I make a suggestion 0:09:30.240 --> 0:09:32.439 to you in a way that's useful to you while 0:09:32.480 --> 0:09:36.000 you use that digital product and not creepy and not intrusive. 0:09:36.360 --> 0:09:39.439 So yeah, that's another thing that we're seeing with chatchept 0:09:40.080 --> 0:09:44.199 is a good coming together of technology and mathematics and 0:09:44.240 --> 0:09:47.080 statistical models, but also just a nice product that people 0:09:47.080 --> 0:09:48.280 are enjoying musing. 0:09:48.280 --> 0:09:51.240 What you're describing, Chris, is a predictive model. But so 0:09:51.280 --> 0:09:54.720 many people, particularly since a year ago today when chatchipt 0:09:55.040 --> 0:09:57.880 came out and sort of collectively blew the world's minds, 0:09:58.520 --> 0:10:02.280 it feels like we're talking to a machine that is 0:10:02.320 --> 0:10:05.240 actually thinking, that is actually sentient, and it's in fact 0:10:05.320 --> 0:10:09.040 design that way, and that has societal implications, some of 0:10:09.080 --> 0:10:11.440 the societal implications that you were referring to earlier. 0:10:11.520 --> 0:10:15.040 VELAs know, you're pointing out the critical kind of missing 0:10:15.080 --> 0:10:17.800 element in what Chris described as what AI is today. 0:10:18.240 --> 0:10:20.319 At the end of the day, every version of AI 0:10:20.360 --> 0:10:23.280 that we have today. It's a mathematical model that predicts 0:10:23.280 --> 0:10:27.560 what happens next based on what's happened before. It doesn't reason, 0:10:27.880 --> 0:10:31.800 it doesn't think, it doesn't have agency, it doesn't have preferences, 0:10:32.480 --> 0:10:34.439 all of the things that people now try to scare 0:10:34.520 --> 0:10:36.240 us with that. I imagine we'll talk a little bit 0:10:36.240 --> 0:10:40.280 about that crazy term AGI. Today's AI is none of that. 0:10:40.760 --> 0:10:42.400 I often like to say, do you all remember that 0:10:42.480 --> 0:10:44.320 movie Honey, I Stroked the Kids? 0:10:44.480 --> 0:10:46.320 Yeah, Rick Moranez, Yeah. 0:10:46.240 --> 0:10:48.840 Great movie. Right, Today's AI. All it is is this, 0:10:49.280 --> 0:10:51.280 take everything that's ever been written, put it in a 0:10:51.320 --> 0:10:54.360 giant library, beam a rate gun at that library, and 0:10:54.480 --> 0:10:57.000 enough power to power a small city for months and months, 0:10:57.080 --> 0:10:59.280 and say, how do we compress that entire library down? 0:10:59.280 --> 0:11:01.800 And give us one little map? And all the math 0:11:01.880 --> 0:11:04.880 does is it says this. If before, when people said 0:11:04.920 --> 0:11:08.400 a word, they often followed it with another word, that's 0:11:08.440 --> 0:11:11.079 all we're going to do right now. Now, what that does. 0:11:11.160 --> 0:11:14.240 It's an amazing magic trick. It's a great illusion. It 0:11:14.320 --> 0:11:16.560 makes you think you're talking to somebody who wants to 0:11:16.559 --> 0:11:18.520 talk back to you. But at the end of the day, 0:11:18.559 --> 0:11:20.800 all the machine is doing is predicting what the next 0:11:20.800 --> 0:11:23.520 word and the answer should be. This is so critical 0:11:24.120 --> 0:11:27.000 because it reframes how we engage with these tools, and 0:11:27.040 --> 0:11:28.960 that's really all they are. They're just tools. They're not, 0:11:29.480 --> 0:11:32.320 you know, all knowing entities. They're not partners, they're not 0:11:32.440 --> 0:11:36.120 conversational and sparring buddies. They're just tools that help us 0:11:36.280 --> 0:11:37.520 maybe be better. 