1 00:00:04,440 --> 00:00:12,920 Speaker 1: Welcome to tech Stuff, a production from iHeartRadio. Heydarren, Welcome 2 00:00:12,960 --> 00:00:16,200 Speaker 1: to tech Stuff. I'm your host, Jonathan Strickland. I'm an 3 00:00:16,239 --> 00:00:20,800 Speaker 1: executive producer with iHeartRadio and how the tech are you? So? 4 00:00:21,320 --> 00:00:25,080 Speaker 1: Last week I said I would do an episode about 5 00:00:25,079 --> 00:00:30,440 Speaker 1: doctor Jeffrey Hinton, the so called godfather of AI. My 6 00:00:30,600 --> 00:00:33,360 Speaker 1: dog is very interested in this, as he winds in 7 00:00:33,400 --> 00:00:37,960 Speaker 1: the background, and as I published this. We're into the 8 00:00:38,000 --> 00:00:40,720 Speaker 1: fifth month of twenty twenty three, and I still feel 9 00:00:40,720 --> 00:00:43,960 Speaker 1: pretty good about calling this the Year of AI. While 10 00:00:44,080 --> 00:00:49,159 Speaker 1: artificial intelligence has obviously been a discipline for decades, with 11 00:00:49,880 --> 00:00:54,680 Speaker 1: lots of impressive displays and exhibitions and developments throughout the years, 12 00:00:55,280 --> 00:00:59,240 Speaker 1: the buzz around and attention to aifields has really hit 13 00:00:59,240 --> 00:01:02,840 Speaker 1: a high point this year, largely driven by stuff like 14 00:01:03,320 --> 00:01:07,960 Speaker 1: large language models LMS, as well as the chatbots built 15 00:01:08,000 --> 00:01:12,600 Speaker 1: on top of them that seem to be pretty knowledgeable, 16 00:01:12,680 --> 00:01:17,080 Speaker 1: almost human in their capabilities. Plus you throw in some 17 00:01:17,319 --> 00:01:20,520 Speaker 1: image and video and audio capabilities that allow us to 18 00:01:20,600 --> 00:01:23,319 Speaker 1: use a machine to create all sorts of stuff, and 19 00:01:23,360 --> 00:01:26,120 Speaker 1: you got yourself something that the average person can at 20 00:01:26,240 --> 00:01:30,600 Speaker 1: least recognize as AI. A lot of AI applications historically 21 00:01:31,240 --> 00:01:36,040 Speaker 1: have been so far behind the scenes that you might 22 00:01:36,080 --> 00:01:39,280 Speaker 1: not even recognize it as artificial intelligence, or you might 23 00:01:39,319 --> 00:01:41,800 Speaker 1: not think of it in that context. But now we're 24 00:01:41,800 --> 00:01:45,880 Speaker 1: getting to a point where there's at least the appearance 25 00:01:46,040 --> 00:01:51,280 Speaker 1: of machines behaving similarly to people within certain contexts, and 26 00:01:51,320 --> 00:01:54,840 Speaker 1: it becomes way easier for the average person to say, Wow, 27 00:01:55,560 --> 00:01:58,360 Speaker 1: hang on, what's going on now. I say that because, 28 00:01:58,760 --> 00:02:00,360 Speaker 1: as I have pointed out in this show so many 29 00:02:00,360 --> 00:02:03,560 Speaker 1: many times, there are lots of different aspects of artificial intelligence, 30 00:02:03,960 --> 00:02:07,160 Speaker 1: some of which have been around for many years, and 31 00:02:07,240 --> 00:02:10,480 Speaker 1: some of them have even been causing problems for many years. 