1 00:00:00,200 --> 00:00:03,640 Speaker 1: Cancer Straight Talk is a podcast for Memorial Sloan Kettering 2 00:00:03,760 --> 00:00:08,360 Speaker 1: Cancer Center. We're host doctor Diane Reedy. Lagunis has intimate 3 00:00:08,440 --> 00:00:13,480 Speaker 1: conversations with patients and experts about topics like dating and sex, 4 00:00:14,000 --> 00:00:18,319 Speaker 1: exercise and diet, the power of gratitude, and more. I 5 00:00:18,480 --> 00:00:21,439 Speaker 1: love being her guest back in April. Listen to Cancer 6 00:00:21,480 --> 00:00:29,760 Speaker 1: Straight Talk. You'll learn so much. Hi everyone, I'm Kitty 7 00:00:29,840 --> 00:00:36,680 Speaker 1: Kuric and this is next question. So I have to 8 00:00:36,760 --> 00:00:40,760 Speaker 1: confess friends that if I sound weird, it's because I 9 00:00:40,800 --> 00:00:45,919 Speaker 1: have a terrible cold. So I apologize in advance. Luckily, 10 00:00:46,120 --> 00:00:50,200 Speaker 1: my plus one isn't sitting next to me catching my germs, 11 00:00:50,280 --> 00:00:54,800 Speaker 1: but she is at a remote location. Where are you, Vivian. 12 00:00:55,000 --> 00:00:56,560 Speaker 2: I'm in Bethesta, Maryland, in my home. 13 00:00:56,880 --> 00:01:01,600 Speaker 1: Oh nice, Well, Vivian Schiller, as you probably heard, is 14 00:01:01,680 --> 00:01:05,160 Speaker 1: my plus one today. And Vivian, I thought i'd start 15 00:01:05,200 --> 00:01:08,920 Speaker 1: by telling folks how we know each other? Do you 16 00:01:08,959 --> 00:01:09,559 Speaker 1: want to start? 17 00:01:09,880 --> 00:01:13,679 Speaker 2: Well? Actually, I think we knew each other when you 18 00:01:13,760 --> 00:01:16,160 Speaker 2: were at CNN, but you would not remember me. Well, 19 00:01:16,319 --> 00:01:17,200 Speaker 2: you knew my husband? 20 00:01:17,480 --> 00:01:18,920 Speaker 1: Yes? Was this in Atlanta? 21 00:01:19,080 --> 00:01:21,600 Speaker 2: This is in Atlanta? You worked with my husband. 22 00:01:21,440 --> 00:01:24,800 Speaker 1: In the early days of CNN. Was that at Take two. 23 00:01:24,920 --> 00:01:26,800 Speaker 2: In the mid eighties, So I think that's how we 24 00:01:26,840 --> 00:01:27,559 Speaker 2: initially met. 25 00:01:27,800 --> 00:01:31,679 Speaker 1: Vivian and I have cross paths at various times in 26 00:01:31,720 --> 00:01:34,760 Speaker 1: our lives. I think Vivian is the only person I 27 00:01:34,800 --> 00:01:38,120 Speaker 1: know who has worked at more news organizations than I have. 28 00:01:38,959 --> 00:01:41,120 Speaker 1: Vivian give us The Rundown. 29 00:01:41,000 --> 00:01:48,040 Speaker 2: CNN, New York Times, NPR, NBC, The Guardian well as 30 00:01:48,040 --> 00:01:51,640 Speaker 2: a board member, and also at Twitter doing a news 31 00:01:51,720 --> 00:01:55,320 Speaker 2: role there, and at Discovery running a news documentary channel. 32 00:01:55,480 --> 00:01:59,000 Speaker 1: So basically, like me, Vivian cannot hold a job. So 33 00:01:59,360 --> 00:02:03,440 Speaker 1: we are going to have a conversation today that actually 34 00:02:03,600 --> 00:02:07,400 Speaker 1: was prompted by a conversation I heard at an Aspen 35 00:02:07,480 --> 00:02:11,400 Speaker 1: Institute board meeting. Vivian and I are both very involved 36 00:02:11,480 --> 00:02:14,280 Speaker 1: in the Aspen Institute. In fact, she's got a paying 37 00:02:14,400 --> 00:02:17,760 Speaker 1: job there. Vivian, what exactly is your role at Aspen? 38 00:02:18,400 --> 00:02:21,400 Speaker 2: I run a program at the Aspens too, called Aspen Digital, 39 00:02:21,400 --> 00:02:24,680 Speaker 2: and our focus is on all things technology and media 40 00:02:24,720 --> 00:02:27,160 Speaker 2: and their impact on society, so exactly all the stuff 41 00:02:27,160 --> 00:02:28,040 Speaker 2: we're talking about today. 42 00:02:28,360 --> 00:02:30,480 Speaker 1: And Vivian and I got to know each other even 43 00:02:30,520 --> 00:02:34,880 Speaker 1: better when we both served on the Aspen Institute Commission 44 00:02:35,120 --> 00:02:38,680 Speaker 1: on Disinformation, which has a more formal title, which is 45 00:02:38,720 --> 00:02:39,359 Speaker 1: go Ahead. 46 00:02:39,200 --> 00:02:41,720 Speaker 2: Vivian Commission on Information Disorder. 47 00:02:41,919 --> 00:02:44,240 Speaker 1: Thank you, Yes, And we got to know each other 48 00:02:44,360 --> 00:02:47,400 Speaker 1: well during that time, but we've known each other for 49 00:02:47,440 --> 00:02:51,120 Speaker 1: a long time, so we are excited to have this 50 00:02:51,200 --> 00:02:54,400 Speaker 1: conversation for all of you. I learned a great deal 51 00:02:54,720 --> 00:02:58,160 Speaker 1: and we have two incredible experts who are coming on 52 00:02:58,440 --> 00:03:03,000 Speaker 1: to talk about not all what AI is, but obviously 53 00:03:03,280 --> 00:03:07,000 Speaker 1: the promise and the perils of this new technology. And 54 00:03:07,040 --> 00:03:10,600 Speaker 1: of course we're going to touch on Sam Altman's auster 55 00:03:10,960 --> 00:03:15,160 Speaker 1: and then reinstatement by the board at Open AI, which 56 00:03:15,280 --> 00:03:19,000 Speaker 1: is full of all sorts of intrigue. And this is 57 00:03:19,040 --> 00:03:22,920 Speaker 1: hopefully a podcast that is AI for dummies, but I 58 00:03:22,960 --> 00:03:26,200 Speaker 1: fear the only dummy in the conversation is yours truly. 59 00:03:26,560 --> 00:03:30,560 Speaker 1: So without further ado, let's invite in our guests, Chris 60 00:03:30,600 --> 00:03:34,480 Speaker 1: Wiggins and the Last Star. Welcome to the podcast. Thank 61 00:03:34,520 --> 00:03:37,040 Speaker 1: you so much for being here. I should note that 62 00:03:37,280 --> 00:03:42,840 Speaker 1: November thirtieth today is the one year anniversary of chat GPT, 63 00:03:43,160 --> 00:03:46,520 Speaker 1: so we actually have a newspeg for this podcast. And 64 00:03:46,600 --> 00:03:49,840 Speaker 1: before we dig in, I thought i'd ask you briefly 65 00:03:50,120 --> 00:03:53,400 Speaker 1: about what you all do and why you're qualified to 66 00:03:53,480 --> 00:03:55,960 Speaker 1: have this conversation Chris, Why don't we start with you? 67 00:03:56,480 --> 00:03:56,760 Speaker 3: Sure? 68 00:03:56,840 --> 00:03:59,480 Speaker 4: Fair question. So for the last twenty two years have 69 00:03:59,520 --> 00:04:02,480 Speaker 4: been on the fact ficulty. At Columbia, I teach applied mathematics. 70 00:04:02,480 --> 00:04:05,120 Speaker 4: My research is in machine learning, mostly apply to biology. 71 00:04:05,480 --> 00:04:07,320 Speaker 4: For the last ten years, I've also been the Chief 72 00:04:07,400 --> 00:04:09,760 Speaker 4: Data Scientist to The New York Times, which means I 73 00:04:09,840 --> 00:04:12,360 Speaker 4: lead a team that develops and deployees machine learning. 74 00:04:12,240 --> 00:04:14,560 Speaker 3: And I'm a lost star. Twenty odd years ago, I 75 00:04:14,600 --> 00:04:17,479 Speaker 3: started my career as a computer science researcher, working on 76 00:04:17,520 --> 00:04:20,800 Speaker 3: what we called artificial intuligence back then. Since then, I've 77 00:04:20,800 --> 00:04:24,279 Speaker 3: spent my career building private and public organizations that focus 78 00:04:24,320 --> 00:04:26,760 Speaker 3: on using tech as a way to advance justice and equity, 79 00:04:27,080 --> 00:04:29,919 Speaker 3: and now lead probably one of the largest film propic 80 00:04:30,040 --> 00:04:32,800 Speaker 3: organizations focused on funding AI that makes the world a 81 00:04:32,800 --> 00:04:33,360 Speaker 3: better place. 82 00:04:33,760 --> 00:04:36,279 Speaker 1: Well, I'm very excited to have you both, as well 83 00:04:36,320 --> 00:04:39,760 Speaker 1: as my friend Vivian, who you both know well. And 84 00:04:40,000 --> 00:04:43,560 Speaker 1: I thought i'd start with a very basic question, what 85 00:04:43,720 --> 00:04:46,800 Speaker 1: is AI? Who wants to take a shot at that? 86 00:04:47,320 --> 00:04:49,520 Speaker 4: I could try a historical view. AI is one of 87 00:04:49,520 --> 00:04:52,640 Speaker 4: my favorite drifting targets meeting It's a term that means 88 00:04:52,680 --> 00:04:56,400 Speaker 4: different things to different people in different decades, in different communities. 89 00:04:56,880 --> 00:04:59,880 Speaker 4: So when the term was coined in nineteen fifty five 90 00:05:00,000 --> 00:05:03,640 Speaker 4: by John McCarthy, it was a proposal that the idea 91 00:05:03,760 --> 00:05:06,880 Speaker 4: that any feature of intelligence can be, in principle, be 92 00:05:07,040 --> 00:05:09,279 Speaker 4: so precisely described that a machine can be made to 93 00:05:09,320 --> 00:05:12,640 Speaker 4: simulate it. So the conception of what artificial intelligence meant 94 00:05:12,640 --> 00:05:15,400 Speaker 4: in nineteen fifty five is so different than what it's 95 00:05:15,440 --> 00:05:18,680 Speaker 4: come to me now even in twenty twenty one, let 96 00:05:18,720 --> 00:05:21,960 Speaker 4: alone a year ago today when chat TPT was launched, 97 00:05:21,960 --> 00:05:25,520 Speaker 4: And now everybody when they think of AI, they're thinking 98 00:05:25,560 --> 00:05:28,200 Speaker 4: of a chatbot, which is really chatbot is a small 99 00:05:28,480 --> 00:05:31,960 Speaker 4: example of machine learning, which is a small example of 100 00:05:32,080 --> 00:05:35,240 Speaker 4: artificial intelligence. So the term has come to mean different 101 00:05:35,240 --> 00:05:37,960 Speaker 4: things in different times, which is why the term never 102 00:05:38,040 --> 00:05:40,159 Speaker 4: feels like you're standing on solid ground when you're saying it, 103 00:05:40,200 --> 00:05:42,560 Speaker 4: because different audiences can mean very different things. When you 104 00:05:42,560 --> 00:05:44,200 Speaker 4: say those two letters. 105 00:05:44,080 --> 00:05:45,880 Speaker 3: You know, I'll agree with that. I agree with everything 106 00:05:45,960 --> 00:05:48,080 Speaker 3: Chris said. I mean, on one side, AI is a 107 00:05:48,200 --> 00:05:51,280 Speaker 3: technology conversation. It's a new set of tools that let 108 00:05:51,279 --> 00:05:54,560 Speaker 3: computers do what people have tradisally thought only we could do. 109 00:05:54,960 --> 00:05:58,200 Speaker 3: But it's also something much bigger. It's a social phenomenon. 110 00:05:58,520 --> 00:06:00,839 Speaker 3: It's a moment now where we get to test and 111 00:06:00,960 --> 00:06:04,440 Speaker 3: examine some pretty basic assumptions about what it means to 112 00:06:04,480 --> 00:06:07,840 Speaker 3: have an economy, a political society, about what it means 113 00:06:07,920 --> 00:06:10,839 Speaker 3: to be human. And that's why we're seeing this amazing 114 00:06:10,880 --> 00:06:12,760 Speaker 3: grounds full of interest in what AI is. 115 00:06:13,320 --> 00:06:17,839 Speaker 1: When you think about AI, can you explain in very 116 00:06:18,560 --> 00:06:22,080 Speaker 1: sort of eighth grade terms, how it works, how these 117 00:06:22,320 --> 00:06:28,920 Speaker 1: large language models are assembled, and how machine learning enables 118 00:06:29,400 --> 00:06:33,640 Speaker 1: technology to spew out things that make sense. Chris, can 119 00:06:33,680 --> 00:06:34,440 Speaker 1: you help me with that? 120 00:06:34,880 --> 00:06:37,720 Speaker 4: Sure? I think again. History is really useful here. One 121 00:06:37,760 --> 00:06:41,760 Speaker 4: example of how you might build a chatbot statistically was 122 00:06:41,960 --> 00:06:46,919 Speaker 4: Claude Shannon in probably nineteen forty four was thinking about 123 00:06:47,040 --> 00:06:49,680 Speaker 4: this model where you generate words at random. Imagine that 124 00:06:49,720 --> 00:06:52,159 Speaker 4: you're reading a book and you find some word, and 125 00:06:52,200 --> 00:06:53,839 Speaker 4: then you keep reading in that book until you find 126 00:06:53,880 --> 00:06:55,440 Speaker 4: that word again, and then write down the word that 127 00:06:55,480 --> 00:06:58,320 Speaker 4: follows it. Then keep reading the book, wait until you 128 00:06:58,320 --> 00:07:01,160 Speaker 4: find that word again and write down the word that 129 00:07:01,160 --> 00:07:04,200 Speaker 4: that's the basic nexus of a small language model. So 130 00:07:04,240 --> 00:07:06,559 Speaker 4: you're predicting the next word based on the previous word 131 00:07:07,240 --> 00:07:09,359 Speaker 4: you can think about what's being done today as a 132 00:07:09,520 --> 00:07:12,760 Speaker 4: very large version of that same small language model. It's 133 00:07:12,800 --> 00:07:16,360 Speaker 4: a statistical prediction model. And an important part there is 134 00:07:16,360 --> 00:07:18,760 Speaker 4: that it really matters what book you're training it on, 135 00:07:19,080 --> 00:07:22,120 Speaker 4: and so you need a very large corpus of training data. 136 00:07:22,200 --> 00:07:24,160 Speaker 4: In this case. One of the things that makes large 137 00:07:24,200 --> 00:07:27,280 Speaker 4: language models possible is the vast amount of information that's 138 00:07:27,320 --> 00:07:31,880 Speaker 4: available online, and so computer programs automatically ingest all of 139 00:07:31,920 --> 00:07:35,160 Speaker 4: the text on the web could be from Reddit, newsgroups 140 00:07:35,240 --> 00:07:37,320 Speaker 4: or Wikipedia, or hey, they. 141 00:07:37,280 --> 00:07:39,920 Speaker 1: Use my book the best advice I ever got for 142 00:07:40,040 --> 00:07:43,040 Speaker 1: chat GPT, and nobody asked my permission. By the way, 143 00:07:43,360 --> 00:07:43,800 Speaker 1: that's right. 144 00:07:43,920 --> 00:07:45,080 Speaker 2: That's a whole other issue. 145 00:07:45,200 --> 00:07:47,680 Speaker 4: That's exactly right. That's all other issues is how this 146 00:07:47,720 --> 00:07:50,720 Speaker 4: relate to the rights of the authors. But the statistical 147 00:07:50,720 --> 00:07:54,200 Speaker 4: problem is one of training from data. So the data 148 00:07:54,240 --> 00:07:57,760 Speaker 4: are central and it's counterintuitive, I think to many people 149 00:07:57,760 --> 00:08:00,880 Speaker 4: who think that computers are about writing down rules, and 150 00:08:00,920 --> 00:08:02,720 Speaker 4: when you write down the rules about how we think 151 00:08:02,760 --> 00:08:05,520 Speaker 4: we think, then you'll get something that acts like how 152 00:08:05,560 --> 00:08:07,840 Speaker 4: we think we think. And in fact, for most of 153 00:08:07,920 --> 00:08:11,120 Speaker 4: artificial intelligence as a field, for the last seventy years, 154 00:08:11,320 --> 00:08:13,840 Speaker 4: that's how people thought we were going to achieve artificial 155 00:08:13,840 --> 00:08:16,360 Speaker 4: intelligence was by understanding how we think we think, and 156 00:08:16,360 --> 00:08:18,360 Speaker 4: then you would just simulate it or just program it. 157 00:08:18,840 --> 00:08:20,640 Speaker 4: And the truth is, it's been a realization in the 158 00:08:20,720 --> 00:08:23,000 Speaker 4: last two decades that the way that we are able 159 00:08:23,040 --> 00:08:26,800 Speaker 4: to achieve such exciting results is from taking really large 160 00:08:26,880 --> 00:08:29,520 Speaker 4: data sets and learning from the data how to build 161 00:08:29,520 --> 00:08:33,400 Speaker 4: a computer that emulates, really imitates what we sound like 162 00:08:33,440 --> 00:08:34,559 Speaker 4: when we are intelligent. 