0:11:38.040 --> 0:11:40.880 Let me ask you this because I thought this was interesting. 0:11:41.640 --> 0:11:45.800 Bill Gates recently noted that AI as it exists today 0:11:46.000 --> 0:11:50.120 is quote still pretty dumb. Chris. Do you agree with that? 0:11:50.960 --> 0:11:51.160 Yeah? 0:11:51.160 --> 0:11:54.400 Absolutely so. I think what VELAs is saying is apt, 0:11:54.440 --> 0:11:57.880 which is language, and the ability to produce language is 0:11:57.880 --> 0:12:01.120 a great imitation of what thinking. And in fact, I 0:12:01.240 --> 0:12:04.120 use the word imitation because in Hellan Turing's original nineteen 0:12:04.160 --> 0:12:08.000 fifty paper on can MA Machines think he basically set 0:12:08.000 --> 0:12:11.040 out this problem. Imagine a computer that could imitate what 0:12:11.080 --> 0:12:13.079 it's like to talk to somebody. That's sort of an 0:12:13.080 --> 0:12:16.040 operaginalization of what it means to think. But there's still 0:12:16.040 --> 0:12:20.199 many things a I can't do. As often described, planning 0:12:20.240 --> 0:12:24.920 is difficult, Compositional thinking is difficult, Working with multiple modes 0:12:24.920 --> 0:12:27.640 at once is difficult. Meaning like words and images together. 0:12:27.880 --> 0:12:30.600 So I think you're right that it's uncanny. Right, we're 0:12:30.600 --> 0:12:33.679 in the uncanny valley of conversations right now with chatbots. 0:12:34.040 --> 0:12:37.080 But it's very difficult for people not to impose this 0:12:37.200 --> 0:12:40.880 belief that it is intelligent or thoughtful. And the truth 0:12:41.000 --> 0:12:43.559 is people have been having that experience for as long 0:12:43.600 --> 0:12:46.280 as they've been building chatbots. Even in the nineteen sixties, 0:12:46.520 --> 0:12:49.880 people were building chatbots based on simple rules, and users 0:12:50.160 --> 0:12:52.560 using that chatbot had the same experience of feeling like 0:12:52.600 --> 0:12:54.319 even though they knew it was just a very simple 0:12:54.320 --> 0:12:57.280 computer program, there was the emotional resonance was one as 0:12:57.280 --> 0:12:59.120 though you were talking to an intelligent agent. 0:12:59.760 --> 0:13:03.400 Hear this word sentient a lot, which of course is 0:13:03.480 --> 0:13:07.679 capable of sensing or feeling conscious of, or responsive to 0:13:07.720 --> 0:13:12.360 the sensations of seeing, hearing, feeling, tasting, or smelling sentient beings, 0:13:12.480 --> 0:13:14.760 which is really what it means to be a human. 0:13:15.960 --> 0:13:19.880 Does AI have the capacity to be sentient? 0:13:20.280 --> 0:13:21.760 I think what we've shown is that it does a 0:13:21.760 --> 0:13:23.719 great imitation of it. But I think it's important for 0:13:23.800 --> 0:13:26.080 us all to remember that it is, as Vela said, 0:13:26.360 --> 0:13:29.240 just math right. It is a mathematical model that spits 0:13:29.240 --> 0:13:32.400 out words and it's optimized for generating words that sound 0:13:32.520 --> 0:13:35.200 like what a human being would say and given the 0:13:35.200 --> 0:13:37.640 same prompt. But we should remember that it is a 0:13:37.679 --> 0:13:41.800 machine and it's executing a mathematical act that we trained 0:13:41.800 --> 0:13:42.040 it on. 0:13:42.360 --> 0:13:44.360 But yet there's a lot of experts out there and 0:13:44.440 --> 0:13:47.760 researchers and some pretty serious people who are trying to 0:13:47.800 --> 0:13:52.600 warn us that these machines may become sentient. Now, is 0:13:52.640 --> 0:13:55.080 that just a matter of seeing too many sci fi 0:13:55.120 --> 0:14:00.120 movies or is that something that is possible obviously not today, 0:14:00.160 --> 0:14:01.040 but on the horizon. 