32 00:02:11,000 --> 00:02:15,480 Speaker 1: See also facial recognition technology and the fact that bias 33 00:02:15,560 --> 00:02:19,840 Speaker 1: and systems can lead to really terrible consequences in the 34 00:02:19,880 --> 00:02:22,880 Speaker 1: real world. And today I wanted to talk about how 35 00:02:22,919 --> 00:02:24,720 Speaker 1: some of the folks in the AI field are voice 36 00:02:24,720 --> 00:02:29,040 Speaker 1: and concerns that they have around AI and AI's evolution 37 00:02:29,320 --> 00:02:32,840 Speaker 1: and also its deployment and how it could be a 38 00:02:32,880 --> 00:02:35,919 Speaker 1: destructive tool in the future. Now, if you've been listening 39 00:02:35,960 --> 00:02:38,440 Speaker 1: to me for a while, you know I try to 40 00:02:38,480 --> 00:02:42,280 Speaker 1: take a very thoughtful approach to this. I think it's 41 00:02:42,320 --> 00:02:46,200 Speaker 1: important to understand the capabilities of AI, and it's also 42 00:02:46,280 --> 00:02:49,920 Speaker 1: important to understand the potential misuses of AI, or at 43 00:02:50,000 --> 00:02:54,480 Speaker 1: least the unintended consequences of using AI. But I also 44 00:02:54,520 --> 00:02:59,080 Speaker 1: want to try to avoid fud that is, fear, uncertainty 45 00:02:59,240 --> 00:03:03,000 Speaker 1: and doubt that can air on the side of being 46 00:03:03,080 --> 00:03:07,040 Speaker 1: an alarmist. So I think that we should be concerned, 47 00:03:07,440 --> 00:03:10,640 Speaker 1: But so far I haven't been ready to push the 48 00:03:10,720 --> 00:03:14,760 Speaker 1: panic button just yet for AI. But maybe that's about 49 00:03:14,800 --> 00:03:17,800 Speaker 1: to change, because while I've been trying to wrap my 50 00:03:17,880 --> 00:03:21,959 Speaker 1: brain around this, a person like doctor Jeffrey Hinton has 51 00:03:22,000 --> 00:03:24,480 Speaker 1: come forward with his own concerns about AI. And if 52 00:03:24,480 --> 00:03:29,119 Speaker 1: doctor Hinton is concerned, I should probably listen. And that's 53 00:03:29,160 --> 00:03:32,080 Speaker 1: because doctor Hinton has been at the cutting edge of 54 00:03:32,160 --> 00:03:37,320 Speaker 1: AI development for years for decades, particularly in fields like 55 00:03:37,560 --> 00:03:41,920 Speaker 1: artificial neural networks and deep neural networks. In particular, he 56 00:03:42,000 --> 00:03:45,040 Speaker 1: recently resigned from his position at Google, where he had 57 00:03:45,080 --> 00:03:48,480 Speaker 1: been working in AI research, at age seventy five. He's 58 00:03:48,480 --> 00:03:50,720 Speaker 1: certainly at a point in his life where retirement would 59 00:03:50,720 --> 00:03:53,440 Speaker 1: seem pretty natural. You would just think that he would 60 00:03:53,440 --> 00:03:55,680 Speaker 1: come to the conclusion of, yes, it's time for me 61 00:03:55,800 --> 00:03:58,680 Speaker 1: to rest. But his decision was made at least in 62 00:03:58,760 --> 00:04:03,000 Speaker 1: part so that he could speak out about AI and 63 00:04:03,400 --> 00:04:08,120 Speaker 1: the dangers he considers to be important without considering how 64 00:04:08,160 --> 00:04:12,760 Speaker 1: it would impact Google. And that's from doctor Hinton himself. 65 00:04:12,800 --> 00:04:15,480 Speaker 1: He posted that on Twitter, where he said he was 66 00:04:15,960 --> 00:04:19,400 Speaker 1: doing this without considering how it would impact on Google. 