163 00:08:34,640 --> 00:08:38,880 Speaker 1: In words, sometimes when I'm writing emails, these words like 164 00:08:39,040 --> 00:08:41,920 Speaker 1: so much. I must use that all the time, thank 165 00:08:41,920 --> 00:08:46,120 Speaker 1: you so much. It shows up in my email if 166 00:08:46,160 --> 00:08:48,560 Speaker 1: I want to just kind of press a button and 167 00:08:48,600 --> 00:08:52,679 Speaker 1: not write anymore. Is that an example a rudimentary example 168 00:08:52,880 --> 00:08:54,240 Speaker 1: of AI. 169 00:08:53,920 --> 00:08:54,640 Speaker 4: Yes, very much. 170 00:08:54,640 --> 00:08:54,720 Speaker 2: So. 171 00:08:54,760 --> 00:08:57,200 Speaker 4: There's the math behind it, which is how are you 172 00:08:57,240 --> 00:08:59,360 Speaker 4: going to predict the next word? But the other thing 173 00:08:59,400 --> 00:09:02,000 Speaker 4: about it is the product and sort of the user interface. 174 00:09:02,320 --> 00:09:04,880 Speaker 4: People like to talk about how in the late fifties 175 00:09:05,080 --> 00:09:08,720 Speaker 4: night at Stanford there was John McCarthy who was working 176 00:09:08,720 --> 00:09:11,079 Speaker 4: on the mathematics of AI, and then there were people 177 00:09:11,080 --> 00:09:13,839 Speaker 4: like Doug Engelbart who were working on the product of AI. 178 00:09:14,160 --> 00:09:16,000 Speaker 4: How are we going to make an interface that allows 179 00:09:16,000 --> 00:09:18,440 Speaker 4: people to interact with the computer. Well, so when you 180 00:09:18,520 --> 00:09:20,920 Speaker 4: just saw it, there was a good example of good math. 181 00:09:21,240 --> 00:09:23,040 Speaker 4: And the math could be as simple as counting the 182 00:09:23,120 --> 00:09:25,160 Speaker 4: number of times that the word what follows the word, 183 00:09:25,240 --> 00:09:27,720 Speaker 4: So it's very simple math. But as well as the 184 00:09:27,720 --> 00:09:30,160 Speaker 4: product idea, which is, how do I make a suggestion 185 00:09:30,240 --> 00:09:32,439 Speaker 4: to you in a way that's useful to you while 186 00:09:32,480 --> 00:09:36,000 Speaker 4: you use that digital product and not creepy and not intrusive. 187 00:09:36,360 --> 00:09:39,439 Speaker 4: So yeah, that's another thing that we're seeing with chatchept 188 00:09:40,080 --> 00:09:44,199 Speaker 4: is a good coming together of technology and mathematics and 189 00:09:44,240 --> 00:09:47,080 Speaker 4: statistical models, but also just a nice product that people 190 00:09:47,080 --> 00:09:48,280 Speaker 4: are enjoying musing. 191 00:09:48,280 --> 00:09:51,240 Speaker 2: What you're describing, Chris, is a predictive model. But so 192 00:09:51,280 --> 00:09:54,720 Speaker 2: many people, particularly since a year ago today when chatchipt 193 00:09:55,040 --> 00:09:57,880 Speaker 2: came out and sort of collectively blew the world's minds, 194 00:09:58,520 --> 00:10:02,280 Speaker 2: it feels like we're talking to a machine that is 195 00:10:02,320 --> 00:10:05,240 Speaker 2: actually thinking, that is actually sentient, and it's in fact 196 00:10:05,320 --> 00:10:09,040 Speaker 2: design that way, and that has societal implications, some of 197 00:10:09,080 --> 00:10:11,440 Speaker 2: the societal implications that you were referring to earlier. 198 00:10:11,520 --> 00:10:15,040 Speaker 3: VELAs know, you're pointing out the critical kind of missing 199 00:10:15,080 --> 00:10:17,800 Speaker 3: element in what Chris described as what AI is today. 200 00:10:18,240 --> 00:10:20,319 Speaker 3: At the end of the day, every version of AI 201 00:10:20,360 --> 00:10:23,280 Speaker 3: that we have today. It's a mathematical model that predicts 202 00:10:23,280 --> 00:10:27,560 Speaker 3: what happens next based on what's happened before. It doesn't reason, 203 00:10:27,880 --> 00:10:31,800 Speaker 3: it doesn't think, it doesn't have agency, it doesn't have preferences, 204 00:10:32,480 --> 00:10:34,439 Speaker 3: all of the things that people now try to scare 205 00:10:34,520 --> 00:10:36,240 Speaker 3: us with that. I imagine we'll talk a little bit 206 00:10:36,240 --> 00:10:40,280 Speaker 3: about that crazy term AGI. Today's AI is none of that. 207 00:10:40,760 --> 00:10:42,400 Speaker 3: I often like to say, do you all remember that 208 00:10:42,480 --> 00:10:44,320 Speaker 3: movie Honey, I Stroked the Kids? 209 00:10:44,480 --> 00:10:46,320 Speaker 1: Yeah, Rick Moranez, Yeah. 210 00:10:46,240 --> 00:10:48,840 Speaker 3: Great movie. Right, Today's AI. All it is is this, 211 00:10:49,280 --> 00:10:51,280 Speaker 3: take everything that's ever been written, put it in a 212 00:10:51,320 --> 00:10:54,360 Speaker 3: giant library, beam a rate gun at that library, and 213 00:10:54,480 --> 00:10:57,000 Speaker 3: enough power to power a small city for months and months, 214 00:10:57,080 --> 00:10:59,280 Speaker 3: and say, how do we compress that entire library down? 215 00:10:59,280 --> 00:11:01,800 Speaker 3: And give us one little map? And all the math 216 00:11:01,880 --> 00:11:04,880 Speaker 3: does is it says this. If before, when people said 217 00:11:04,920 --> 00:11:08,400 Speaker 3: a word, they often followed it with another word, that's 218 00:11:08,440 --> 00:11:11,079 Speaker 3: all we're going to do right now. Now, what that does. 219 00:11:11,160 --> 00:11:14,240 Speaker 3: It's an amazing magic trick. It's a great illusion. It 220 00:11:14,320 --> 00:11:16,560 Speaker 3: makes you think you're talking to somebody who wants to 221 00:11:16,559 --> 00:11:18,520 Speaker 3: talk back to you. But at the end of the day, 222 00:11:18,559 --> 00:11:20,800 Speaker 3: all the machine is doing is predicting what the next 223 00:11:20,800 --> 00:11:23,520 Speaker 3: word and the answer should be. This is so critical 224 00:11:24,120 --> 00:11:27,000 Speaker 3: because it reframes how we engage with these tools, and 225 00:11:27,040 --> 00:11:28,960 Speaker 3: that's really all they are. They're just tools. They're not, 226 00:11:29,480 --> 00:11:32,320 Speaker 3: you know, all knowing entities. They're not partners, they're not 227 00:11:32,440 --> 00:11:36,120 Speaker 3: conversational and sparring buddies. They're just tools that help us 228 00:11:36,280 --> 00:11:37,520 Speaker 3: maybe be better. 229 00:11:38,040 --> 00:11:40,880 Speaker 1: Let me ask you this because I thought this was interesting. 230 00:11:41,640 --> 00:11:45,800 Speaker 1: Bill Gates recently noted that AI as it exists today 231 00:11:46,000 --> 00:11:50,120 Speaker 1: is quote still pretty dumb. Chris. Do you agree with that? 232 00:11:50,960 --> 00:11:51,160 Speaker 2: Yeah? 233 00:11:51,160 --> 00:11:54,400 Speaker 4: Absolutely so. I think what VELAs is saying is apt, 234 00:11:54,440 --> 00:11:57,880 Speaker 4: which is language, and the ability to produce language is 235 00:11:57,880 --> 00:12:01,120 Speaker 4: a great imitation of what thinking. And in fact, I 236 00:12:01,240 --> 00:12:04,120 Speaker 4: use the word imitation because in Hellan Turing's original nineteen 237 00:12:04,160 --> 00:12:08,000 Speaker 4: fifty paper on can MA Machines think he basically set 238 00:12:08,000 --> 00:12:11,040 Speaker 4: out this problem. Imagine a computer that could imitate what 239 00:12:11,080 --> 00:12:13,079 Speaker 4: it's like to talk to somebody. That's sort of an 240 00:12:13,080 --> 00:12:16,040 Speaker 4: operaginalization of what it means to think. But there's still 241 00:12:16,040 --> 00:12:20,199 Speaker 4: many things a I can't do. As often described, planning 242 00:12:20,240 --> 00:12:24,920 Speaker 4: is difficult, Compositional thinking is difficult, Working with multiple modes 243 00:12:24,920 --> 00:12:27,640 Speaker 4: at once is difficult. Meaning like words and images together. 244 00:12:27,880 --> 00:12:30,600 Speaker 4: So I think you're right that it's uncanny. Right, we're 245 00:12:30,600 --> 00:12:33,679 Speaker 4: in the uncanny valley of conversations right now with chatbots. 246 00:12:34,040 --> 00:12:37,080 Speaker 4: But it's very difficult for people not to impose this 247 00:12:37,200 --> 00:12:40,880 Speaker 4: belief that it is intelligent or thoughtful. And the truth 248 00:12:41,000 --> 00:12:43,559 Speaker 4: is people have been having that experience for as long 249 00:12:43,600 --> 00:12:46,280 Speaker 4: as they've been building chatbots. Even in the nineteen sixties, 250 00:12:46,520 --> 00:12:49,880 Speaker 4: people were building chatbots based on simple rules, and users 251 00:12:50,160 --> 00:12:52,560 Speaker 4: using that chatbot had the same experience of feeling like 252 00:12:52,600 --> 00:12:54,319 Speaker 4: even though they knew it was just a very simple 253 00:12:54,320 --> 00:12:57,280 Speaker 4: computer program, there was the emotional resonance was one as 254 00:12:57,280 --> 00:12:59,120 Speaker 4: though you were talking to an intelligent agent. 255 00:12:59,760 --> 00:13:03,400 Speaker 1: Hear this word sentient a lot, which of course is 256 00:13:03,480 --> 00:13:07,679 Speaker 1: capable of sensing or feeling conscious of, or responsive to 257 00:13:07,720 --> 00:13:12,360 Speaker 1: the sensations of seeing, hearing, feeling, tasting, or smelling sentient beings, 258 00:13:12,480 --> 00:13:14,760 Speaker 1: which is really what it means to be a human. 259 00:13:15,960 --> 00:13:19,880 Speaker 1: Does AI have the capacity to be sentient? 260 00:13:20,280 --> 00:13:21,760 Speaker 4: I think what we've shown is that it does a 261 00:13:21,760 --> 00:13:23,719 Speaker 4: great imitation of it. But I think it's important for 262 00:13:23,800 --> 00:13:26,080 Speaker 4: us all to remember that it is, as Vela said, 263 00:13:26,360 --> 00:13:29,240 Speaker 4: just math right. It is a mathematical model that spits 264 00:13:29,240 --> 00:13:32,400 Speaker 4: out words and it's optimized for generating words that sound 265 00:13:32,520 --> 00:13:35,200 Speaker 4: like what a human being would say and given the 266 00:13:35,200 --> 00:13:37,640 Speaker 4: same prompt. But we should remember that it is a 267 00:13:37,679 --> 00:13:41,800 Speaker 4: machine and it's executing a mathematical act that we trained 268 00:13:41,800 --> 00:13:42,040 Speaker 4: it on. 269 00:13:42,360 --> 00:13:44,360 Speaker 2: But yet there's a lot of experts out there and 270 00:13:44,440 --> 00:13:47,760 Speaker 2: researchers and some pretty serious people who are trying to 271 00:13:47,800 --> 00:13:52,600 Speaker 2: warn us that these machines may become sentient. Now, is 272 00:13:52,640 --> 00:13:55,080 Speaker 2: that just a matter of seeing too many sci fi 273 00:13:55,120 --> 00:14:00,120 Speaker 2: movies or is that something that is possible obviously not today, 274 00:14:00,160 --> 00:14:01,040 Speaker 2: but on the horizon. 275 00:14:01,320 --> 00:14:03,480 Speaker 4: I think there's a couple of reasons why people are 276 00:14:03,520 --> 00:14:06,280 Speaker 4: warning us of that possibility. I mean, again, part of 277 00:14:06,280 --> 00:14:08,560 Speaker 4: it is wordplay, and I think that's what Alan during 278 00:14:08,640 --> 00:14:10,480 Speaker 4: was getting at in nineteen fifty when he said, can 279 00:14:10,559 --> 00:14:12,959 Speaker 4: machines think? Is an ill posed problem, so let's try 280 00:14:13,000 --> 00:14:15,800 Speaker 4: to operationalize it. But I think the warnings are often 281 00:14:16,160 --> 00:14:21,560 Speaker 4: distractions from real problems that automated inequality and other downsides 282 00:14:21,560 --> 00:14:24,240 Speaker 4: of using algorithms today are causing. In the here and now. 283 00:14:24,400 --> 00:14:28,040 Speaker 4: It's sometimes very difficult to think about our existing challenges 284 00:14:28,040 --> 00:14:32,240 Speaker 4: in sociotechnical systems, and somehow more pleasant to think about 285 00:14:32,280 --> 00:14:35,800 Speaker 4: this terminator doomsday future which is not cured yet. I 286 00:14:35,840 --> 00:14:36,800 Speaker 4: think also there's. 287 00:14:36,640 --> 00:14:39,200 Speaker 1: A yet yet that's scary, that's right. 288 00:14:39,440 --> 00:14:41,680 Speaker 4: I think there's also a concern that people are putting 289 00:14:41,680 --> 00:14:44,360 Speaker 4: forward this idea of a doomsday because there's only a 290 00:14:44,360 --> 00:14:46,960 Speaker 4: small number of companies at present which are able to 291 00:14:47,000 --> 00:14:51,760 Speaker 4: afford amassing lots of data and producing really good products. 292 00:14:52,000 --> 00:14:53,920 Speaker 4: And often these companies are saying, we are the ones 293 00:14:53,920 --> 00:14:56,400 Speaker 4: who can tell you how to regulate this. So there's 294 00:14:56,440 --> 00:14:59,000 Speaker 4: concern that some of the doomsday scenario might be coupled 295 00:14:59,000 --> 00:15:02,720 Speaker 4: to re trey capture or getting ahead of potential regulation 296 00:15:02,840 --> 00:15:04,800 Speaker 4: both domestically and internationally. 297 00:15:05,120 --> 00:15:07,040 Speaker 3: Let's zoom out a little bit. I agree with what 298 00:15:07,160 --> 00:15:09,800 Speaker 3: Chris said, but maybe and you spoke to experts and 299 00:15:09,880 --> 00:15:12,360 Speaker 3: serious people who are trying to scare us all, and 300 00:15:12,400 --> 00:15:14,600 Speaker 3: I just want to put us in context. There are, 301 00:15:14,880 --> 00:15:17,760 Speaker 3: you know, eight billion people on the planet. There are 302 00:15:17,760 --> 00:15:21,440 Speaker 3: maybe a few hundred thousand that really understand how AI works. 303 00:15:21,960 --> 00:15:24,960 Speaker 3: And there's maybe on the order of a few thousand 304 00:15:25,000 --> 00:15:27,480 Speaker 3: people who have decided that what they care about more 305 00:15:27,520 --> 00:15:30,800 Speaker 3: than anything else is the existential risk to humanity. Let's 306 00:15:30,840 --> 00:15:33,360 Speaker 3: just put that in context. There are billions of people 307 00:15:33,360 --> 00:15:35,720 Speaker 3: on the planet today that could use what AI offers 308 00:15:35,760 --> 00:15:39,120 Speaker 3: as a promise to fundamentally change what their lives look like. 309 00:15:39,160 --> 00:15:42,080 Speaker 3: They're access to economic opportunity, to any number of other things. 310 00:15:42,480 --> 00:15:45,640 Speaker 3: There are hundreds of thousands or millions of people who 311 00:15:45,680 --> 00:15:49,479 Speaker 3: work in companies that could fundamentally change their relationship with customers. 312 00:15:49,840 --> 00:15:51,720 Speaker 3: So what do we have. We have a small set 313 00:15:51,720 --> 00:15:54,640 Speaker 3: of people with direct access to some of these tools, 314 00:15:54,640 --> 00:15:56,920 Speaker 3: and we'll talk about how that came about shortly, who 315 00:15:56,960 --> 00:15:58,920 Speaker 3: have come up with this common idea that the thing 316 00:15:58,920 --> 00:16:00,640 Speaker 3: we should care about more than any else is that 317 00:16:00,680 --> 00:16:03,120 Speaker 3: AI will kill us all. And in the meantime, we're 318 00:16:03,160 --> 00:16:05,560 Speaker 3: living in a world where there are so many opportunities 319 00:16:05,560 --> 00:16:07,840 Speaker 3: and challenges here and now that we should be spending 320 00:16:07,840 --> 00:16:08,760 Speaker 3: our time thinking about. 