0:14:01.320 --> 0:14:03.480 I think there's a couple of reasons why people are 0:14:03.520 --> 0:14:06.280 warning us of that possibility. I mean, again, part of 0:14:06.280 --> 0:14:08.560 it is wordplay, and I think that's what Alan during 0:14:08.640 --> 0:14:10.480 was getting at in nineteen fifty when he said, can 0:14:10.559 --> 0:14:12.959 machines think? Is an ill posed problem, so let's try 0:14:13.000 --> 0:14:15.800 to operationalize it. But I think the warnings are often 0:14:16.160 --> 0:14:21.560 distractions from real problems that automated inequality and other downsides 0:14:21.560 --> 0:14:24.240 of using algorithms today are causing. In the here and now. 0:14:24.400 --> 0:14:28.040 It's sometimes very difficult to think about our existing challenges 0:14:28.040 --> 0:14:32.240 in sociotechnical systems, and somehow more pleasant to think about 0:14:32.280 --> 0:14:35.800 this terminator doomsday future which is not cured yet. I 0:14:35.840 --> 0:14:36.800 think also there's. 0:14:36.640 --> 0:14:39.200 A yet yet that's scary, that's right. 0:14:39.440 --> 0:14:41.680 I think there's also a concern that people are putting 0:14:41.680 --> 0:14:44.360 forward this idea of a doomsday because there's only a 0:14:44.360 --> 0:14:46.960 small number of companies at present which are able to 0:14:47.000 --> 0:14:51.760 afford amassing lots of data and producing really good products. 0:14:52.000 --> 0:14:53.920 And often these companies are saying, we are the ones 0:14:53.920 --> 0:14:56.400 who can tell you how to regulate this. So there's 0:14:56.440 --> 0:14:59.000 concern that some of the doomsday scenario might be coupled 0:14:59.000 --> 0:15:02.720 to re trey capture or getting ahead of potential regulation 0:15:02.840 --> 0:15:04.800 both domestically and internationally. 0:15:05.120 --> 0:15:07.040 Let's zoom out a little bit. I agree with what 0:15:07.160 --> 0:15:09.800 Chris said, but maybe and you spoke to experts and 0:15:09.880 --> 0:15:12.360 serious people who are trying to scare us all, and 0:15:12.400 --> 0:15:14.600 I just want to put us in context. There are, 0:15:14.880 --> 0:15:17.760 you know, eight billion people on the planet. There are 0:15:17.760 --> 0:15:21.440 maybe a few hundred thousand that really understand how AI works. 0:15:21.960 --> 0:15:24.960 And there's maybe on the order of a few thousand 0:15:25.000 --> 0:15:27.480 people who have decided that what they care about more 0:15:27.520 --> 0:15:30.800 than anything else is the existential risk to humanity. Let's 0:15:30.840 --> 0:15:33.360 just put that in context. There are billions of people 0:15:33.360 --> 0:15:35.720 on the planet today that could use what AI offers 0:15:35.760 --> 0:15:39.120 as a promise to fundamentally change what their lives look like. 0:15:39.160 --> 0:15:42.080 They're access to economic opportunity, to any number of other things. 0:15:42.480 --> 0:15:45.640 There are hundreds of thousands or millions of people who 0:15:45.680 --> 0:15:49.479 work in companies that could fundamentally change their relationship with customers. 