67 00:04:19,839 --> 00:04:22,320 Speaker 1: He was addressing a New York Times article that implied 68 00:04:22,360 --> 00:04:25,040 Speaker 1: he had left Google so that he could criticize Google 69 00:04:25,080 --> 00:04:27,760 Speaker 1: in particular. He was quick to say that he felt 70 00:04:27,800 --> 00:04:31,679 Speaker 1: Google had been pretty responsible in its pursuit of AI, 71 00:04:31,880 --> 00:04:36,599 Speaker 1: at least arguably until relatively recently. So let's learn a 72 00:04:36,640 --> 00:04:40,360 Speaker 1: bit about doctor Hinton and his background, the work that 73 00:04:40,400 --> 00:04:44,120 Speaker 1: he pursued, and what his concerns around AI actually cover, 74 00:04:44,560 --> 00:04:47,800 Speaker 1: and maybe along the way we'll figure out some questions 75 00:04:47,839 --> 00:04:52,120 Speaker 1: that we need to answer at some point, implications that 76 00:04:52,200 --> 00:04:56,120 Speaker 1: need to be considered, and perhaps choices we absolutely should 77 00:04:56,160 --> 00:04:59,039 Speaker 1: not make if we want to create helpful AI that 78 00:04:59,080 --> 00:05:02,240 Speaker 1: provides a net benefit rather than something that you know, 79 00:05:03,120 --> 00:05:07,160 Speaker 1: creates the terminator or how or whatever. And I am 80 00:05:07,200 --> 00:05:10,200 Speaker 1: being a bit flippant, but there are reasons we should 81 00:05:10,279 --> 00:05:14,279 Speaker 1: have some concerns, even if they don't involve single minded 82 00:05:14,440 --> 00:05:20,119 Speaker 1: cyborg soldiers. Jeffrey Hinton was born in nineteen forty seven 83 00:05:20,279 --> 00:05:24,800 Speaker 1: in London, England. He attended the University of Edinburgh and 84 00:05:24,880 --> 00:05:28,400 Speaker 1: graduated with a degree in psychology in nineteen sixty nine, 85 00:05:28,440 --> 00:05:31,240 Speaker 1: which is an interesting starting point for someone who had 86 00:05:31,240 --> 00:05:35,080 Speaker 1: become deeply involved in computer science, and the background in 87 00:05:35,120 --> 00:05:38,400 Speaker 1: psychology is probably an important component for someone who would 88 00:05:38,440 --> 00:05:42,560 Speaker 1: contribute to the advancement of artificial intelligence in general and 89 00:05:42,720 --> 00:05:46,560 Speaker 1: neural networks in particular. In nineteen seventy eight, he earned 90 00:05:46,560 --> 00:05:50,479 Speaker 1: a PhD in artificial intelligence at the University of Sussex. 91 00:05:50,920 --> 00:05:54,880 Speaker 1: He transitioned into being an AI researcher, but it was 92 00:05:54,960 --> 00:05:57,440 Speaker 1: kind of a tough go in the UK. There just 93 00:05:57,520 --> 00:06:01,320 Speaker 1: really wasn't that much support and funding for AI research 94 00:06:01,480 --> 00:06:04,400 Speaker 1: over in the UK, so it was hard for him 95 00:06:04,440 --> 00:06:07,480 Speaker 1: to make much progress. So he decided to immigrate to 96 00:06:07,520 --> 00:06:10,680 Speaker 1: the United States, where he first worked as a researcher 97 00:06:10,720 --> 00:06:13,880 Speaker 1: at the University of California in San Diego, and then 98 00:06:13,920 --> 00:06:18,039 Speaker 1: he moved on to Carnegie Mellon University and he worked 99 00:06:18,080 --> 00:06:20,960 Speaker 1: at Carnegie Mellon as a professor from nineteen eighty two 100 00:06:21,080 --> 00:06:24,360 Speaker 1: to nineteen eighty seven, but by the late eighties he 101 00:06:24,440 --> 00:06:29,200 Speaker 1: made a decision to relocate to Canada and you might say, well, 102 00:06:29,200 --> 00:06:32,039 Speaker 1: why would you go to Canada