321 00:16:09,400 --> 00:16:12,640 Speaker 2: So there are the here and now, there is the 322 00:16:12,720 --> 00:16:16,960 Speaker 2: long term future, and there's that whole middle ground that 323 00:16:17,000 --> 00:16:19,240 Speaker 2: I think many of us haven't addressed. But before we 324 00:16:19,280 --> 00:16:21,760 Speaker 2: come back to that, just to help our listeners a 325 00:16:21,760 --> 00:16:26,480 Speaker 2: little bit, we hear the term AGI artificial general intelligence, 326 00:16:26,640 --> 00:16:30,600 Speaker 2: not to be confused with generative artificial intelligence. Very confusing 327 00:16:30,800 --> 00:16:33,400 Speaker 2: even for people that pay attention to these things. AGI 328 00:16:33,600 --> 00:16:36,280 Speaker 2: is this sort of robot overlord saying we're talking about 329 00:16:36,280 --> 00:16:38,040 Speaker 2: that may or may not ever happen. Is that right? 330 00:16:38,320 --> 00:16:40,880 Speaker 1: Yeah? What is that? You guys help me out. I'm 331 00:16:40,920 --> 00:16:41,760 Speaker 1: the dumb one here. 332 00:16:42,320 --> 00:16:44,280 Speaker 4: I think one thing that's useful is to remember the 333 00:16:44,320 --> 00:16:48,720 Speaker 4: two different g's in those two acronyms. The GAI is generitive, 334 00:16:48,760 --> 00:16:51,960 Speaker 4: but AGI is general. So when people talk about AGI 335 00:16:52,120 --> 00:16:55,640 Speaker 4: for general intelligence, part of which is exciting is the 336 00:16:55,680 --> 00:16:59,640 Speaker 4: idea that in the last fifty years we've done species, 337 00:17:00,120 --> 00:17:02,480 Speaker 4: not me personally, but the human being species have done 338 00:17:02,480 --> 00:17:04,479 Speaker 4: a really good job building algorithms that are good for 339 00:17:04,520 --> 00:17:07,560 Speaker 4: individual tasks. Like you can build an app that can 340 00:17:07,880 --> 00:17:09,720 Speaker 4: take a picture and say does this have a dog 341 00:17:09,760 --> 00:17:11,640 Speaker 4: face in it? Or a cat face in it? That's 342 00:17:11,680 --> 00:17:14,560 Speaker 4: a specific use of statistical modeling, which is good for 343 00:17:14,600 --> 00:17:18,080 Speaker 4: that specific use case. So the dream of AGI is 344 00:17:18,119 --> 00:17:21,399 Speaker 4: that you can produce one algorithm, one machine, one model 345 00:17:21,680 --> 00:17:25,360 Speaker 4: that's good not only for disinbiguating cat faces from dog faces, 346 00:17:25,359 --> 00:17:29,679 Speaker 4: but also composing a sonnet or enjoying strawberries and cream, 347 00:17:29,960 --> 00:17:33,639 Speaker 4: or whatever general problem you would like the machine to solve. 348 00:17:33,800 --> 00:17:36,600 Speaker 4: So that's the g of agis general. It's very easy 349 00:17:36,600 --> 00:17:38,880 Speaker 4: for us to make machine learning models that are good 350 00:17:38,880 --> 00:17:41,199 Speaker 4: for one specific task. It's much harder to make a 351 00:17:41,240 --> 00:17:43,800 Speaker 4: machine learning model that's general and is able to do 352 00:17:43,920 --> 00:17:46,359 Speaker 4: anything that we consider an intelligent task. 353 00:17:46,840 --> 00:17:50,040 Speaker 1: So where are we in terms of And I didn't 354 00:17:50,080 --> 00:17:54,080 Speaker 1: really understand that explanation, Chris, can you try it again 355 00:17:54,440 --> 00:17:57,439 Speaker 1: in a more like I'm not a Columbia student, or 356 00:17:57,760 --> 00:18:01,320 Speaker 1: just pretend like I'm in sixth grade? Help me out, sure, 357 00:18:01,400 --> 00:18:02,360 Speaker 1: help me out. 358 00:18:02,840 --> 00:18:07,960 Speaker 4: So for many decades, we've been able to build specific 359 00:18:08,119 --> 00:18:12,440 Speaker 4: machine learning models. So a specific algorithm that can tell 360 00:18:12,480 --> 00:18:15,480 Speaker 4: the difference between a picture of a dog and a 361 00:18:15,520 --> 00:18:17,919 Speaker 4: picture of a cat, say, that's a specific problem. And 362 00:18:18,000 --> 00:18:21,119 Speaker 4: we've been very good for decades at building algorithms that 363 00:18:21,160 --> 00:18:24,439 Speaker 4: can do very specific and focused tasks. One of the 364 00:18:24,480 --> 00:18:27,479 Speaker 4: things that we've seen with chatbots that are trained on 365 00:18:27,560 --> 00:18:30,600 Speaker 4: a wide variety of documents is that you can have 366 00:18:30,680 --> 00:18:34,160 Speaker 4: a plausible conversation with a chatbot about a wide variety 367 00:18:34,200 --> 00:18:37,320 Speaker 4: of topics. So if you've trained a chatbot only on 368 00:18:37,400 --> 00:18:40,840 Speaker 4: chemistry textbooks, you will have a great conversation about chemistry 369 00:18:40,880 --> 00:18:44,280 Speaker 4: and not about any other subject. But by training a 370 00:18:44,359 --> 00:18:48,200 Speaker 4: chatbot on a wide variety of topics chemistry, philosophy, and 371 00:18:48,240 --> 00:18:52,240 Speaker 4: all points in between, people are experiencing this shock that 372 00:18:52,320 --> 00:18:56,280 Speaker 4: you can interact with an AI, meaning an algorithm that 373 00:18:56,680 --> 00:19:01,359 Speaker 4: works not only solving a specific problem, solving a general problem, 374 00:19:01,440 --> 00:19:04,479 Speaker 4: in this case, the general problem of having having an 375 00:19:04,520 --> 00:19:08,840 Speaker 4: intelligent sounding conversation about a general breath of topics. 376 00:19:10,680 --> 00:19:13,040 Speaker 1: After a quick break, we'll be back with my co 377 00:19:13,320 --> 00:19:17,080 Speaker 1: pilot and plus one Vivian Shiller, talking to Chris Wiggins 378 00:19:17,200 --> 00:19:22,200 Speaker 1: and velost Star. If you want to get smarter every 379 00:19:22,240 --> 00:19:25,320 Speaker 1: morning with a breakdown of the news and fascinating takes 380 00:19:25,359 --> 00:19:28,480 Speaker 1: on health and wellness and pop culture, sign up for 381 00:19:28,520 --> 00:19:31,640 Speaker 1: our daily newsletter, Wake Up Call by going to Katiecuric 382 00:19:31,720 --> 00:19:38,360 Speaker 1: dot com. We're back with Chris Wiggins and velost Star, 383 00:19:38,520 --> 00:19:43,119 Speaker 1: along with my plus one Vivian Schiller. Have you guys 384 00:19:43,240 --> 00:19:47,160 Speaker 1: used chat GPT or bard. Have you tried to have 385 00:19:47,240 --> 00:19:50,480 Speaker 1: it write speeches for you or come up with any 386 00:19:50,600 --> 00:19:54,320 Speaker 1: kind of documents. I'm sure you've tested it, Vivian, what 387 00:19:54,359 --> 00:19:55,720 Speaker 1: has your experience been like. 388 00:19:56,200 --> 00:20:00,640 Speaker 2: I've used chat gpt to develop an itinerary. I took 389 00:20:00,680 --> 00:20:02,720 Speaker 2: it to Japan. I knew I needed I had some 390 00:20:02,840 --> 00:20:05,840 Speaker 2: time between two places I needed to be, and I 391 00:20:05,880 --> 00:20:08,280 Speaker 2: had certain things that I was interested in, certain things 392 00:20:08,320 --> 00:20:10,960 Speaker 2: I was less interested in, didn't know how long it 393 00:20:11,000 --> 00:20:13,440 Speaker 2: took to get paced the place, and actually chat Gipt 394 00:20:13,600 --> 00:20:15,919 Speaker 2: gave me an amazing itinerary, so it was very useful. 395 00:20:16,119 --> 00:20:18,879 Speaker 1: Travel agents probably don't like that, how about you, guys? 396 00:20:19,480 --> 00:20:21,840 Speaker 3: Yeah, you know, I mean, I've used every LM out there, 397 00:20:21,880 --> 00:20:23,199 Speaker 3: and so I'm like, I'm no longer, you know. I 398 00:20:23,200 --> 00:20:25,080 Speaker 3: wish I could say it was at the emergent frontier 399 00:20:25,080 --> 00:20:26,679 Speaker 3: of AI, but I'm no longer Now I have a 400 00:20:26,680 --> 00:20:28,560 Speaker 3: different role. But I spent a lot of time with 401 00:20:28,600 --> 00:20:30,240 Speaker 3: the smartest people that are working on this stuff, and 402 00:20:30,240 --> 00:20:32,639 Speaker 3: I've used them all. I've used them to do really 403 00:20:32,680 --> 00:20:36,200 Speaker 3: basic and pedantic things like oh, give me some talking points. 404 00:20:36,400 --> 00:20:38,080 Speaker 3: I no longer do that, having tried it a few 405 00:20:38,119 --> 00:20:40,359 Speaker 3: times and realizing how bad it is. I spent a 406 00:20:40,400 --> 00:20:42,840 Speaker 3: lot of time using these generative AI tools my nieces 407 00:20:42,840 --> 00:20:45,600 Speaker 3: and nephews. I'm doing really fun things like saying, hey, 408 00:20:45,680 --> 00:20:48,560 Speaker 3: let's come up with married a scene and let's see 409 00:20:48,560 --> 00:20:50,560 Speaker 3: if we can get an EI to draw it for us, 410 00:20:50,840 --> 00:20:52,600 Speaker 3: and then ask the question, hey, is this kind of 411 00:20:52,600 --> 00:20:54,639 Speaker 3: what you pictured in your mind? Side? How do we 412 00:20:54,720 --> 00:20:57,200 Speaker 3: make it better? And we actually iterate with genitive AI 413 00:20:57,280 --> 00:21:00,840 Speaker 3: to create new artworks or even basic things sometimes like Hey, 414 00:21:00,920 --> 00:21:02,919 Speaker 3: I want to tell you a bedtime story, what do 415 00:21:02,920 --> 00:21:05,040 Speaker 3: you want it to be about? And then we work 416 00:21:05,119 --> 00:21:06,359 Speaker 3: with an AI to kind of come up with a 417 00:21:06,440 --> 00:21:09,200 Speaker 3: nice little Kate. Look, these are all really fun, But again, 418 00:21:09,200 --> 00:21:10,679 Speaker 3: I want to make sure that we understand that we're 419 00:21:10,760 --> 00:21:12,920 Speaker 3: kind of missing the point a little bit, right. These 420 00:21:12,960 --> 00:21:15,639 Speaker 3: tools that have changed our lives, and Vivian said, have 421 00:21:15,800 --> 00:21:18,600 Speaker 3: done so really in an amazing way, but not because 422 00:21:18,680 --> 00:21:21,720 Speaker 3: the technology has already changed our lives, but because it's 423 00:21:21,760 --> 00:21:24,680 Speaker 3: opened our eyes to what's possible when these tools get 424 00:21:24,720 --> 00:21:28,040 Speaker 3: to be really amazing. We talk all the time about 425 00:21:28,040 --> 00:21:30,560 Speaker 3: how these tools have hallucinations, right, the idea that you 426 00:21:30,640 --> 00:21:33,120 Speaker 3: might ask it a question and it doesn't check whether 427 00:21:33,119 --> 00:21:34,760 Speaker 3: the answer is real or not. It just kind of 428 00:21:34,760 --> 00:21:37,320 Speaker 3: spews some language back at you and you say, okay, 429 00:21:37,359 --> 00:21:40,240 Speaker 3: well that sounds reasonable and you move on. The tools 430 00:21:40,240 --> 00:21:42,920 Speaker 3: that we have today aren't products yet. There's still kind 431 00:21:42,920 --> 00:21:45,360 Speaker 3: of the very early days of what generative AI will 432 00:21:45,359 --> 00:21:47,639 Speaker 3: look like. And my hope is when we start training 433 00:21:47,680 --> 00:21:50,600 Speaker 3: these models on medical data that includes all of the 434 00:21:50,640 --> 00:21:53,359 Speaker 3: kind of published medical literature, we'll get to a much 435 00:21:53,400 --> 00:21:55,520 Speaker 3: better sense of what a generative AI can do to 436 00:21:55,520 --> 00:21:57,879 Speaker 3: help the doctor diagnose it. But at the end of 437 00:21:57,920 --> 00:22:00,680 Speaker 3: the day, I can't imagine a world in which we say, 438 00:22:00,720 --> 00:22:03,879 Speaker 3: the genitor of AIS we have today are directly diagnosing 439 00:22:03,960 --> 00:22:06,600 Speaker 3: a patient. The only thing they can do is help 440 00:22:06,680 --> 00:22:10,040 Speaker 3: a doctor or a medical professional whose trained use it 441 00:22:10,040 --> 00:22:12,320 Speaker 3: as an input into their process to figure out what's 442 00:22:12,359 --> 00:22:14,720 Speaker 3: going on. And that's the moment that we're stuck in 443 00:22:14,800 --> 00:22:16,199 Speaker 3: right now, because I know so many of us want 444 00:22:16,240 --> 00:22:18,240 Speaker 3: to jump into a future where we say that AIS 445 00:22:18,240 --> 00:22:20,800 Speaker 3: are going to do everything for us, but we're really 446 00:22:20,880 --> 00:22:22,560 Speaker 3: in a moment where we're saying, the only way this 447 00:22:22,640 --> 00:22:25,639 Speaker 3: works is that the AIS support human decision makers. They 448 00:22:25,640 --> 00:22:28,760 Speaker 3: can use what the technology gives them, but their own experience, 449 00:22:28,800 --> 00:22:31,560 Speaker 3: their lived wisdom. They're you know, working with patients for 450 00:22:31,600 --> 00:22:33,600 Speaker 3: hour many years to actually make a decision. 451 00:22:33,880 --> 00:22:34,080 Speaker 2: You know. 452 00:22:34,200 --> 00:22:36,760 Speaker 1: I tried to get chat ept to write a poem 453 00:22:36,800 --> 00:22:41,760 Speaker 1: for my husband's birthday and it was very, honestly not 454 00:22:42,000 --> 00:22:46,480 Speaker 1: very good. I gave it information about my husband, but 455 00:22:46,600 --> 00:22:49,879 Speaker 1: it was quite pedantic and not very clever. It was 456 00:22:50,040 --> 00:22:54,520 Speaker 1: sort of honestly Hallmark CARDI quality. And I think it's 457 00:22:54,560 --> 00:22:58,200 Speaker 1: because it didn't have the breadth of knowledge about him 458 00:22:58,760 --> 00:23:02,440 Speaker 1: that I do so so it couldn't really compete with that, 459 00:23:02,880 --> 00:23:05,760 Speaker 1: but it was fun to try it. And another example, 460 00:23:06,119 --> 00:23:09,320 Speaker 1: when I interviewed Carl Rove at the Aspen Ideas Festival, 461 00:23:09,600 --> 00:23:12,280 Speaker 1: I was trying to come up with a fun title 462 00:23:12,359 --> 00:23:16,640 Speaker 1: for the conversation. I asked chat GPT and it came 463 00:23:16,760 --> 00:23:19,720 Speaker 1: up with a great title, which was the Elephant in 464 00:23:19,800 --> 00:23:22,520 Speaker 1: the Room because it was on the future of the 465 00:23:22,560 --> 00:23:27,000 Speaker 1: Republican Party. And I was like, that is genius. So, 466 00:23:27,520 --> 00:23:29,760 Speaker 1: you know, I think you're right what you were saying, 467 00:23:29,800 --> 00:23:34,280 Speaker 1: Velss about it being helpful but not determinative. And one 468 00:23:34,320 --> 00:23:37,720 Speaker 1: example is, you know, I'm very into cancer screening and 469 00:23:37,800 --> 00:23:39,760 Speaker 1: some of the things that they're going to be able 470 00:23:39,800 --> 00:23:44,399 Speaker 1: to do that is beyond the ability of a human 471 00:23:44,520 --> 00:23:47,760 Speaker 1: to see things, is to take these massive data sets 472 00:23:48,400 --> 00:23:53,600 Speaker 1: and look at scans and figure out actually predict if 473 00:23:53,640 --> 00:23:56,639 Speaker 1: someone may or may not get breast cancer in the 474 00:23:56,680 --> 00:24:00,480 Speaker 1: next five years. I mean, that really blows my mind. 475 00:24:01,000 --> 00:24:04,520 Speaker 1: But that obviously has to be done in conjunction with 476 00:24:04,600 --> 00:24:08,080 Speaker 1: an experienced medical professional. Right, So is that what you 477 00:24:08,160 --> 00:24:10,240 Speaker 1: mean veloc by kind of being an aid? 478 00:24:10,920 --> 00:24:13,800 Speaker 3: It is, And let me add a little bit of nuance, o, Katie. 