0:15:49.840 --> 0:15:51.720 So what do we have. We have a small set 0:15:51.720 --> 0:15:54.640 of people with direct access to some of these tools, 0:15:54.640 --> 0:15:56.920 and we'll talk about how that came about shortly, who 0:15:56.960 --> 0:15:58.920 have come up with this common idea that the thing 0:15:58.920 --> 0:16:00.640 we should care about more than any else is that 0:16:00.680 --> 0:16:03.120 AI will kill us all. And in the meantime, we're 0:16:03.160 --> 0:16:05.560 living in a world where there are so many opportunities 0:16:05.560 --> 0:16:07.840 and challenges here and now that we should be spending 0:16:07.840 --> 0:16:08.760 our time thinking about. 0:16:09.400 --> 0:16:12.640 So there are the here and now, there is the 0:16:12.720 --> 0:16:16.960 long term future, and there's that whole middle ground that 0:16:17.000 --> 0:16:19.240 I think many of us haven't addressed. But before we 0:16:19.280 --> 0:16:21.760 come back to that, just to help our listeners a 0:16:21.760 --> 0:16:26.480 little bit, we hear the term AGI artificial general intelligence, 0:16:26.640 --> 0:16:30.600 not to be confused with generative artificial intelligence. Very confusing 0:16:30.800 --> 0:16:33.400 even for people that pay attention to these things. AGI 0:16:33.600 --> 0:16:36.280 is this sort of robot overlord saying we're talking about 0:16:36.280 --> 0:16:38.040 that may or may not ever happen. Is that right? 0:16:38.320 --> 0:16:40.880 Yeah? What is that? You guys help me out. I'm 0:16:40.920 --> 0:16:41.760 the dumb one here. 0:16:42.320 --> 0:16:44.280 I think one thing that's useful is to remember the 0:16:44.320 --> 0:16:48.720 two different g's in those two acronyms. The GAI is generitive, 0:16:48.760 --> 0:16:51.960 but AGI is general. So when people talk about AGI 0:16:52.120 --> 0:16:55.640 for general intelligence, part of which is exciting is the 0:16:55.680 --> 0:16:59.640 idea that in the last fifty years we've done species, 0:17:00.120 --> 0:17:02.480 not me personally, but the human being species have done 0:17:02.480 --> 0:17:04.479 a really good job building algorithms that are good for 0:17:04.520 --> 0:17:07.560 individual tasks. Like you can build an app that can 0:17:07.880 --> 0:17:09.720 take a picture and say does this have a dog 0:17:09.760 --> 0:17:11.640 face in it? Or a cat face in it? That's 0:17:11.680 --> 0:17:14.560 a specific use of statistical modeling, which is good for 0:17:14.600 --> 0:17:18.080 that specific use case. So the dream of AGI is 0:17:18.119 --> 0:17:21.399 that you can produce one algorithm, one machine, one model 0:17:21.680 --> 0:17:25.360 that's good not only for disinbiguating cat faces from dog faces, 0:17:25.359 --> 0:17:29.679 but also composing a sonnet or enjoying strawberries and cream, 0:17:29.960 --> 0:17:33.639 or whatever general problem you would like the machine to solve. 0:17:33.800 --> 0:17:36.600 So that's the g of agis general. It's very easy 0:17:36.600 --> 0:17:38.880 for us to make machine learning models that are good 0:17:38.880 --> 0:17:41.199 for one specific task. It's much harder to make a 0:17:41.240 --> 0:17:43.800 machine learning model that's general and is able to do 0:17:43.920 --> 0:17:46.359 anything that we consider an intelligent task. 0:17:46.840 --> 0:17:50.040 So where are we in terms of And I didn't 0:17:50.080 --> 0:17:54.080 really understand that explanation, Chris, can you try it again 0:17:54.440 --> 0:17:57.439 in a more like I'm not a Columbia student, or 0:17:57.760 --> 0:18:01.320 just pretend like I'm in sixth grade? Help me out, sure, 0:18:01.400 --> 0:18:02.360 help me out. 0:18:02.840 --> 0:18:07.960 So for many decades, we've been able to build specific 0:18:08.119 --> 0:18:12.440 machine learning models. So a specific algorithm that can tell 0:18:12.480 --> 0:18:15.480 the difference between a picture of a dog and a 0:18:15.520 --> 0:18:17.919 picture of a cat, say, that's a specific problem. And 0:18:18.000 --> 0:18:21.119 we've been very good for decades at building algorithms that 0:18:21.160 --> 0:18:24.439 can do very specific and focused tasks. One of the 0:18:24.480 --> 0:18:27.479 things that we've seen with chatbots that are trained on 0:18:27.560 --> 0:18:30.600 a wide variety of documents is that you can have 0:18:30.680 --> 0:18:34.160 a plausible conversation with a chatbot about a wide variety 0:18:34.200 --> 0:18:37.320 of topics. So if you've trained a chatbot only on 0:18:37.400 --> 0:18:40.840 chemistry textbooks, you will have a great conversation about chemistry 0:18:40.880 --> 0:18:44.280 and not about any other subject. But by training a 0:18:44.359 --> 0:18:48.200 chatbot on a wide variety of topics chemistry, philosophy, and 0:18:48.240 --> 0:18:52.240 all points in between, people are experiencing this shock that 0:18:52.320 --> 0:18:56.280 you can interact with an AI, meaning an algorithm that 0:18:56.680 --> 0:19:01.359 works not only solving a specific problem, solving a general problem, 0:19:01.440 --> 0:19:04.479 in this case, the general problem of having having an 0:19:04.520 --> 0:19:08.840 intelligent sounding conversation about a general breath of topics. 0:19:10.680 --> 0:19:13.040 After a quick break, we'll be back with my co 0:19:13.320 --> 0:19:17.080 pilot and plus one Vivian Shiller, talking to Chris Wiggins 0:19:17.200 --> 0:19:22.200 and velost Star. If you want to get smarter every 0:19:22.240 --> 0:19:25.320 morning with a breakdown of the news and fascinating takes 0:19:25.359 --> 0:19:28.480 on health and wellness and pop culture, sign up for 0:19:28.520 --> 0:19:31.640 our daily newsletter, Wake Up Call by going to Katiecuric 0:19:31.720 --> 0:19:38.360 dot com. We're back with Chris Wiggins and velost Star, 0:19:38.520 --> 0:19:43.119 along with my plus one Vivian Schiller. Have you guys 0:19:43.240 --> 0:19:47.160 used chat GPT or bard. Have you tried to have 0:19:47.240 --> 0:19:50.480 it write speeches for you or come up with any 0:19:50.600 --> 0:19:54.320 kind of documents. I'm sure you've tested it, Vivian, what 0:19:54.359 --> 0:19:55.720 has your experience been like. 0:19:56.200 --> 0:20:00.640 I've used chat gpt to develop an itinerary. I took 0:20:00.680 --> 0:20:02.720 it to Japan. I knew I needed I had some 0:20:02.840 --> 0:20:05.840 time between two places I needed to be, and I 0:20:05.880 --> 0:20:08.280 had certain things that I was interested in, certain things 0:20:08.320 --> 0:20:10.960 I was less interested in, didn't know how long it 0:20:11.000 --> 0:20:13.440 took to get paced the place, and actually chat Gipt 0:20:13.600 --> 0:20:15.919 gave me an amazing itinerary, so it was very useful. 0:20:16.119 --> 0:20:18.879 Travel agents probably don't like that, how about you, guys? 0:20:19.480 --> 0:20:21.840 Yeah, you know, I mean, I've used every LM out there, 0:20:21.880 --> 0:20:23.199 and so I'm like, I'm no longer, you know. I 0:20:23.200 --> 0:20:25.080 wish I could say it was at the emergent frontier 0:20:25.080 --> 0:20:26.679 of AI, but I'm no longer Now I have a 0:20:26.680 --> 0:20:28.560 different role. But I spent a lot of time with 0:20:28.600 --> 0:20:30.240