when you were already 103 00:06:32,040 --> 00:06:34,240 Speaker 1: working in AI in the United States. I mean, the 104 00:06:34,320 --> 00:06:39,320 Speaker 1: US has spent billions of dollars in research and development 105 00:06:39,440 --> 00:06:44,400 Speaker 1: in the technology field. Well, his primary reason was because 106 00:06:44,480 --> 00:06:47,640 Speaker 1: the main source for research funding in AI at the 107 00:06:47,680 --> 00:06:51,400 Speaker 1: time came from the Department of Defense, and doctor Hinton 108 00:06:51,600 --> 00:06:54,520 Speaker 1: wasn't comfortable with the idea of working on machine intelligence 109 00:06:55,160 --> 00:06:59,960 Speaker 1: that was through military backing, because the presumption is whatever 110 00:07:00,000 --> 00:07:03,279 Speaker 1: our work you create is ultimately going to be put 111 00:07:03,320 --> 00:07:06,200 Speaker 1: to use by the Department of Defense, and it's reasonable 112 00:07:06,240 --> 00:07:09,920 Speaker 1: to assume that at least some of those uses could 113 00:07:09,920 --> 00:07:13,720 Speaker 1: be weaponized, and Hinton didn't want to contribute to work 114 00:07:13,760 --> 00:07:17,400 Speaker 1: that could later be used to harm or kill others, 115 00:07:17,880 --> 00:07:21,720 Speaker 1: so he would rather sidestep that and get funding from 116 00:07:21,760 --> 00:07:26,720 Speaker 1: other sources. So he settled in Toronto, Canada. He took 117 00:07:26,760 --> 00:07:30,600 Speaker 1: on more academic roles. He continued to be professor. He 118 00:07:30,680 --> 00:07:33,040 Speaker 1: also continued to work in the field of AI research, 119 00:07:33,160 --> 00:07:37,600 Speaker 1: specifically in artificial neural networks and deep learning approaches, and 120 00:07:37,640 --> 00:07:40,320 Speaker 1: we will talk more about those in just a little bit. 121 00:07:40,800 --> 00:07:44,239 Speaker 1: In twenty twelve, he co founded a company with two 122 00:07:44,360 --> 00:07:49,080 Speaker 1: of his students after publishing a paper on deep learning. 123 00:07:49,440 --> 00:07:53,320 Speaker 1: So his paper got the attention of some really smart 124 00:07:53,360 --> 00:07:57,000 Speaker 1: people around the world, and before Hinton knew it, he 125 00:07:57,120 --> 00:08:00,640 Speaker 1: was being courted by some really big companies, companies that 126 00:08:00,680 --> 00:08:04,520 Speaker 1: had super deep pockets and wanted to hire him and 127 00:08:04,560 --> 00:08:07,560 Speaker 1: a couple of his students on to work in the 128 00:08:07,560 --> 00:08:11,120 Speaker 1: field of AI research. So he got one offer from 129 00:08:11,200 --> 00:08:14,080 Speaker 1: the Chinese company Baidu that would have had him and 130 00:08:14,160 --> 00:08:16,760 Speaker 1: his two students work for the company for a few 131 00:08:16,840 --> 00:08:21,360 Speaker 1: years in return for twelve million smackaroo's worth of compensation. 