479 00:24:13,920 --> 00:24:16,200 Speaker 3: I mean, you've been such a courageous kind of leader 480 00:24:16,240 --> 00:24:19,040 Speaker 3: on this topic. When we started looking at breast cancer 481 00:24:19,040 --> 00:24:21,399 Speaker 3: in particular with AI through our lens as a civil 482 00:24:21,440 --> 00:24:25,080 Speaker 3: society institution, we learned about this fundamental problem that's just 483 00:24:25,119 --> 00:24:27,400 Speaker 3: going to kind of blow your hairback. We have all 484 00:24:27,400 --> 00:24:29,800 Speaker 3: these algorithms today that have been trained to do exactly 485 00:24:29,800 --> 00:24:32,480 Speaker 3: what you described to take a mammogram or a scan 486 00:24:33,000 --> 00:24:35,840 Speaker 3: and say hey, can we do early prediction of cancer risk? 487 00:24:36,359 --> 00:24:38,760 Speaker 3: But all of these tools we learned very quickly have 488 00:24:38,840 --> 00:24:42,159 Speaker 3: been trained on global north populations. They've been trained on 489 00:24:42,200 --> 00:24:45,639 Speaker 3: American data and European data, and so when an organization 490 00:24:45,880 --> 00:24:48,440 Speaker 3: like Instituto Protea, which is a partner of ours, took 491 00:24:48,440 --> 00:24:51,040 Speaker 3: these to Brazil and tried to use them on low 492 00:24:51,040 --> 00:24:53,760 Speaker 3: cost machines that were already recent settings, they found the 493 00:24:53,760 --> 00:24:57,200 Speaker 3: algorithms didn't work at all. So even in that aspirational 494 00:24:57,200 --> 00:24:59,119 Speaker 3: moment that you've created this idea that we might have 495 00:24:59,200 --> 00:25:01,960 Speaker 3: this massive break through, we come back to a very 496 00:25:02,160 --> 00:25:06,000 Speaker 3: human kind of fundamental problem that until we train this 497 00:25:06,160 --> 00:25:08,560 Speaker 3: data on ways that are representative about all of the 498 00:25:08,560 --> 00:25:10,520 Speaker 3: people in the world, not just those who have privileged 499 00:25:10,560 --> 00:25:13,800 Speaker 3: access to Western medicine, right, we're never going to realize 500 00:25:13,840 --> 00:25:15,840 Speaker 3: the promise. You talked about. 501 00:25:15,440 --> 00:25:20,800 Speaker 1: How biased is AI? How biased are these large language models, 502 00:25:20,920 --> 00:25:25,120 Speaker 1: because I remember doing a documentary on our tech addiction, 503 00:25:25,440 --> 00:25:28,359 Speaker 1: and this was just starting to be talked about, and 504 00:25:28,400 --> 00:25:31,480 Speaker 1: I think this was like in twenty eighteen, Chris, do 505 00:25:31,520 --> 00:25:34,200 Speaker 1: you see this as a major problem that it doesn't 506 00:25:34,280 --> 00:25:38,120 Speaker 1: really represent people like so much in society? 507 00:25:38,440 --> 00:25:40,760 Speaker 4: I mean, the problem is always how something is used 508 00:25:40,880 --> 00:25:43,560 Speaker 4: or interpreted. I would say in the context of medical 509 00:25:43,680 --> 00:25:48,359 Speaker 4: usage of AI, there's additional challenges around responsibility or attribution 510 00:25:48,720 --> 00:25:51,879 Speaker 4: and decision making. So I think for all of these tools, 511 00:25:52,320 --> 00:25:55,200 Speaker 4: they're going through this very inefficient part of our hype cycle. 512 00:25:55,320 --> 00:25:57,439 Speaker 4: So in a hype cycle, there's a moment where you 513 00:25:57,480 --> 00:25:59,840 Speaker 4: discover a technology and you have this moment of irrationally 514 00:26:00,040 --> 00:26:01,960 Speaker 4: zuprints and you think it's going to be great, and 515 00:26:02,000 --> 00:26:04,119 Speaker 4: then you have some trough of despair as you realize 516 00:26:04,119 --> 00:26:06,359 Speaker 4: it's actually not that good about generating a poem about 517 00:26:06,400 --> 00:26:08,600 Speaker 4: your husband in your case, And then we get to 518 00:26:08,640 --> 00:26:10,760 Speaker 4: some efficient place where we all have an understanding of 519 00:26:10,760 --> 00:26:14,280 Speaker 4: what these technologies can do and cannot do. So I 520 00:26:14,280 --> 00:26:16,280 Speaker 4: do think we all need to limit our trust in 521 00:26:16,280 --> 00:26:19,560 Speaker 4: all these technologies in in that'stry for technology in general. 522 00:26:19,560 --> 00:26:21,160 Speaker 4: But I think Veloso is making a good, great point, 523 00:26:21,160 --> 00:26:24,000 Speaker 4: which is form machine learning in general, which again is 524 00:26:24,040 --> 00:26:27,399 Speaker 4: the strategy that actually works for artificial intelligence. Where you 525 00:26:27,440 --> 00:26:30,720 Speaker 4: train an algorithm on lots of data, it is extremely biased, 526 00:26:30,720 --> 00:26:33,160 Speaker 4: and this is that it's well suited to the data 527 00:26:33,160 --> 00:26:35,879 Speaker 4: st you have, and there are many complex problems in 528 00:26:35,920 --> 00:26:38,080 Speaker 4: the world where when you train it on one data set, 529 00:26:38,200 --> 00:26:41,399 Speaker 4: it will not generalize to some other very different data set. 530 00:26:41,560 --> 00:26:43,760 Speaker 4: And the different data set could be you've trained a 531 00:26:43,840 --> 00:26:46,040 Speaker 4: language model in chemistry and then you try to test 532 00:26:46,040 --> 00:26:48,240 Speaker 4: it on poetry, or it could be that you've trained 533 00:26:48,280 --> 00:26:51,359 Speaker 4: it on genetic information from one demographic group and then 534 00:26:51,440 --> 00:26:54,880 Speaker 4: you realize it says nothing about, say, predicting phenotype from 535 00:26:54,880 --> 00:26:58,240 Speaker 4: genotype for a different demographic group. That is a real problem. 536 00:26:58,320 --> 00:27:01,199 Speaker 4: It often undergoes the name of buis, but in the 537 00:27:01,240 --> 00:27:04,200 Speaker 4: case of machine learning, it's built into the system. If 538 00:27:04,240 --> 00:27:06,119 Speaker 4: you train it on one data set, you're going to 539 00:27:06,200 --> 00:27:07,840 Speaker 4: have a bad time if you try to use it 540 00:27:07,840 --> 00:27:09,000 Speaker 4: on a very different data set. 541 00:27:09,680 --> 00:27:11,520 Speaker 2: Let me follow up with that to both the Loss 542 00:27:11,560 --> 00:27:14,239 Speaker 2: and Chris, which is how much of that has to 543 00:27:14,280 --> 00:27:18,439 Speaker 2: do with the people who are selecting the data sets, 544 00:27:18,560 --> 00:27:22,600 Speaker 2: who are creating the technology, who are deploying technology, most 545 00:27:22,640 --> 00:27:27,080 Speaker 2: of whom are in Silicon Valley. Are they maybe in 546 00:27:27,119 --> 00:27:31,000 Speaker 2: a few centralized companies. How much of that is an issue? 547 00:27:31,040 --> 00:27:32,760 Speaker 2: And how do we get out of that jam? 548 00:27:32,840 --> 00:27:34,320 Speaker 3: Yeah, Vivian, I'm going to take that. I'm going to 549 00:27:34,359 --> 00:27:36,000 Speaker 3: go one step bigger. I'm going to give you an 550 00:27:36,000 --> 00:27:38,960 Speaker 3: example for it. We talk a lot about you've probably 551 00:27:39,040 --> 00:27:42,000 Speaker 3: heard about hiring algorithms, about how companies are using AI 552 00:27:42,040 --> 00:27:44,639 Speaker 3: to screen resumes about who they want to hire. And 553 00:27:44,680 --> 00:27:47,159 Speaker 3: there's a story that's been well told there about the 554 00:27:47,200 --> 00:27:49,760 Speaker 3: fact that these algorithms are often biased, they often pick 555 00:27:50,280 --> 00:27:54,600 Speaker 3: men over women particular types of technical competency. That's one story, 556 00:27:54,640 --> 00:27:57,160 Speaker 3: and we get it. But there's a bigger story here 557 00:27:57,160 --> 00:27:59,960 Speaker 3: that we have a really hard time engaging with those algori. 558 00:28:00,320 --> 00:28:03,560 Speaker 3: We're trained on twenty years of data about how human 559 00:28:03,640 --> 00:28:07,439 Speaker 3: recruiters picked candidates, and yet we never talk about the 560 00:28:07,440 --> 00:28:09,760 Speaker 3: fact that for twenty years we've lived in a world 561 00:28:09,800 --> 00:28:12,800 Speaker 3: where our own recruiters are showing these biases day in 562 00:28:12,880 --> 00:28:15,560 Speaker 3: and day out. The question we should be asking is 563 00:28:15,600 --> 00:28:18,280 Speaker 3: not why is the algorithm biased? It's why is a 564 00:28:18,359 --> 00:28:20,800 Speaker 3: society have we been so okay for twenty years with 565 00:28:20,840 --> 00:28:22,800 Speaker 3: this set of outcomes, and now that we have a 566 00:28:22,840 --> 00:28:25,880 Speaker 3: tool that shows us just how bias we've been, we're 567 00:28:25,880 --> 00:28:29,040 Speaker 3: not having a public conversation about restructuring our entire hiring 568 00:28:29,080 --> 00:28:32,640 Speaker 3: mechanism across the private sect. This is just one analog 569 00:28:32,720 --> 00:28:34,240 Speaker 3: of a lot of things like this that I think 570 00:28:34,280 --> 00:28:37,800 Speaker 3: are emerging across the board, where AI, because of the 571 00:28:37,840 --> 00:28:40,880 Speaker 3: bias in the algorithm, is putting a spotlight on the 572 00:28:40,920 --> 00:28:44,240 Speaker 3: bias in our human behavior. We should be using AI 573 00:28:44,240 --> 00:28:47,120 Speaker 3: as an investigative tool, as a magnifying glass that lets 574 00:28:47,160 --> 00:28:49,480 Speaker 3: us look at all kinds of decisions and say, how 575 00:28:49,480 --> 00:28:52,880 Speaker 3: do we build a more just and equitable society. Let's 576 00:28:53,160 --> 00:28:55,960 Speaker 3: have a conversation with a bias in AI. We absolutely should, 577 00:28:55,960 --> 00:28:57,280 Speaker 3: and the answer to that, we kind of know what 578 00:28:57,320 --> 00:29:00,960 Speaker 3: the answer is, right. More representative data, presentive talent that 579 00:29:01,040 --> 00:29:04,400 Speaker 3: designs these algorithms, making sure there's public compute that allows 580 00:29:04,400 --> 00:29:07,920 Speaker 3: these people to develop products. But let's take the bigger 581 00:29:08,040 --> 00:29:10,160 Speaker 3: picture here. This is what we're going into over the 582 00:29:10,160 --> 00:29:13,040 Speaker 3: next twenty years is a world in which these tools 583 00:29:13,040 --> 00:29:16,440 Speaker 3: demonstrate to us why we're okay with the society we've built, 584 00:29:17,040 --> 00:29:18,959 Speaker 3: and let us question if we actually want to make 585 00:29:19,000 --> 00:29:20,400 Speaker 3: some fundamental changes in them. 586 00:29:20,720 --> 00:29:24,120 Speaker 1: But maybe AI can be an instrument for change for losso. 587 00:29:24,160 --> 00:29:27,360 Speaker 1: I mean, you know that is such a massive undertaking 588 00:29:27,520 --> 00:29:32,040 Speaker 1: to uproot bias in society. I mean, it's so baked in, 589 00:29:32,360 --> 00:29:36,080 Speaker 1: so maybe this is one entry way to address it. 590 00:29:36,480 --> 00:29:39,200 Speaker 3: Absolutely, It's one of the things I'm most optimistic about 591 00:29:39,280 --> 00:29:41,600 Speaker 3: right is when we look at things like we're going 592 00:29:41,640 --> 00:29:43,400 Speaker 3: to have a conversation I'm sure here about some of 593 00:29:43,440 --> 00:29:46,280 Speaker 3: the recent developments in the AI world, one of which 594 00:29:46,280 --> 00:29:50,600 Speaker 3: has just been the continued silencing of women underrepresented characters 595 00:29:50,640 --> 00:29:54,080 Speaker 3: in building these tools. I'm deeply optimistic about the fact 596 00:29:54,120 --> 00:29:56,480 Speaker 3: that we could invest in creating a new capacity to 597 00:29:56,520 --> 00:29:59,680 Speaker 3: build AI that's really representative a lot of those problems 598 00:29:59,720 --> 00:30:02,400 Speaker 3: would one we signed a spotlight on them, and two 599 00:30:02,400 --> 00:30:04,240 Speaker 3: we'd very quickly move to fix them. 600 00:30:04,760 --> 00:30:07,240 Speaker 1: Well, when we come back, we're going to talk about 601 00:30:07,320 --> 00:30:11,880 Speaker 1: how do you regulate artificial intelligence, What in the world 602 00:30:12,040 --> 00:30:16,680 Speaker 1: is going on with Sam Altman and open AI, and 603 00:30:17,640 --> 00:30:22,240 Speaker 1: how quickly is this technology going to evolve. That's right 604 00:30:22,280 --> 00:30:28,520 Speaker 1: after this, I want to tell you all about the 605 00:30:28,560 --> 00:30:32,440 Speaker 1: Cancer Straight Talk podcast for Memorial Sloan Cattering Cancer Center 606 00:30:32,520 --> 00:30:36,880 Speaker 1: with MSK oncologist doctor Diane Reedy Lagunis. I was a 607 00:30:36,920 --> 00:30:39,640 Speaker 1: guest and we had a totally candid conversation about my 608 00:30:39,760 --> 00:30:44,520 Speaker 1: family's experiences with cancer, including my husband's illness, my own 609 00:30:44,560 --> 00:30:47,280 Speaker 1: treatment for breast cancer, and of course that time I 610 00:30:47,320 --> 00:30:51,240 Speaker 1: got a colonoscopy. On TV, Cancer straight Talk features life 611 00:30:51,280 --> 00:30:56,240 Speaker 1: affirming conversations with experts and patients alike about topics affecting 612 00:30:56,360 --> 00:30:59,600 Speaker 1: everyone touched by cancer. If that includes you, I hope 613 00:30:59,600 --> 00:31:03,240 Speaker 1: you'll listen into my episode and every episode of Cancer 614 00:31:03,280 --> 00:31:11,400 Speaker 1: Straight Talk. We're back with Chris Wiggins and Velos star 615 00:31:11,560 --> 00:31:16,120 Speaker 1: along with my plus one Vivian Shiller. Chris is the 616 00:31:16,240 --> 00:31:19,440 Speaker 1: Chief Data Science of The New York Times, Associate Professor 617 00:31:19,480 --> 00:31:23,680 Speaker 1: of Applied Mathematics and Systems Biology at Columbia, and he 618 00:31:23,720 --> 00:31:26,920 Speaker 1: wrote the book How Data Happened, A History from the 619 00:31:26,960 --> 00:31:30,640 Speaker 1: Age of Reason to the Age of Algorithms, which frankly 620 00:31:30,760 --> 00:31:34,720 Speaker 1: I read in two days, just kidding. Chris the Loss 621 00:31:35,160 --> 00:31:38,600 Speaker 1: is President and trustee of the Patrick J. McGovern Foundation, 622 00:31:38,840 --> 00:31:42,520 Speaker 1: which focuses on AI and data solutions. And my plus 623 00:31:42,520 --> 00:31:46,680 Speaker 1: one today is my good friend Vivian Schiller, who has 624 00:31:46,760 --> 00:31:50,640 Speaker 1: worked in many media organizations and has really dug into 625 00:31:50,800 --> 00:31:55,840 Speaker 1: AI and technology, media and society. So you gave me 626 00:31:55,920 --> 00:31:59,480 Speaker 1: the perfect segue the loss in our last conversation before 627 00:31:59,520 --> 00:32:03,560 Speaker 1: the break, and that was what is happening right now 628 00:32:04,040 --> 00:32:08,680 Speaker 1: in various technology companies. So, Chris and velos, who wants 629 00:32:08,720 --> 00:32:13,120 Speaker 1: to kind of explain this Sam Altman drama which is 630 00:32:13,200 --> 00:32:17,840 Speaker 1: being watched with baited breath by everyone in technology and 631 00:32:17,960 --> 00:32:20,400 Speaker 1: I think in media right now. Chris, you want to 632 00:32:20,440 --> 00:32:21,440 Speaker 1: give it a shot, I. 633 00:32:21,360 --> 00:32:23,640 Speaker 4: Can try, with the warning that you know, we're all 634 00:32:23,680 --> 00:32:26,320 Speaker 4: outside the company and so all of it is speculative. 635 00:32:26,440 --> 00:32:28,240 Speaker 4: You know, there's a set of about four people who 636 00:32:28,280 --> 00:32:30,320 Speaker 4: really know what happened. There are the people who were 637 00:32:30,320 --> 00:32:33,520 Speaker 4: on the board that were voting to oust the CEO, 638 00:32:34,120 --> 00:32:36,360 Speaker 4: and so there's a very small number of people who 639 00:32:36,400 --> 00:32:38,560 Speaker 4: really were in the room when it happened and can 640 00:32:38,600 --> 00:32:38,959 Speaker 4: tell us. 641 00:32:39,040 --> 00:32:42,160 Speaker 1: Having said that, Chris, though there's been some pretty strong 642 00:32:42,240 --> 00:32:45,560 Speaker 1: reporting on it that I've read, and let me try 643 00:32:45,560 --> 00:32:47,960 Speaker 1: to set it up if I could. So, Sam Altman, 644 00:32:48,160 --> 00:32:52,440 Speaker 1: this young genius head of open Ai, who I think 645 00:32:52,520 --> 00:32:55,840 Speaker 1: is very well liked by the press, considered obviously a 646 00:32:55,880 --> 00:32:59,400 Speaker 1: real leader in the field, was the CEO of open Ai. 647 00:33:00,320 --> 00:33:03,600 Speaker 1: Two members of the board who were very concerned that 648 00:33:04,040 --> 00:33:08,880 Speaker 1: the business model was superseding the ethical considerations of AI. 649 00:33:09,320 --> 00:33:12,400 Speaker 1: Is my understanding. Okay, Vivian, you look like you want 650 00:33:12,400 --> 00:33:13,720 Speaker 1: to add something, Is that right? 651 00:33:14,240 --> 00:33:17,640 Speaker 2: Uh? Well, all they have said publicly, and I think 652 00:33:17,640 --> 00:33:18,960 Speaker 2: I've seen a lot of that there has been some 653 00:33:19,040 --> 00:33:24,040 Speaker 2: fantastic reporting, is that sam Aldman was not communicating in 654 00:33:24,080 --> 00:33:25,800 Speaker 2: a way that made the boy I forget the wordy exactly, 655 00:33:25,880 --> 00:33:27,920 Speaker 2: but communicating to the board in a way that made 656 00:33:27,960 --> 00:33:31,240 Speaker 2: them feel comfortable. They didn't specifically say they were worried 657 00:33:31,760 --> 00:33:35,040 Speaker 2: the AI was getting out ahead of his skis. I 658 00:33:35,080 --> 00:33:38,080 Speaker 2: think there's one other interesting twist in all of this, 659 00:33:38,200 --> 00:33:40,840 Speaker 2: which is not to get too technical, but the structure 660 00:33:40,880 --> 00:33:42,240 Speaker 2: of open AI is fascinating. 661 00:33:42,320 --> 00:33:44,920 Speaker 1: Well, it's really important, I think to mention that. 662 00:33:45,080 --> 00:33:47,920 Speaker 2: Yeah, it's a not for profit organization of five oh 663 00:33:47,960 --> 00:33:50,800 Speaker 2: one C three, which, as someone that has been part 664 00:33:50,840 --> 00:33:52,760 Speaker 2: of and led five O one C three, has very 665 00:33:52,840 --> 00:33:55,360 Speaker 2: specific governance. They have a governance to a mission, a 666 00:33:55,480 --> 00:33:58,720 Speaker 2: stated mission that is part of how the organization is 667 00:33:58,760 --> 00:33:59,200 Speaker 2: set up. 668 00:33:59,440 --> 00:34:02,360 Speaker 1: Let me just say interject that their work should benefit 669 00:34:02,440 --> 00:34:05,880 Speaker 1: quote unquote humanity as a whole exactly. 670 00:34:05,960 --> 00:34:08,400 Speaker 2: So a not for profit organization is not there to 671 00:34:08,440 --> 00:34:11,520 Speaker 2: return shareheld value. It is there for the greater good. 672 00:34:11,560 --> 00:34:13,800 Speaker 2: In this case, the exact words that you just quoted, 673 00:34:14,440 --> 00:34:18,160 Speaker 2: that not for profit owned, among other things, this for 674 00:34:18,200 --> 00:34:22,360 Speaker 2: profit entity that was set up because the resources that 675 00:34:22,360 --> 00:34:24,600 Speaker 2: are needed in order to continue to evolve open AI 676 00:34:25,200 --> 00:34:28,719 Speaker 2: requires tremendous billions of dollars of resources. So they've set 677 00:34:28,719 --> 00:34:32,920 Speaker 2: this up and that entity was able to then bring 678 00:34:32,960 --> 00:34:36,000 Speaker 2: in a lot of outside money, billions of dollars to 679 00:34:36,040 --> 00:34:39,000 Speaker 2: continue to evolve and see the developments that we've seen 680 00:34:39,440 --> 00:34:42,680 Speaker 2: come out of Open Eye AI since then, Chat, GPT 681 00:34:42,800 --> 00:34:45,920 Speaker 2: and many many, many other tools. That's not that unusual 682 00:34:46,000 --> 00:34:47,879 Speaker 2: a set up. There are other organizations that are set 683 00:34:47,960 --> 00:34:50,200 Speaker 2: up like that and worked just fine. But in this 684 00:34:50,320 --> 00:34:53,719 Speaker 2: case there was really a lack of alignment, and that 685 00:34:54,000 --> 00:34:57,800 Speaker 2: not for profit organization management that I think it was 686 00:34:57,840 --> 00:35:01,840 Speaker 2: a four person board decided they were either not in 687 00:35:01,840 --> 00:35:03,840 Speaker 2: a loop or not comfortable with where the for profit 688 00:35:03,960 --> 00:35:09,560 Speaker 2: entity was, and so they apparently without any consultation with anyone, fired. 689 00:35:09,239 --> 00:35:10,440 Speaker 1: The booted him. 690 00:35:10,560 --> 00:35:13,080 Speaker 2: They booted him, and they didn't really understand what well anyway, 691 00:35:13,160 --> 00:35:16,360 Speaker 2: they clearly didn't foresee what the rebound would be. 692 00:35:16,520 --> 00:35:19,719 Speaker 1: Then there's a huge uprising among the employees. I think 693 00:35:19,800 --> 00:35:22,040 Speaker 1: eighty percent said they were going to quit if he 694 00:35:22,160 --> 00:35:25,040 Speaker 1: was gone, and they were going to follow him to Microsoft, 695 00:35:25,400 --> 00:35:29,440 Speaker 1: and then suddenly he's back in business at open AI. 696 00:35:30,040 --> 00:35:32,439 Speaker 1: Can you guys help us make sense of it? Chris, 697 00:35:32,520 --> 00:35:33,520 Speaker 1: do you want to start. 698 00:35:33,560 --> 00:35:35,360 Speaker 4: Yeah again with the warning that a lot of this 699 00:35:35,480 --> 00:35:38,880 Speaker 4: is speculation because only you know the four members who 700 00:35:39,000 --> 00:35:41,239 Speaker 4: voted to out him, and the six board members total 701 00:35:41,320 --> 00:35:43,440 Speaker 4: really know what happened in the room where it happens. 702 00:35:43,520 --> 00:35:46,440 Speaker 4: But the popular understanding right now is that it was 703 00:35:46,480 --> 00:35:51,319 Speaker 4: a concern over movie too fast versus having safeguards. But 704 00:35:51,640 --> 00:35:53,920 Speaker 4: it may come out with future reporting that it was 705 00:35:54,280 --> 00:35:57,640 Speaker 4: about product moves, or about the decision to open up 706 00:35:57,719 --> 00:35:59,880 Speaker 4: so much of the access to the technology that they 707 00:36:00,360 --> 00:36:02,959 Speaker 4: had to slow down new signups. I mean, I've seen 708 00:36:03,080 --> 00:36:06,040 Speaker 4: many people speculate on what the causes were. Also the 709 00:36:06,040 --> 00:36:09,799 Speaker 4: possibility that some sort of particularly quantum leap in the 710 00:36:09,840 --> 00:36:13,160 Speaker 4: technology caused the board to have anxiety, but at this 711 00:36:13,200 --> 00:36:16,920 Speaker 4: point we don't know. I think future reporting, good investigative, 712 00:36:16,920 --> 00:36:19,680 Speaker 4: shoe works, shoe leather work right is needed right now 713 00:36:19,680 --> 00:36:21,320 Speaker 4: to figure out what actually went. 714 00:36:21,160 --> 00:36:24,279 Speaker 1: Down the loss. I know that Chris just mentioned sort 715 00:36:24,320 --> 00:36:26,879 Speaker 1: of a new technology, and I've been reading about this 716 00:36:27,000 --> 00:36:30,760 Speaker 1: project q asterisk. I don't even know how you say it, Vivian, 717 00:36:30,840 --> 00:36:31,560 Speaker 1: how do you say that? 718 00:36:31,760 --> 00:36:32,240 Speaker 4: Q star? 719 00:36:33,200 --> 00:36:36,440 Speaker 1: Q star has been described as a major breakthrough in 720 00:36:36,480 --> 00:36:41,759 Speaker 1: the company's pursuit of artificial general intelligence. So can you 721 00:36:41,800 --> 00:36:45,279 Speaker 1: help me understand veloss, what the hell that means and 722 00:36:45,320 --> 00:36:47,359 Speaker 1: what that technology was? Do you know? 723 00:36:47,920 --> 00:36:50,040 Speaker 3: Sure? Super happy too. I've read, I mean, all of 724 00:36:50,040 --> 00:36:51,960 Speaker 3: the public reporting and some of the peoper is behind it. 725 00:36:52,000 --> 00:36:54,200 Speaker 3: But can I give you my spicy take first before 726 00:36:54,239 --> 00:36:55,280 Speaker 3: I tell you about q start. 727 00:36:55,440 --> 00:36:58,239 Speaker 1: Oh we love spicy takes. Here a question the. 728 00:36:58,200 --> 00:37:01,880 Speaker 3: Two line here, like this is the telenovella of twenty 729 00:37:01,960 --> 00:37:05,520 Speaker 3: twenty three. None of this matters, right, but we love 730 00:37:05,600 --> 00:37:08,440 Speaker 3: our tabloid headlines that we have spent so much time 731 00:37:08,640 --> 00:37:10,880 Speaker 3: I got to say, hundreds of millions of hours of 732 00:37:10,960 --> 00:37:14,840 Speaker 3: human time thinking about Sam Altman and open Ai. Let's 733 00:37:14,840 --> 00:37:17,319 Speaker 3: put this in context, and it's so important that we 734 00:37:17,360 --> 00:37:20,320 Speaker 3: get this right. Open Ai is a company that was 735 00:37:20,360 --> 00:37:23,160 Speaker 3: based on a public paper that taught you how to 736 00:37:23,200 --> 00:37:26,800 Speaker 3: do LLLMS large language models, these like these chat GPT 737 00:37:26,920 --> 00:37:30,719 Speaker 3: type things. Right. They raised billions of dollars, which they 738 00:37:30,760 --> 00:37:34,120 Speaker 3: spent pretty much exclusively on what we call compute right 739 00:37:34,200 --> 00:37:37,080 Speaker 3: access to a bunch of computers, and they built the 740 00:37:37,120 --> 00:37:41,880 Speaker 3: first product that people could see. Nothing revolutionary happened at 741 00:37:41,920 --> 00:37:43,920 Speaker 3: open ai except for the fact that they took this 742 00:37:44,000 --> 00:37:47,080 Speaker 3: incredible paper that was done by some amazing scientists and 743 00:37:47,120 --> 00:37:49,960 Speaker 3: then just threw money at the problem. And once they did, 744 00:37:49,960 --> 00:37:51,920 Speaker 3: what did everybody else do? While then Microsoft threw a 745 00:37:51,920 --> 00:37:53,480 Speaker 3: lot of money at the problem at Facebook threw a 746 00:37:53,480 --> 00:37:54,840 Speaker 3: lot of money at it, Google threw a lot of 747 00:37:54,880 --> 00:37:56,960 Speaker 3: money at it, and they all came up with technologies 748 00:37:56,960 --> 00:37:59,920 Speaker 3: that are pretty similar, some are slightly better than ours. Okay, 749 00:38:00,360 --> 00:38:02,439 Speaker 3: it's important for us to say this because we spend 750 00:38:02,480 --> 00:38:05,359 Speaker 3: a lot of time daifying open ai as if it's 751 00:38:05,400 --> 00:38:08,120 Speaker 3: the most amazing thing that's ever happened. And it turns 752 00:38:08,160 --> 00:38:10,080 Speaker 3: out that when you have a pretty complex problem and 753 00:38:10,160 --> 00:38:12,600 Speaker 3: a pretty complex way to solve it, and you spend 754 00:38:12,600 --> 00:38:15,520 Speaker 3: a couple of billion dollars, you can come out with 755 00:38:15,560 --> 00:38:18,600 Speaker 3: an answer pretty easily. Okay, I say all that to 756 00:38:18,640 --> 00:38:20,680 Speaker 3: you and excuse the mini rant, because now we have 757 00:38:20,719 --> 00:38:23,800 Speaker 3: a real question in front of us, right, why is it, Katie, 758 00:38:23,840 --> 00:38:26,000 Speaker 3: that we're okay with the world in which a technology 759 00:38:26,040 --> 00:38:28,520 Speaker 3: that could change every human life on the planet is 760 00:38:28,560 --> 00:38:31,400 Speaker 3: held by seven companies that have these kinds of like 761 00:38:31,640 --> 00:38:34,880 Speaker 3: human personal dramas that drive what will happen with them. 762 00:38:35,000 --> 00:38:38,520 Speaker 1: And that's all about regulation. But philosoph before you do that, 763 00:38:38,600 --> 00:38:41,279 Speaker 1: you can to start what is QStar before we talk 764 00:38:41,320 --> 00:38:44,640 Speaker 1: about regulation, because I've read about it and I'm it's 765 00:38:44,680 --> 00:38:48,279 Speaker 1: sort of shrouded in mystery and interest and. 766 00:38:48,280 --> 00:38:50,040 Speaker 3: Again, right, and one of the things I really appreciate 767 00:38:50,040 --> 00:38:52,080 Speaker 3: about Chris and neither of us really want to be 768 00:38:52,160 --> 00:38:54,359 Speaker 3: a pundit. Right, We've both been experts in this field 769 00:38:54,360 --> 00:38:56,520 Speaker 3: for a long time. What I can glean from what's 770 00:38:56,560 --> 00:38:58,879 Speaker 3: been publicly reported and from some of the sources I've 771 00:38:58,880 --> 00:39:01,880 Speaker 3: talked to, is that it's a shift from focusing on 772 00:39:02,040 --> 00:39:05,560 Speaker 3: language as a predictive model to being able to focus 773 00:39:05,560 --> 00:39:09,120 Speaker 3: on things like math problems as a reasoning model. So 774 00:39:09,160 --> 00:39:11,279 Speaker 3: instead of saying, hey, I've got a sentence you know 775 00:39:11,840 --> 00:39:14,560 Speaker 3: twinkle twinkle, Well, we know the next words are probably 776 00:39:14,600 --> 00:39:17,160 Speaker 3: little star, it's instead a way to say, well, what 777 00:39:17,280 --> 00:39:21,560 Speaker 3: is two plus two? And you might say probabilistically, because 778 00:39:21,600 --> 00:39:23,400 Speaker 3: I've looked at everything humans have ever written, Well, when 779 00:39:23,440 --> 00:39:25,799 Speaker 3: you say two plus two, it usually followed my equals four. 780 00:39:26,480 --> 00:39:28,600 Speaker 3: But if somebody along the way, in some book had 781 00:39:28,600 --> 00:39:31,799 Speaker 3: written two plus two equals five, then one in ten 782 00:39:31,880 --> 00:39:34,759 Speaker 3: million times the large language model might say, oh, two 783 00:39:34,760 --> 00:39:37,120 Speaker 3: plus two equals five. We're trying to fix that, And 784 00:39:37,160 --> 00:39:39,759 Speaker 3: so q star says, can we actually reason if we 785 00:39:39,840 --> 00:39:42,080 Speaker 3: have two of one thing and two of another what 786 00:39:42,160 --> 00:39:45,040 Speaker 3: happens when you put them together. This is a big breakthrough. 