132 00:08:22,120 --> 00:08:26,280 Speaker 1: That's a healthy salary. But Hinton also had other folks 133 00:08:26,480 --> 00:08:29,000 Speaker 1: who were potentially interested in his work, and he also 134 00:08:29,080 --> 00:08:33,000 Speaker 1: figured it would be far more lucrative if he created 135 00:08:33,080 --> 00:08:37,120 Speaker 1: a company with his students, if they made a company 136 00:08:37,160 --> 00:08:41,200 Speaker 1: together that could then be acquired, so instead of getting 137 00:08:41,320 --> 00:08:44,520 Speaker 1: hired as individuals, they would have a company that would 138 00:08:44,520 --> 00:08:46,800 Speaker 1: have to be purchased. And so that's when he and 139 00:08:46,920 --> 00:08:52,520 Speaker 1: these two students incorporated into DNN Research. The DNN stands 140 00:08:52,520 --> 00:08:57,760 Speaker 1: for deep Neural Networks. Hinton then took this brand new 141 00:08:57,800 --> 00:09:02,640 Speaker 1: company which really just had three employees, including himself, and 142 00:09:02,760 --> 00:09:06,800 Speaker 1: had no products and no services, no business plan, nothing 143 00:09:06,840 --> 00:09:09,760 Speaker 1: other than the fact that it was incorporated, and then 144 00:09:09,760 --> 00:09:13,560 Speaker 1: he put it up for auction. The actual auction took 145 00:09:13,600 --> 00:09:18,560 Speaker 1: place in Lake Tahoe during a conference on AI and 146 00:09:18,640 --> 00:09:24,120 Speaker 1: machine learning, and by Do participated, but so did Microsoft, Google, 147 00:09:24,320 --> 00:09:27,960 Speaker 1: and an AI research company called DeepMind, which a couple 148 00:09:28,000 --> 00:09:32,600 Speaker 1: of years later would become a Google subsidiary of its own. Now, 149 00:09:32,640 --> 00:09:36,280 Speaker 1: DeepMind was the first company to bow out of the auction. 150 00:09:36,400 --> 00:09:40,040 Speaker 1: It just did not have the resources of these three 151 00:09:40,240 --> 00:09:45,360 Speaker 1: giant tech companies. Microsoft then followed. Actually, Microsoft kind of 152 00:09:45,360 --> 00:09:47,120 Speaker 1: bounced in and out of the auction a couple of 153 00:09:47,160 --> 00:09:50,959 Speaker 1: times before finally throwing in the towel, and the bidding 154 00:09:51,040 --> 00:09:53,360 Speaker 1: war came down to Google versus by Do, and it 155 00:09:53,440 --> 00:09:56,280 Speaker 1: just kept going and going and going. Once the price 156 00:09:56,400 --> 00:10:00,640 Speaker 1: hit an astounding forty four million dollars. And keep in 157 00:10:00,679 --> 00:10:04,520 Speaker 1: mind DNN Research had only been around for a very 158 00:10:04,559 --> 00:10:07,640 Speaker 1: short while and had no products or services to its name, 159 00:10:08,320 --> 00:10:11,240 Speaker 1: doctor Hinton called the auction closed and the company went 160 00:10:11,280 --> 00:10:14,720 Speaker 1: to its new owner, that of Google. Reportedly, the people 161 00:10:14,760 --> 00:10:18,079 Speaker 1: at Google were actually surprised that he stopped the auction 162 00:10:18,200 --> 00:10:21,400 Speaker 1: at that point, because they figured that he was leaving 163 00:10:21,440 --> 00:10:23,960 Speaker 1: millions of dollars on the table that the bidding war 164 00:10:24,080 --> 00:10:27,160 Speaker 1: would have continued between Google and by Doo, and he 165 00:10:27,200 --> 00:10:30,320 Speaker 1: could have gotten more for it, but doctor Hinton was 166 00:10:30,440 --> 00:10:33,439 Speaker 1: more concerned with working for Google rather than for by 167 00:10:33,520 --> 00:10:36,000 Speaker 1: Do and felt that forty four million dollars was more 168 00:10:36,080 --> 00:10:40,760 Speaker 1: than enough. So that's a pretty you know, mature approach 169 00:10:40,840 --> 00:10:44,000 Speaker 1: as opposed to let's take every cent we can grab. 170 