787 00:39:45,120 --> 00:39:47,160 Speaker 3: It is something that gets us closer to what Chris 788 00:39:47,200 --> 00:39:49,839 Speaker 3: described as AGI right, that idea that you've got one 789 00:39:49,840 --> 00:39:52,080 Speaker 3: model that can talk about language and I can do 790 00:39:52,120 --> 00:39:54,960 Speaker 3: a little bit of math. We don't have any sense yet. 791 00:39:54,960 --> 00:39:57,120 Speaker 3: There hasn't been public reporting yet of just how good 792 00:39:57,120 --> 00:39:59,640 Speaker 3: of a breakthrough this is. But again, you take seven 793 00:39:59,719 --> 00:40:02,120 Speaker 3: or eight hundred really smart people, you give them a 794 00:40:02,200 --> 00:40:04,239 Speaker 3: lot of compute, and you say, hey, go figure some 795 00:40:04,280 --> 00:40:06,319 Speaker 3: stuff out, and this is what the next breakthrough looks like. 796 00:40:07,239 --> 00:40:09,080 Speaker 3: I don't think it's the thing that's going to lead 797 00:40:09,120 --> 00:40:11,759 Speaker 3: to terminator style robots and helicopters that are out there 798 00:40:11,800 --> 00:40:13,240 Speaker 3: trying to kill us all. That's all I'm saying. 799 00:40:13,520 --> 00:40:16,319 Speaker 1: That's good to know. I appreciate that. Well, I think 800 00:40:16,360 --> 00:40:19,880 Speaker 1: you raised the big question, and that obviously is regulation 801 00:40:20,120 --> 00:40:23,359 Speaker 1: something that Vivian and I dealt with a lot when 802 00:40:23,400 --> 00:40:26,200 Speaker 1: we were on this asping commission. 803 00:40:26,000 --> 00:40:28,600 Speaker 2: They ask in Commission on Information Disorder. 804 00:40:28,239 --> 00:40:32,200 Speaker 1: Thank you very much, Vivian, where it's very, very difficult 805 00:40:32,600 --> 00:40:38,080 Speaker 1: to regulate these things. And maybe I see veloss your scowling, 806 00:40:38,560 --> 00:40:41,480 Speaker 1: and so you think that's not an accurate statement. I'm 807 00:40:41,520 --> 00:40:43,640 Speaker 1: good at reading facial expressions, Belosto. 808 00:40:45,160 --> 00:40:47,319 Speaker 3: Let's start with the question, though, like, why are we 809 00:40:47,400 --> 00:40:50,080 Speaker 3: so focused on regulating? Right? What does it mean to regulate? 810 00:40:50,160 --> 00:40:52,279 Speaker 3: It means figure out all the ways it can harm 811 00:40:52,360 --> 00:40:54,440 Speaker 3: us and limit them from doing so, let me ask 812 00:40:54,480 --> 00:40:56,400 Speaker 3: you a different question, like, look, I grew up in 813 00:40:56,480 --> 00:40:59,600 Speaker 3: rural Illinois as like a very proud American, but my 814 00:40:59,680 --> 00:41:02,120 Speaker 3: parents or not well off. For me, the biggest thing 815 00:41:02,160 --> 00:41:04,160 Speaker 3: in the world was being able to access a library. 816 00:41:04,600 --> 00:41:06,480 Speaker 3: And I'll tell you why this matters, right, I'd go 817 00:41:06,520 --> 00:41:08,440 Speaker 3: to a library that was paid for by a pedance 818 00:41:08,480 --> 00:41:11,400 Speaker 3: of tax dollars, that took books and knowledge and all 819 00:41:11,400 --> 00:41:13,600 Speaker 3: of these public assets and made them available to me 820 00:41:13,640 --> 00:41:16,359 Speaker 3: as a curious YOUMKID. Today, if we're sitting here talking 821 00:41:16,400 --> 00:41:19,239 Speaker 3: about AI, you and I are fixed in a conversation 822 00:41:19,280 --> 00:41:22,719 Speaker 3: that says AI is owned by private companies. We don't 823 00:41:22,800 --> 00:41:25,840 Speaker 3: know how private companies make decisions. Well, our tool is 824 00:41:25,880 --> 00:41:29,000 Speaker 3: to regulate them. What if we ask a different question, 825 00:41:29,120 --> 00:41:33,040 Speaker 3: why are governments investing in building public purpose AI that's 826 00:41:33,040 --> 00:41:36,200 Speaker 3: done transparently, that's actually said. This is like a library, 827 00:41:36,200 --> 00:41:38,960 Speaker 3: it's a part of public infrastructure. And when we make 828 00:41:39,040 --> 00:41:41,960 Speaker 3: decisions about how AI will be used, that's a public 829 00:41:42,000 --> 00:41:45,360 Speaker 3: and democratic conversation, not for a board of four people. 830 00:41:45,640 --> 00:41:47,760 Speaker 3: We've seen what happens when you let them make decisions 831 00:41:47,800 --> 00:41:48,800 Speaker 3: about AI companies. 832 00:41:49,000 --> 00:41:52,000 Speaker 1: Well, by Chris, why isn't government getting more involved? 833 00:41:52,320 --> 00:41:54,640 Speaker 4: Well, there's a couple of reasons. I mean. One is 834 00:41:55,120 --> 00:41:57,120 Speaker 4: at the scale of the US federal government, which I 835 00:41:57,120 --> 00:42:00,640 Speaker 4: think is what you mean by government. The response by 836 00:42:00,680 --> 00:42:04,279 Speaker 4: the US federal government is often reactive and sectoral. So 837 00:42:04,320 --> 00:42:07,080 Speaker 4: what I mean by that is reactive, meaning that often 838 00:42:07,440 --> 00:42:09,640 Speaker 4: the US federal government doesn't move in a large scale 839 00:42:09,680 --> 00:42:12,920 Speaker 4: until something clearly bad has happened, and something that's so 840 00:42:12,960 --> 00:42:15,480 Speaker 4: bad that everyone accepts that it was bad, and thereafter 841 00:42:15,560 --> 00:42:17,960 Speaker 4: the US federal government will make a new agency to 842 00:42:18,280 --> 00:42:21,680 Speaker 4: govern a particular sector. So bisectoral, what I mean is 843 00:42:22,080 --> 00:42:24,840 Speaker 4: we have a sector of the law and in a 844 00:42:24,840 --> 00:42:29,560 Speaker 4: branch of the US government around say finance or transportation 845 00:42:30,360 --> 00:42:33,239 Speaker 4: or other sectors of our lives, rather than having a 846 00:42:33,280 --> 00:42:36,560 Speaker 4: branch of government that works on technologies read large. A 847 00:42:36,719 --> 00:42:38,799 Speaker 4: counter example to what I just said is FTC. So 848 00:42:39,000 --> 00:42:42,640 Speaker 4: Federal Trade Commission works on antitrust, but under the current 849 00:42:42,680 --> 00:42:45,480 Speaker 4: leadership of FTC they have sort of reasserted that part 850 00:42:45,520 --> 00:42:47,600 Speaker 4: of the purpose of FDC is to think about consumer 851 00:42:47,680 --> 00:42:50,280 Speaker 4: protection as well, so there's an option there for FTC 852 00:42:50,760 --> 00:42:54,080 Speaker 4: to be responsive. That said, there are other ways that 853 00:42:54,120 --> 00:42:58,080 Speaker 4: the US government operates other than laws, for example, executive orders, 854 00:42:58,360 --> 00:43:00,479 Speaker 4: which can be an opportunity for the President to say 855 00:43:00,520 --> 00:43:03,040 Speaker 4: this is really important, and I'm demanding that other people 856 00:43:03,040 --> 00:43:05,680 Speaker 4: who are in the White House respond or commission reports 857 00:43:05,680 --> 00:43:08,640 Speaker 4: on something, and by the spending power of the US government. 858 00:43:08,719 --> 00:43:11,400 Speaker 4: So when the US government says we will no longer 859 00:43:11,520 --> 00:43:13,919 Speaker 4: give money to any company that doesn't meet this bar 860 00:43:14,480 --> 00:43:17,360 Speaker 4: in terms of safety or transparency or other things that 861 00:43:17,400 --> 00:43:20,600 Speaker 4: we may want from technologies in general, that has a 862 00:43:20,680 --> 00:43:24,479 Speaker 4: huge market effect because without passing any laws, the White 863 00:43:24,480 --> 00:43:27,880 Speaker 4: House in this case can actually drive companies to behave 864 00:43:28,040 --> 00:43:29,960 Speaker 4: differently for market reasons. 865 00:43:30,400 --> 00:43:32,840 Speaker 3: Here's the thing, right again. I know it's a big statement, 866 00:43:32,840 --> 00:43:34,520 Speaker 3: and we're kind of nipping around the edges and we're 867 00:43:34,560 --> 00:43:36,480 Speaker 3: kind of saying about what can government do today? But 868 00:43:36,520 --> 00:43:38,440 Speaker 3: I'm going to ask the question again, why are we 869 00:43:38,480 --> 00:43:40,520 Speaker 3: so okay with the fact that we've just given up 870 00:43:40,560 --> 00:43:42,759 Speaker 3: as a public citizen rey to say that we could 871 00:43:42,760 --> 00:43:45,280 Speaker 3: actually own and build these tools. There's three things government 872 00:43:45,280 --> 00:43:47,480 Speaker 3: could do that I don't think risk touchedock. The first 873 00:43:47,600 --> 00:43:50,520 Speaker 3: is that could invest in public compute resources to make 874 00:43:50,760 --> 00:43:53,839 Speaker 3: supercomputing available to lots of communities and groups that are 875 00:43:53,840 --> 00:43:56,160 Speaker 3: working on AI. I worked with an amazing group called 876 00:43:56,200 --> 00:43:59,640 Speaker 3: Indigity Genius. It's a number of indigenous AI scientists. We're 877 00:43:59,640 --> 00:44:02,440 Speaker 3: building tools for people to use on reservations that use 878 00:44:02,480 --> 00:44:06,000 Speaker 3: AI for their public purpose. There's not compute resources between 879 00:44:06,040 --> 00:44:09,399 Speaker 3: Boise and Chicago that they can get access to. Right 880 00:44:09,520 --> 00:44:11,000 Speaker 3: we should be spending money on this. There's a bill 881 00:44:11,040 --> 00:44:14,320 Speaker 3: in Congress right now. The second is data representation, mandating 882 00:44:14,320 --> 00:44:18,000 Speaker 3: that these companies actually include public data sets that are 883 00:44:18,040 --> 00:44:21,640 Speaker 3: truly represented with guidelines. This could happen through regulation, it 884 00:44:21,680 --> 00:44:24,120 Speaker 3: could happen through policy, it could happen through an EO. 885 00:44:24,239 --> 00:44:27,480 Speaker 3: And the last is talent. Why are we so confident 886 00:44:27,520 --> 00:44:29,760 Speaker 3: that the only way that you can make a career 887 00:44:29,760 --> 00:44:31,640 Speaker 3: in AI to go get a degree and then go 888 00:44:31,680 --> 00:44:34,480 Speaker 3: work for one of these companies making whatever six figures. 889 00:44:34,800 --> 00:44:37,120 Speaker 3: What if we built a public service core of computer 890 00:44:37,160 --> 00:44:39,680 Speaker 3: scientists and data scientists and we're seeing the start of 891 00:44:39,680 --> 00:44:42,359 Speaker 3: that under the Biden Prris administration, to actually say, let's 892 00:44:42,360 --> 00:44:44,200 Speaker 3: go work in government and let's work in communities to 893 00:44:44,200 --> 00:44:46,480 Speaker 3: build AI products. These are three things we could do 894 00:44:46,520 --> 00:44:48,840 Speaker 3: that actually have nothing to do with limiting the safety 895 00:44:48,880 --> 00:44:51,799 Speaker 3: of EI tools. That's important, but that can't be the 896 00:44:51,800 --> 00:44:54,520 Speaker 3: only conversation. And it feels like it is right now. 897 00:44:54,719 --> 00:44:56,840 Speaker 1: And Vivian, don't you think it's weird that this is 898 00:44:56,880 --> 00:44:59,920 Speaker 1: all handled by the FTC? I mean, why isn't there 899 00:45:00,239 --> 00:45:05,560 Speaker 1: cabinet level position kind of overseeing technology. It seems to 900 00:45:05,600 --> 00:45:09,600 Speaker 1: me it's such a huge issue that, you know, new 901 00:45:09,680 --> 00:45:14,040 Speaker 1: departments have been established historically, the Department of the Interior, 902 00:45:14,120 --> 00:45:18,040 Speaker 1: you know, HHS. I don't even know when they were established, 903 00:45:18,040 --> 00:45:20,600 Speaker 1: but it seems to me it's time to establish a 904 00:45:20,719 --> 00:45:25,359 Speaker 1: new cabinet level position and a whole infrastructure that can 905 00:45:25,400 --> 00:45:27,239 Speaker 1: help manage these issues. Right. 906 00:45:27,480 --> 00:45:30,439 Speaker 2: Yeah, Well, Biden's executive order doesn't quite go that far, 907 00:45:30,560 --> 00:45:33,080 Speaker 2: but it's starting to walk towards those space sort of. 908 00:45:33,120 --> 00:45:35,200 Speaker 2: Among the many things that the Executive Order says is 909 00:45:35,320 --> 00:45:40,440 Speaker 2: deep coordination among various parts of government, more AI expertise 910 00:45:40,800 --> 00:45:43,200 Speaker 2: in all of these federal offices. I mean, that's part 911 00:45:43,200 --> 00:45:45,919 Speaker 2: of the problem. You don't have people that understand the technology, 912 00:45:45,960 --> 00:45:47,920 Speaker 2: it's going to be hard to make do any kind 913 00:45:47,960 --> 00:45:51,200 Speaker 2: of regulation. I think they're also limited by what can 914 00:45:51,280 --> 00:45:54,279 Speaker 2: be done without the ascent of Congress, since Congress doesn't 915 00:45:54,320 --> 00:45:56,520 Speaker 2: seem to be assenting to just about anything right now. 916 00:45:56,760 --> 00:45:59,920 Speaker 4: There's good and bad this idea of focusing new creation 917 00:46:00,040 --> 00:46:02,279 Speaker 4: and of branches of government on AI. I like the 918 00:46:02,280 --> 00:46:05,600 Speaker 4: idea that government is taking consumer protection seriously. Like that 919 00:46:05,680 --> 00:46:08,640 Speaker 4: sounds good, but a loss there is realizing the ways 920 00:46:08,640 --> 00:46:11,480 Speaker 4: in which AI is just another technology. So we already 921 00:46:11,480 --> 00:46:14,880 Speaker 4: have a Presidential Office of Science and Technology Policy. We 922 00:46:14,960 --> 00:46:18,880 Speaker 4: already have funding agencies. I'll show my biases as an academic, 923 00:46:18,920 --> 00:46:21,640 Speaker 4: but we have the National Science Foundation. It's been writing 924 00:46:21,680 --> 00:46:24,240 Speaker 4: checks since the mid fifties. So there are already ways 925 00:46:24,280 --> 00:46:27,080 Speaker 4: for the US government to spur innovation. So I like 926 00:46:27,160 --> 00:46:30,200 Speaker 4: the idea of US recognizing that AI is having a 927 00:46:30,200 --> 00:46:33,680 Speaker 4: big impact. Again, that's partly about technology, but also part 928 00:46:33,719 --> 00:46:36,200 Speaker 4: of the power of markets in our own norms. But 929 00:46:36,719 --> 00:46:39,319 Speaker 4: I also don't want to make AI so exceptional that 930 00:46:39,360 --> 00:46:42,319 Speaker 4: we don't profit from the lessons learned for dealing with 931 00:46:42,320 --> 00:46:46,319 Speaker 4: technologies in general. We've regulated and made safe and made 932 00:46:46,320 --> 00:46:49,279 Speaker 4: productive so many forms of technology through both positive and 933 00:46:49,360 --> 00:46:51,840 Speaker 4: negative regulation. So I feel like there's lessons to be 934 00:46:51,920 --> 00:46:54,440 Speaker 4: learned there that we might lose out at if we 935 00:46:54,480 --> 00:46:56,640 Speaker 4: somehow think of AI as being magic and not just 936 00:46:56,680 --> 00:46:57,920 Speaker 4: another form of technology. 937 00:46:58,120 --> 00:47:00,080 Speaker 2: HETI I have a quick follow up, which is the 938 00:47:00,120 --> 00:47:03,720 Speaker 2: issue with speed. These tech companies are moving really fast, 939 00:47:03,920 --> 00:47:07,759 Speaker 2: and government, often for very good reasons, move slowly. Governments, 940 00:47:07,800 --> 00:47:10,279 Speaker 2: I should say, because you've also got actions coming out 941 00:47:10,280 --> 00:47:12,160 Speaker 2: of the European Union, in the UK, other parts of 942 00:47:12,160 --> 00:47:15,400 Speaker 2: the inter governmental organizations like the United Nations, which I 943 00:47:15,440 --> 00:47:17,799 Speaker 2: know you're part of that group that's working on this philosophy, 944 00:47:18,040 --> 00:47:20,200 Speaker 2: I mean, can they possibly keep up, let alone get 945 00:47:20,200 --> 00:47:20,719 Speaker 2: out ahead of this. 946 00:47:21,080 --> 00:47:22,839 Speaker 3: Yeah, you know, I think you're asking exactly the right question. 947 00:47:22,920 --> 00:47:25,080 Speaker 3: I'm going to disagree with Chris just in a matter 948 00:47:25,120 --> 00:47:27,920 Speaker 3: of degree, which is the sense that AI is exceptional 949 00:47:28,000 --> 00:47:30,359 Speaker 3: only in exactly what you refer to, Vivian is the 950 00:47:30,440 --> 00:47:33,520 Speaker 3: speed of transformation that's creating in our society, and so yes, 951 00:47:33,560 --> 00:47:35,120 Speaker 3: there's a lot to be learned by how we've dealt 952 00:47:35,120 --> 00:47:37,160 Speaker 3: with this in the past. But we don't have one 953 00:47:37,200 --> 00:47:39,279 Speaker 3: hundred years between the introduction of the Cotton gen and 954 00:47:39,320 --> 00:47:42,480 Speaker 3: the creation of the National Labor Relations Board, right, we 955 00:47:42,560 --> 00:47:45,719 Speaker 3: don't have that much time now. Look, I think the 956 00:47:45,840 --> 00:47:48,719 Speaker 3: question is what are we reacting to and why are 957 00:47:48,719 --> 00:47:51,240 Speaker 3: we spending so much more time reacting to tech companies? 958 00:47:51,640 --> 00:47:54,560 Speaker 3: Where is there public leadership that says, what's the vision 959 00:47:54,600 --> 00:47:57,400 Speaker 3: for what AI should be in human society and how 960 00:47:57,440 --> 00:48:00,560 Speaker 3: do we create policy that gets us there. The mandates 961 00:48:00,560 --> 00:48:02,719 Speaker 3: of the Secretary General of the UN, Antonio bu Terras 962 00:48:02,760 --> 00:48:04,800 Speaker 3: has given us on this high level advisory board to 963 00:48:04,800 --> 00:48:07,359 Speaker 3: which I've been avoided is to move beyond just thinking 964 00:48:07,360 --> 00:48:09,560 Speaker 3: about the risks of what happens when AI is deployed 965 00:48:09,560 --> 00:48:11,920 Speaker 3: by private companies and say, what does it actually look 966 00:48:11,960 --> 00:48:14,480 Speaker 3: like to build a governance mechanism they use this AI 967 00:48:14,560 --> 00:48:18,120 Speaker 3: to create a better future. That's not how our particular 968 00:48:18,160 --> 00:48:20,480 Speaker 3: government system is set up at the moment. I think 969 00:48:20,520 --> 00:48:23,040 Speaker 3: the Biden Harris Executive Order, which we refer to a 970 00:48:23,080 --> 00:48:26,239 Speaker 3: couple of times on here, was actually a really meaningful 971 00:48:26,280 --> 00:48:28,719 Speaker 3: attempt to take a lot of this language and push 972 00:48:28,719 --> 00:48:32,319 Speaker 3: it into one hundred page document. It's a start, but 973 00:48:32,400 --> 00:48:34,960 Speaker 3: we need a new public conversation. This isn't something to 974 00:48:35,000 --> 00:48:37,640 Speaker 3: say government should go figure this out. I think we 975 00:48:37,719 --> 00:48:41,279 Speaker 3: need to actually have a public American conversation about what 976 00:48:41,360 --> 00:48:44,560 Speaker 3: a future driven by AI looks like, and we need 977 00:48:44,560 --> 00:48:45,680 Speaker 3: to figure out where to start that. 978 00:48:46,360 --> 00:48:49,440 Speaker 1: I'd love to follow up by asking how quickly is 979 00:48:49,480 --> 00:48:53,360 Speaker 1: it moving? I mean, how different will the world look, say, 980 00:48:53,560 --> 00:48:57,279 Speaker 1: in one to five years, Chris, I mean, what are 981 00:48:57,320 --> 00:49:01,080 Speaker 1: you seeing in terms of how quickly this technology is evolving? 982 00:49:01,600 --> 00:49:04,320 Speaker 1: What's going to look different in a few years. 983 00:49:04,760 --> 00:49:08,239 Speaker 4: I often like to talk about norms, laws, markets, and architecture, 984 00:49:08,280 --> 00:49:11,240 Speaker 4: which is this idea from the legal scholar Larry Lesseik 985 00:49:11,320 --> 00:49:14,359 Speaker 4: about the forces acting on us can be clustered into 986 00:49:14,400 --> 00:49:16,880 Speaker 4: those four groups and they all have their own time scales. 987 00:49:17,080 --> 00:49:19,960 Speaker 4: So architecture in this case includes technology which moves. It 988 00:49:20,000 --> 00:49:22,000 Speaker 4: feels like it moves really faster, and there might be 989 00:49:22,040 --> 00:49:25,000 Speaker 4: some sort of paradigm shift where we're confronted with the 990 00:49:25,000 --> 00:49:28,600 Speaker 4: new technology. Markets react very quickly. For example, we create 991 00:49:28,680 --> 00:49:31,439 Speaker 4: new job titles like prompt engineer and start paying people 992 00:49:31,520 --> 00:49:33,200 Speaker 4: to do that, and we start writing books about how 993 00:49:33,200 --> 00:49:36,200 Speaker 4: to use lllms, and then our norms adapt much more 994 00:49:36,239 --> 00:49:39,480 Speaker 4: slowly as we get to normative statements like should I 995 00:49:39,880 --> 00:49:43,280 Speaker 4: use in court a bunch of citations that were generated 996 00:49:43,280 --> 00:49:46,360 Speaker 4: by llms? Is it okay to write the eulogy for 997 00:49:46,400 --> 00:49:48,759 Speaker 4: my friend using chatgypt? Those are normative things that we 998 00:49:48,800 --> 00:49:51,239 Speaker 4: all have to react to. And then laws, and. 999 00:49:51,160 --> 00:49:53,480 Speaker 1: I'll answer to that question, no, it's not. Go ahead. 1000 00:49:54,600 --> 00:49:57,399 Speaker 4: So our norms constantly evolve, and then the laws, as 1001 00:49:57,440 --> 00:50:01,040 Speaker 4: you pointed out, are generally much slower. Timescale for laws 1002 00:50:01,120 --> 00:50:04,680 Speaker 4: is much longer than timescale for those. So you know, 1003 00:50:04,760 --> 00:50:08,080 Speaker 4: chat GPT was a great product innovation. GPT three had 1004 00:50:08,080 --> 00:50:09,640 Speaker 4: been around for like a year or two before that. 1005 00:50:10,120 --> 00:50:12,480 Speaker 4: I looked at my notes and saw that I was 1006 00:50:12,480 --> 00:50:14,799 Speaker 4: teaching GPT three to my class in the spring of 1007 00:50:14,800 --> 00:50:17,560 Speaker 4: twenty twenty two. And you know, GPT two was around 1008 00:50:17,560 --> 00:50:19,919 Speaker 4: before that, and chatbots, as I said, were around since 1009 00:50:19,960 --> 00:50:22,399 Speaker 4: the sixties. So many of these things are not new, 1010 00:50:22,480 --> 00:50:25,319 Speaker 4: But what changes quickly is our norms and also the 1011 00:50:25,320 --> 00:50:28,560 Speaker 4: way these things become products. So Opdai has done a 1012 00:50:28,560 --> 00:50:32,839 Speaker 4: great job of making GPT four the basis for other 1013 00:50:32,920 --> 00:50:36,120 Speaker 4: plugins and for API access, and other companies have been 1014 00:50:36,160 --> 00:50:38,000 Speaker 4: built sort of on top of that technology. 1015 00:50:38,360 --> 00:50:41,000 Speaker 1: What does that need help translate? What do you mean? 1016 00:50:41,080 --> 00:50:43,759 Speaker 4: Sorry? Yeah, sorry, I was a nerdy tangent there. So 1017 00:50:44,640 --> 00:50:48,000 Speaker 4: katis are application programming interfaces. It basically means I'm going 1018 00:50:48,080 --> 00:50:51,160 Speaker 4: to allow one program to interact with another program, and 1019 00:50:51,200 --> 00:50:53,760 Speaker 4: those two programs could be run by totally different companies. 1020 00:50:54,000 --> 00:50:56,200 Speaker 4: So I could have one company make a computer talk 1021 00:50:56,280 --> 00:50:59,960 Speaker 4: to a different companies computer, and all sorts of creativity 1022 00:51:00,160 --> 00:51:00,720 Speaker 4: is unlocked. 1023 00:51:00,719 --> 00:51:00,919 Speaker 2: There. 1024 00:51:01,040 --> 00:51:02,200 Speaker 1: Can you give me an example. 1025 00:51:02,520 --> 00:51:06,000 Speaker 4: Vivian had an itinerary, Now hook it up to Expedia 1026 00:51:06,120 --> 00:51:08,000 Speaker 4: or Kayak or some other company that will buy the 1027 00:51:08,000 --> 00:51:11,279 Speaker 4: ticket for you. So you had GPT right us at 1028 00:51:11,680 --> 00:51:14,240 Speaker 4: poem hook it up to a company that will already 1029 00:51:14,239 --> 00:51:16,240 Speaker 4: print it for you and mainly your card. 1030 00:51:16,880 --> 00:51:18,319 Speaker 1: Yeah, exactly, got it. 1031 00:51:19,120 --> 00:51:21,520 Speaker 4: So all of those interfaces are such an onlock to 1032 00:51:21,600 --> 00:51:24,799 Speaker 4: different people's creativity, and the people again could be you know, 1033 00:51:25,000 --> 00:51:28,160 Speaker 4: artists or students or other companies. So that's the thing 1034 00:51:28,200 --> 00:51:30,920 Speaker 4: that's easy to move quickly is you know, let's say 1035 00:51:30,920 --> 00:51:33,440 Speaker 4: we were all stuck with GPT three from spring of 1036 00:51:33,440 --> 00:51:36,200 Speaker 4: twenty twenty two. Now that we've had this normative change 1037 00:51:36,200 --> 00:51:38,799 Speaker 4: that everybody has had their eyes open to, their creativity, 1038 00:51:38,840 --> 00:51:41,799 Speaker 4: open to the market, open to which means a bunch 1039 00:51:41,800 --> 00:51:44,759 Speaker 4: of capital flowing to this new opportunity. There's so much 1040 00:51:44,800 --> 00:51:46,719 Speaker 4: room for things to change real fast. Not because the 1041 00:51:46,760 --> 00:51:49,520 Speaker 4: tech is advancing so fast and scientists are so smart, 1042 00:51:49,880 --> 00:51:51,799 Speaker 4: whether or not they are. It's because all of our 1043 00:51:51,880 --> 00:51:54,880 Speaker 4: norms and our markets are changing so fast. There's very 1044 00:51:54,920 --> 00:51:57,319 Speaker 4: little viscosity to stop us from coming up with new 1045 00:51:57,320 --> 00:52:00,719 Speaker 4: ways of doing things now that we have accepted, for example, 1046 00:52:01,360 --> 00:52:06,480 Speaker 4: moderately hallucinatory and somewhat truthful generative technologies. 1047 00:52:07,520 --> 00:52:09,399 Speaker 3: Let me give you two facts that take what Chris 1048 00:52:09,400 --> 00:52:12,160 Speaker 3: said with them in stark relief. He talked about GPT three, 1049 00:52:12,200 --> 00:52:13,880 Speaker 3: a lot a model that came out in twenty twenty 1050 00:52:13,960 --> 00:52:16,240 Speaker 3: out of twenty twenty two. Rather, it took about eighteen 1051 00:52:16,280 --> 00:52:18,600 Speaker 3: months to train that model. I'll spare you what that means, 1052 00:52:18,640 --> 00:52:20,880 Speaker 3: but it took about eighteen months of people working with computers. 1053 00:52:21,320 --> 00:52:24,880 Speaker 3: The newest supercomputer from Nvideo can now train a GPT 1054 00:52:24,920 --> 00:52:29,160 Speaker 3: three equivalent in four minutes. We have increased the amount 1055 00:52:29,239 --> 00:52:31,839 Speaker 3: of compute capacity exists on the planet by fifty five 1056 00:52:32,000 --> 00:52:35,520 Speaker 3: million times in the last ten years. The pace of 1057 00:52:35,680 --> 00:52:39,359 Speaker 3: change is so incredible here, and when Chris talks about 1058 00:52:39,400 --> 00:52:42,560 Speaker 3: the human components of that, the pieces of connecting and creativity, 1059 00:52:42,880 --> 00:52:45,360 Speaker 3: we also have to acknowledge that even just what's possible 1060 00:52:45,440 --> 00:52:47,600 Speaker 3: is changing almost by the day or by the month. 1061 00:52:47,920 --> 00:52:50,359 Speaker 3: Q Star that we talked about wouldn't even have been 1062 00:52:50,440 --> 00:52:54,120 Speaker 3: conceivable two years ago. So who knows what two years 1063 00:52:54,120 --> 00:52:56,879 Speaker 3: from now will look like. And that's my one last 1064 00:52:56,880 --> 00:52:59,680 Speaker 3: thought on regulation is we are so we're working so 1065 00:52:59,760 --> 00:53:03,400 Speaker 3: hard regulating what AI looked like two years ago. Maybe 1066 00:53:03,440 --> 00:53:06,040 Speaker 3: in the most frontier places, the most brilliant congress people 1067 00:53:06,040 --> 00:53:09,040 Speaker 3: are saying, what does AI look like today? We have 1068 00:53:09,120 --> 00:53:11,400 Speaker 3: no idea how to build a policy that regulates what 1069 00:53:11,440 --> 00:53:14,080 Speaker 3: AI will look like in five years, So we take 1070 00:53:14,120 --> 00:53:16,279 Speaker 3: control of building AI ourselves. 1071 00:53:16,840 --> 00:53:20,120 Speaker 1: In fact, I wanted to ask you both. Jeffrey Hinton, 1072 00:53:20,160 --> 00:53:23,600 Speaker 1: known as the godfather of AI, spent decades advancing AI, 1073 00:53:23,840 --> 00:53:28,279 Speaker 1: but we're recently cautioned about the potential existential dangers that 1074 00:53:28,400 --> 00:53:31,480 Speaker 1: could pose. I feel like you all have kind of 1075 00:53:31,560 --> 00:53:35,799 Speaker 1: diminished the bad stuff that goes with AI, and I'm 1076 00:53:35,920 --> 00:53:40,280 Speaker 1: curious if you can give us some sense of how 1077 00:53:40,400 --> 00:53:44,360 Speaker 1: it could be misused or abused in the wrong hands. 1078 00:53:44,840 --> 00:53:47,200 Speaker 4: To be clear, there's lots of bad stuff, it's just 1079 00:53:47,360 --> 00:53:51,200 Speaker 4: not that particular bad stuff. So there's bad stuff happening 1080 00:53:51,280 --> 00:53:53,759 Speaker 4: right now all the time. And so Jeff Hinton and 1081 00:53:53,800 --> 00:53:56,680 Speaker 4: others have portrayed a possible bad thing in the future 1082 00:53:57,239 --> 00:54:00,600 Speaker 4: that has some unknowable but I think very small probability. 1083 00:54:00,880 --> 00:54:03,239 Speaker 4: So there are other existential risks right now that don't 1084 00:54:03,400 --> 00:54:06,520 Speaker 4: evolve anything involving AI. Let's worry about those. But also 1085 00:54:06,560 --> 00:54:08,719 Speaker 4: when we're talking about AI, there's all sorts of bad 1086 00:54:08,719 --> 00:54:11,120 Speaker 4: things happening right now with AI. You know, if you automate, 1087 00:54:11,520 --> 00:54:14,560 Speaker 4: as philosophic saying earlier sexism or it doesn't make it 1088 00:54:14,640 --> 00:54:17,719 Speaker 4: less sexist. Right for you to have a biased algorithm 1089 00:54:17,760 --> 00:54:19,920 Speaker 4: and then you automate it so it can be you know, 1090 00:54:20,040 --> 00:54:22,880 Speaker 4: sexist or show biases at high efficiency at scale, that 1091 00:54:22,920 --> 00:54:25,719 Speaker 4: doesn't make it any less biased, right, It's still bad. 1092 00:54:25,920 --> 00:54:28,279 Speaker 4: So I wouldn't say that we're, at least on my point, 1093 00:54:28,360 --> 00:54:30,560 Speaker 4: trying to minimize the bad stuff. It's just it's not 1094 00:54:30,640 --> 00:54:32,680 Speaker 4: Jeff Hint, it's bad stuff that I'm more concerned about. 1095 00:54:32,920 --> 00:54:34,360 Speaker 3: Let me tell you, when I spend my time on 1096 00:54:34,680 --> 00:54:36,759 Speaker 3: I spend my time on making sure that communities around 1097 00:54:36,800 --> 00:54:39,080 Speaker 3: the world are just totally left out of the air revolution. 