00:10:44,559 --> 00:10:48,200 Speaker 1: Also in twenty twelve, he won an award when he 171 00:10:48,320 --> 00:10:52,280 Speaker 1: co invented a deep learning model called alex Net, named 172 00:10:52,280 --> 00:10:56,840 Speaker 1: after the other co creator, his student, Alex Krzewski. This 173 00:10:57,000 --> 00:10:59,200 Speaker 1: was bigger than just a two person operation. By the way, 174 00:10:59,240 --> 00:11:03,480 Speaker 1: it's not like Alex Krzewski and doctor Hinton were the 175 00:11:03,480 --> 00:11:05,560 Speaker 1: only two to work on it, but they were the 176 00:11:05,640 --> 00:11:10,080 Speaker 1: leads on this project and it was named after Alex. Alex, 177 00:11:10,080 --> 00:11:11,720 Speaker 1: by the way, was also one of the two students 178 00:11:11,720 --> 00:11:15,080 Speaker 1: who was part of DNN research, the other being a 179 00:11:15,120 --> 00:11:20,800 Speaker 1: student named Iliya Sutzkever. And my apologies for the butchering 180 00:11:20,840 --> 00:11:26,120 Speaker 1: of pronunciation, but the learning model Alex Natt focused quite literally, 181 00:11:26,360 --> 00:11:30,080 Speaker 1: i guess you could say, on image recognition and participated 182 00:11:30,120 --> 00:11:32,720 Speaker 1: in a competition in which the model proved to have 183 00:11:32,760 --> 00:11:36,440 Speaker 1: an eighty five percent accuracy rate. And while it's a 184 00:11:36,440 --> 00:11:38,480 Speaker 1: trivial thing for a human to look at a photo 185 00:11:38,600 --> 00:11:41,560 Speaker 1: and say something like that's a bunny rabbit. It's not 186 00:11:41,720 --> 00:11:43,960 Speaker 1: so trivial to create a way for computers to be 187 00:11:44,040 --> 00:11:47,559 Speaker 1: able to do the same thing. So this eighty five 188 00:11:47,600 --> 00:11:52,040 Speaker 1: percent accuracy rate was like that was like a stake 189 00:11:52,120 --> 00:11:56,000 Speaker 1: in the ground saying we have made a massive leap 190 00:11:56,000 --> 00:11:58,920 Speaker 1: ahead with machine learning and artificial intelligence. It was one 191 00:11:58,920 --> 00:12:03,000 Speaker 1: of the reasons why DNA research was so highly sought after, 192 00:12:03,559 --> 00:12:06,839 Speaker 1: and alex Nett wasn't just an impressive approach toward machine learning. 193 00:12:08,280 --> 00:12:11,200 Speaker 1: It really got enough buzz that money began to pour 194 00:12:11,280 --> 00:12:16,200 Speaker 1: into deep learning projects everywhere, not just with doctor Hinton 195 00:12:16,240 --> 00:12:20,080 Speaker 1: and his students, Like we literally started to see more 196 00:12:20,120 --> 00:12:24,240 Speaker 1: development in the discipline as a whole because this was 197 00:12:24,240 --> 00:12:29,000 Speaker 1: such an impressive display. From twenty twelve until just this year, 198 00:12:29,760 --> 00:12:32,679 Speaker 1: doctor Hinton worked in AI research and deep learning, in 199 00:12:32,720 --> 00:12:37,280 Speaker 1: particular over at Google. One of his two students, Ilijas Skiver, 200 00:12:37,640 --> 00:12:41,160 Speaker 1: actually would leave Google to join a little AI nonprofit 201 00:12:41,280 --> 00:12:45,079 Speaker 1: called open Ai. I mean, originally it was a nonprofit, 202 00:12:45,120 --> 00:12:48,320 Speaker 1: and technically the nonprofit part of open ai is still 203 00:12:48,320 --> 00:12:52,720 Speaker 1: a parent organization, but