1098 00:54:39,600 --> 00:54:41,640 Speaker 3: I'd spend my time thinking about making sure that AI 1099 00:54:41,760 --> 00:54:44,960 Speaker 3: decisions that affect people's lives have the contours of human 1100 00:54:45,040 --> 00:54:47,560 Speaker 3: ethics around them. I spend my time making sure that 1101 00:54:47,600 --> 00:54:49,960 Speaker 3: the people who are building these tools are representative of 1102 00:54:50,000 --> 00:54:52,120 Speaker 3: all of us. I spend my time making sure that 1103 00:54:52,160 --> 00:54:54,360 Speaker 3: AI is not being used to run autonomous weapons and 1104 00:54:54,480 --> 00:54:57,319 Speaker 3: run warfare. These are things that we can all spend 1105 00:54:57,360 --> 00:54:59,480 Speaker 3: our time on to make sure that AI doesn't actually 1106 00:54:59,480 --> 00:55:01,800 Speaker 3: make the world worse and maybe makes the world better. 1107 00:55:02,200 --> 00:55:04,240 Speaker 3: I don't have time to be thinking about what happens 1108 00:55:04,239 --> 00:55:07,400 Speaker 3: in twenty five years when one man's conception of a 1109 00:55:07,520 --> 00:55:10,480 Speaker 3: risk comes true. There's a lot of risks that actually 1110 00:55:10,480 --> 00:55:12,800 Speaker 3: affect our daily lives today that we should be spending 1111 00:55:12,800 --> 00:55:13,760 Speaker 3: our time making better. 1112 00:55:14,360 --> 00:55:16,759 Speaker 2: One of the areas that we're very, very focused on 1113 00:55:17,000 --> 00:55:22,160 Speaker 2: at the Aspen Institute is the intersection of artificial intelligence, 1114 00:55:22,640 --> 00:55:26,960 Speaker 2: the upcoming twenty twenty four elections, and societal trust. And 1115 00:55:27,239 --> 00:55:30,160 Speaker 2: it's a big area of concern. We've seen you even 1116 00:55:30,280 --> 00:55:33,600 Speaker 2: just from recent elections outside the United States, recently in 1117 00:55:33,680 --> 00:55:37,640 Speaker 2: Argentina and in Netherlands, Slovakia and Poland. We've seen how 1118 00:55:37,760 --> 00:55:41,000 Speaker 2: some of the parties, the candidates, the campaigns are using AI, 1119 00:55:41,160 --> 00:55:44,960 Speaker 2: and there is some significant concern about our twenty twenty 1120 00:55:44,960 --> 00:55:48,799 Speaker 2: four elections and the ways that AI might impact what 1121 00:55:49,480 --> 00:55:53,359 Speaker 2: population thinks, how they vote, where they show up. Can 1122 00:55:53,400 --> 00:55:55,399 Speaker 2: you just share with us, both of you a little 1123 00:55:55,400 --> 00:55:57,680 Speaker 2: bit about what you're seeing there and what you think 1124 00:55:57,800 --> 00:55:59,480 Speaker 2: we should be most worried about here? 1125 00:55:59,520 --> 00:56:01,680 Speaker 3: If anything, Yeah, a look at a majority of the 1126 00:56:01,719 --> 00:56:03,680 Speaker 3: world's population is going to the polls next year. You 1127 00:56:03,719 --> 00:56:05,920 Speaker 3: just have putted out. What I'm concerned about is not 1128 00:56:05,960 --> 00:56:08,440 Speaker 3: how AI will go and change the elections. It's how 1129 00:56:08,520 --> 00:56:11,360 Speaker 3: bad actors are going to use AI to do what 1130 00:56:11,400 --> 00:56:14,719 Speaker 3: they've already done to perforate trust in our society, but 1131 00:56:14,800 --> 00:56:17,359 Speaker 3: do it even more effectively. I'm worried about things like 1132 00:56:17,480 --> 00:56:19,920 Speaker 3: somebody deciding to send out three hundred and fifty million 1133 00:56:20,040 --> 00:56:23,880 Speaker 3: individualized emails to manipulate the way people are going to vote. 1134 00:56:23,920 --> 00:56:26,040 Speaker 3: And here's the worst part. They don't have to include 1135 00:56:26,080 --> 00:56:29,040 Speaker 3: any misinformation or lies at all, because what they can 1136 00:56:29,120 --> 00:56:32,160 Speaker 3: do is look at real factual information, only give you 1137 00:56:32,200 --> 00:56:35,239 Speaker 3: a version of that story that affects your demographic as 1138 00:56:35,239 --> 00:56:38,040 Speaker 3: they understand you that analyzes your behaviors and tries to 1139 00:56:38,040 --> 00:56:40,080 Speaker 3: get you vote a certain way. We don't even have 1140 00:56:40,200 --> 00:56:43,040 Speaker 3: rules in place on what to do if somebody comes 1141 00:56:43,040 --> 00:56:45,440 Speaker 3: to you with something where not a single fact is incorrect, 1142 00:56:45,800 --> 00:56:48,560 Speaker 3: but is architected to manipulate you in some way. This 1143 00:56:48,600 --> 00:56:50,279 Speaker 3: is where we should be spending our time thinking about 1144 00:56:50,280 --> 00:56:51,200 Speaker 3: policy and regulation. 1145 00:56:51,880 --> 00:56:53,319 Speaker 1: Chris, any thoughts from you. 1146 00:56:53,800 --> 00:56:59,080 Speaker 4: It's been so far, just had to summarize it. Yeah, 1147 00:56:59,120 --> 00:57:01,880 Speaker 4: that's a real concern, and some of it is about 1148 00:57:02,360 --> 00:57:05,080 Speaker 4: AI as we understand it this year, but some of 1149 00:57:05,120 --> 00:57:08,280 Speaker 4: it is about the fact that our marketplace of ideas 1150 00:57:08,280 --> 00:57:11,600 Speaker 4: has become completely algorithmically empowered by a few private companies, 1151 00:57:12,080 --> 00:57:15,120 Speaker 4: and so all of our conceptions about how, you know, 1152 00:57:15,160 --> 00:57:17,320 Speaker 4: having lots of people have a free exchange of ideas, 1153 00:57:17,400 --> 00:57:20,120 Speaker 4: you know, so we're predicated on a very different sort 1154 00:57:20,120 --> 00:57:22,760 Speaker 4: of game theory of the way people are trading ideas. 1155 00:57:22,800 --> 00:57:25,600 Speaker 4: In addition to the fact that the digital assets are 1156 00:57:25,640 --> 00:57:29,480 Speaker 4: so easily manipulated that there's room for creating things that 1157 00:57:29,520 --> 00:57:31,640 Speaker 4: are that look trustworthy. 1158 00:57:31,520 --> 00:57:34,360 Speaker 1: Like Nancy Pelosi intoxicated. 1159 00:57:34,600 --> 00:57:35,920 Speaker 4: That's a good example, or. 1160 00:57:35,920 --> 00:57:40,280 Speaker 1: Tom Hanks talking and saying something any about a dentist 1161 00:57:40,360 --> 00:57:42,400 Speaker 1: or something some dental service, or. 1162 00:57:42,480 --> 00:57:45,760 Speaker 4: Simply taking video game footage from a video game and 1163 00:57:45,800 --> 00:57:48,400 Speaker 4: representing it as being from a war zone, which also 1164 00:57:48,480 --> 00:57:51,920 Speaker 4: has happened time and time again in different military conflicts 1165 00:57:51,960 --> 00:57:54,240 Speaker 4: and continues to happen. So it doesn't even have to 1166 00:57:54,280 --> 00:57:57,360 Speaker 4: be deep fixed, right, It can be absolutely cheap fix that, 1167 00:57:57,520 --> 00:58:00,320 Speaker 4: you know, accelerated by an information platform which is used 1168 00:58:00,400 --> 00:58:03,000 Speaker 4: to optimize engagement. Now I'm going to go down a 1169 00:58:03,520 --> 00:58:06,440 Speaker 4: slightly nerdy rant. I can tell anyways it's bad. So 1170 00:58:06,600 --> 00:58:08,640 Speaker 4: there's a lot of concern there, and there's a very 1171 00:58:08,640 --> 00:58:12,280 Speaker 4: difficult time for academic researchers to investigate it because the 1172 00:58:12,400 --> 00:58:14,800 Speaker 4: digital commons is now owned by a few private companies 1173 00:58:14,800 --> 00:58:17,240 Speaker 4: who are not particularly motivated to share information in a 1174 00:58:17,280 --> 00:58:20,000 Speaker 4: research friendly way. So it's difficult for us to do 1175 00:58:20,040 --> 00:58:22,480 Speaker 4: anything that even looks like experiments, which is the way 1176 00:58:22,480 --> 00:58:24,400 Speaker 4: science has been done for the last century, to do 1177 00:58:24,520 --> 00:58:27,640 Speaker 4: randomized control trials around different treatments. There's sort of no 1178 00:58:27,800 --> 00:58:31,320 Speaker 4: framework for doing that technologically nor ethically. The people who 1179 00:58:31,320 --> 00:58:34,680 Speaker 4: are most concerned about it are not particularly technologically able 1180 00:58:34,800 --> 00:58:37,040 Speaker 4: to get hold of lots of data and do statistical 1181 00:58:37,040 --> 00:58:40,440 Speaker 4: analyzes of them, so it's a concern. I mean, it's 1182 00:58:40,440 --> 00:58:43,520 Speaker 4: a concern politically, it's a concern for researchers who want 1183 00:58:43,560 --> 00:58:45,040 Speaker 4: to understand it. I'm concerned. 1184 00:58:45,440 --> 00:58:47,800 Speaker 2: I'll add one other thing, just as an addendum to this. 1185 00:58:48,040 --> 00:58:50,800 Speaker 2: There's much we can't accomplish between now and the elections 1186 00:58:50,800 --> 00:58:52,960 Speaker 2: next year. But one thing we can do, and this 1187 00:58:53,000 --> 00:58:55,000 Speaker 2: is work that we're taking on a little promotion for 1188 00:58:55,080 --> 00:58:57,680 Speaker 2: our for the Aspen Institute here is bringing groups together 1189 00:58:57,720 --> 00:58:59,640 Speaker 2: who are not talking to each other. We did a 1190 00:58:59,640 --> 00:59:01,800 Speaker 2: deep We spoke to a lot of experts, including both 1191 00:59:01,840 --> 00:59:04,160 Speaker 2: of the experts here. One of them said something that 1192 00:59:04,320 --> 00:59:07,120 Speaker 2: hit us, which is that election officials don't understand what 1193 00:59:07,360 --> 00:59:09,880 Speaker 2: is the potential of what AI can do to cause confusion, 1194 00:59:10,640 --> 00:59:13,560 Speaker 2: and the AI companies don't understand how democracy works. So 1195 00:59:13,600 --> 00:59:16,400 Speaker 2: we can bring these groups together to cross educate, cross trained, 1196 00:59:16,680 --> 00:59:19,840 Speaker 2: to understand each other's risks. That's at least something. 1197 00:59:20,240 --> 00:59:24,200 Speaker 1: And Chris, one of our recommendations from our commission, on 1198 00:59:24,240 --> 00:59:27,520 Speaker 1: which Vivian was a part and I was a co chair, 1199 00:59:27,720 --> 00:59:31,320 Speaker 1: was to open the doors for scientists and researchers to 1200 00:59:31,440 --> 00:59:35,360 Speaker 1: actually study these tech companies. But clearly, Vivian, that hasn't happened, 1201 00:59:35,400 --> 00:59:35,680 Speaker 1: has it? 1202 00:59:36,120 --> 00:59:38,800 Speaker 2: Well? We may need a whole other podcast for that, 1203 00:59:38,920 --> 00:59:41,280 Speaker 2: given the political pressures that are happening on those that 1204 00:59:41,320 --> 00:59:44,880 Speaker 2: are looking into miss and disinformation and the chilling effect 1205 00:59:44,880 --> 00:59:46,680 Speaker 2: that that has, but it's it's troubling. 1206 00:59:47,160 --> 00:59:51,280 Speaker 1: Well. On that note, Happy holidays everybody. Chris and the 1207 00:59:51,400 --> 00:59:54,760 Speaker 1: Loss and Vivian, thank you all so much for this conversation. 1208 00:59:54,880 --> 00:59:58,560 Speaker 1: I hope it's helpful to people who are trying to 1209 00:59:58,560 --> 01:00:03,400 Speaker 1: wrap their arms around this new technology and the ramifications 1210 01:00:03,560 --> 01:00:06,200 Speaker 1: it is going to have on all of us. To 1211 01:00:06,240 --> 01:00:07,960 Speaker 1: all three of you, thank you so much. 1212 01:00:08,240 --> 01:00:16,440 Speaker 4: Thanks for having us, Thank you, Thanks everybody. 1213 01:00:18,480 --> 01:00:21,240 Speaker 1: Vivian, you've become such an expert in this area. Did 1214 01:00:21,280 --> 01:00:24,720 Speaker 1: you hear anything new or interesting or are you as 1215 01:00:24,800 --> 01:00:25,640 Speaker 1: troubled as ever? 1216 01:00:26,720 --> 01:00:29,360 Speaker 2: That's a good question. It's not that I heard anything new, 1217 01:00:29,400 --> 01:00:31,240 Speaker 2: because I spend a lot of time on this space. 1218 01:00:31,520 --> 01:00:34,480 Speaker 2: But what to me was so revealing about this conversation 1219 01:00:35,440 --> 01:00:37,880 Speaker 2: is not sort of all the things that we're worried 1220 01:00:37,880 --> 01:00:41,880 Speaker 2: about are the robot overlords taking over or the deep 1221 01:00:41,920 --> 01:00:43,840 Speaker 2: fake that's going to, you know, make everybody in the 1222 01:00:43,840 --> 01:00:47,160 Speaker 2: world believe it. It's the second and third order effects 1223 01:00:47,440 --> 01:00:52,040 Speaker 2: and the fact that so much control over these incredibly 1224 01:00:52,120 --> 01:00:55,200 Speaker 2: powerful technologies are in the hands of just a few people. 1225 01:00:55,280 --> 01:00:58,480 Speaker 2: I think they both made those points very very strongly, 1226 01:00:59,000 --> 01:01:02,439 Speaker 2: and I think it's it's really hopeful focus on sort 1227 01:01:02,480 --> 01:01:05,760 Speaker 2: of the things that really matter. We get very distracted 1228 01:01:05,760 --> 01:01:09,960 Speaker 2: by shiny objects and maybe not focusing on the fundamentals. 1229 01:01:09,680 --> 01:01:13,959 Speaker 1: Like the telenovella story of Sam Altman, where we need 1230 01:01:14,000 --> 01:01:18,120 Speaker 1: to really focus on the long term implications of all 1231 01:01:18,160 --> 01:01:21,560 Speaker 1: of this. Well, I think they're both really nice, really smart. 1232 01:01:21,680 --> 01:01:24,439 Speaker 1: Thank you for introducing me to them, Vivian, and thank 1233 01:01:24,480 --> 01:01:26,160 Speaker 1: you for being part of the podcast. 1234 01:01:26,760 --> 01:01:29,960 Speaker 2: Well, thank you for letting me share Dinny's with you. Katie. 1235 01:01:30,040 --> 01:01:32,640 Speaker 2: It's an incredibly humbling honor, so thank you so much. 1236 01:01:39,640 --> 01:01:42,880 Speaker 1: Thanks for listening. Everyone. If you have a question for me, 1237 01:01:43,240 --> 01:01:45,760 Speaker 1: a subject you want us to cover, or you want 1238 01:01:45,760 --> 01:01:49,080 Speaker 1: to share your thoughts about how you navigate this crazy world, 1239 01:01:49,480 --> 01:01:52,440 Speaker 1: reach out. You can leave a short message at six 1240 01:01:52,560 --> 01:01:55,640 Speaker 1: oh nine five point two five to five oh five, 1241 01:01:55,960 --> 01:01:58,440 Speaker 1: or you can send me a DM on Instagram. I 1242 01:01:58,480 --> 01:02:01,400 Speaker 1: would love to hear from you. Next Question is a 1243 01:02:01,440 --> 01:02:06,160 Speaker 1: production of iHeartMedia and Katie Couric Media. The executive producers 1244 01:02:06,200 --> 01:02:10,360 Speaker 1: are me, Katie Kuric, and Courtney ltz Our. Supervising producer 1245 01:02:10,560 --> 01:02:14,720 Speaker 1: is Ryan Martx, and our producers are Adriana Fazzio and 1246 01:02:14,840 --> 01:02:20,200 Speaker 1: Meredith Barnes. Julian Weller composed our theme music. For more 1247 01:02:20,200 --> 01:02:23,800 Speaker 1: information about today's episode, or to sign up for my newsletter, 1248 01:02:23,960 --> 01:02:27,280 Speaker 1: wake Up Call, go to the description in the podcast app, 1249 01:02:27,560 --> 01:02:30,800 Speaker 1: or visit us at Katiecuric dot com. You can also 1250 01:02:30,960 --> 01:02:34,120 Speaker 1: find me on Instagram and all my social media channels. 1251 01:02:34,680 --> 01:02:40,160 Speaker 1: For more podcasts from iHeartRadio, visit the iHeartRadio app, Apple Podcasts, 1252 01:02:40,400 --> 01:02:43,080 Speaker 1: or wherever you listen to your favorite shows,