really the for profit arm of 204 00:12:52,720 --> 00:12:56,079 Speaker 1: open AI is in the news way more frequently these days. 205 00:12:56,559 --> 00:12:59,560 Speaker 1: In twenty eighteen, doctor Hinton was a co recipient of 206 00:12:59,600 --> 00:13:03,640 Speaker 1: the award. This is a prestigious honor for those in 207 00:13:03,679 --> 00:13:06,600 Speaker 1: the computing field. Some people even refer to it as 208 00:13:06,640 --> 00:13:11,080 Speaker 1: being the equivalent to a Nobel Prize. And now he's 209 00:13:11,280 --> 00:13:15,520 Speaker 1: stepping forward with concerns relating to the work he dedicated 210 00:13:15,840 --> 00:13:19,000 Speaker 1: his life too. Now we're going to take a quick break. 211 00:13:19,000 --> 00:13:22,640 Speaker 1: When we come back, we're going to talk about deep 212 00:13:22,720 --> 00:13:26,240 Speaker 1: neural networks and what they do within the realm of 213 00:13:26,360 --> 00:13:40,040 Speaker 1: machine learning. But first these messages. Okay, we're back. Let's 214 00:13:40,120 --> 00:13:43,439 Speaker 1: talk about doctor Hinton's work and deep neural networks. Now, 215 00:13:43,480 --> 00:13:47,640 Speaker 1: as you might imagine, this subject gets really complicated. It's 216 00:13:47,679 --> 00:13:51,480 Speaker 1: really nuanced, it's technical, and as I'm sure you have 217 00:13:51,559 --> 00:13:55,120 Speaker 1: no need to imagine, my understanding of deep neural networks 218 00:13:55,200 --> 00:13:57,959 Speaker 1: is pretty limited. I mean, you could call it surface 219 00:13:58,040 --> 00:14:00,440 Speaker 1: level and I wouldn't be able to disagree. So we're 220 00:14:00,440 --> 00:14:04,720 Speaker 1: going to paint this topic in broad strokes. And I'm 221 00:14:04,720 --> 00:14:08,160 Speaker 1: doing this not to dumb it down, but rather to 222 00:14:08,200 --> 00:14:10,360 Speaker 1: do my best to kind of get across the general 223 00:14:10,440 --> 00:14:14,199 Speaker 1: way it works without making too many egregious errors. Along 224 00:14:14,240 --> 00:14:18,800 Speaker 1: the way. So first up, the goal of a deep 225 00:14:18,960 --> 00:14:23,600 Speaker 1: neural network is to provide a learning mechanism that mimics 226 00:14:23,640 --> 00:14:27,920 Speaker 1: the human brain, but using a computer rather than a 227 00:14:27,960 --> 00:14:31,520 Speaker 1: human brain. So, for the purposes of an overly simple 228 00:14:31,560 --> 00:14:35,720 Speaker 1: thought experiment, imagine you've got a black box. It's an 229 00:14:35,760 --> 00:14:39,320 Speaker 1: opaque black box. Now, one side of the box allows 230 00:14:39,360 --> 00:14:42,000 Speaker 1: you to put something in, and the other side of 231 00:14:42,000 --> 00:14:46,640 Speaker 1: the box allows stuff to come out. And let's say 232 00:14:46,640 --> 00:14:51,920 Speaker 1: that you are putting in one thing and it transforms 233 00:14:51,960 --> 00:14:53,800 Speaker 1: in some way inside the box and comes out as 234 00:14:53,800 --> 00:14:57,520 Speaker 1: something else. That's the general thought here, because I'm feeling 235 00:14:57,800 --> 00:15:01,800 Speaker 1: a little peckish and a little puckish. Let's say that 236 00:15:01,880 --> 00:15:06,000 Speaker 1: you decide to put in the inputs as the ingredients 237 00:15:06,000 --> 00:15:08,080 Speaker 1: you would need for a pizza, and you're shoving that 238 00:15:08,120 --> 00:15:10,920 Speaker 1: into the box. So we're talking stuff like pizza dough 239 00:15:11,400 --> 00:15:15,400 Speaker 1: and some sauce and some cheese and any toppings you like, 240 00:15:16,040 --> 00:15:18,280 Speaker 1: and you shove that into the input in the box, 241 00:15:18,280 --> 00:15:21,600 Speaker 1: and then the output shoots out a cowl zone. Well shucks, 242 00:15:22,080 --> 00:15:24,720 Speaker 1: you think, unless you're Ben Wyatt from Parks and rec 243 00:15:24,760 --> 00:15:27,400 Speaker 1: in which case you celebrate because you think col zones 244 00:15:27,440 --> 00:15:30,040 Speaker 1: are superior to pizza in every way. But assuming you're 245 00:15:30,120 --> 00:15:33,720 Speaker 1: not Ben Wyatt, you say, that's not what I wanted. 246 00:15:33,880 --> 00:15:36,400 Speaker 1: I wanted to get a pizza, not a calzone. So 247 00:15:36,440 --> 00:15:38,680 Speaker 1: you have to open up the box, you have to 248 00:15:38,720 --> 00:15:41,560 Speaker 1: adjust some stuff inside it, you have to close it 249 00:15:41,600 --> 00:15:44,040 Speaker 1: all up and try it again, and you keep doing 250 00:15:44,080 --> 00:15:46,600 Speaker 1: this over and over until you get a pizza, a 251 00:15:46,720 --> 00:15:50,680 Speaker 1: properly cooked and prepared pizza. This is sort of similar 252 00:15:50,720 --> 00:15:56,320 Speaker 1: to how computer scientists perform supervised learning with artificial neural networks, 253 00:15:56,760 --> 00:16:01,800 Speaker 1: because that box represents what we call hidden layers. They 254 00:16:01,840 --> 00:16:05,000 Speaker 1: could be lots of hidden layers, and these are layers 255 00:16:05,080 --> 00:16:10,160 Speaker 1: of computer nodes that serve as artificial neurons, and pathways 256 00:16:10,280 --> 00:16:14,760 Speaker 1: form between these different nodes as they process information. So 257 00:16:15,080 --> 00:16:18,680 Speaker 1: when you put input into the system, that input goes 258 00:16:18,720 --> 00:16:21,240 Speaker 1: to a node and it begins to sort the data 259 00:16:21,400 --> 00:16:24,560 Speaker 1: based on some criteria that the system has been trained on. 260 00:16:24,640 --> 00:16:28,000 Speaker 1: Whatever the purpose of the system actually is. It's kind 261 00:16:28,040 --> 00:16:30,880 Speaker 1: of like, you know, let's say it's for recognizing bunnies, 262 00:16:30,920 --> 00:16:33,880 Speaker 1: since we use that example earlier. So you feed it 263 00:16:33,960 --> 00:16:37,240 Speaker 1: a whole bunch of images, and the node takes the 264 00:16:37,320 --> 00:16:40,360 Speaker 1: data and passes the data to another node. A layer 265 00:16:40,440 --> 00:16:43,600 Speaker 1: down and it does it. It chooses the node based 266 00:16:43,680 --> 00:16:48,720 Speaker 1: on some transformational function at that artificial neuron, right, so 267 00:16:48,760 --> 00:16:52,320 Speaker 1: you can think data comes in, the neuron, performs a 268 00:16:52,440 --> 00:16:55,800 Speaker 1: transformational function on this data. Based on that result, it 269 00:16:55,840 --> 00:16:59,640 Speaker 1: goes to one node or another, and then the process 270 00:16:59,680 --> 00:17:02,200 Speaker 1: repeat and it does this again and again until it 271 00:17:02,240 --> 00:17:06,480 Speaker 1: comes out the output side. Where like in our example, 272 00:17:07,480 --> 00:17:10,360 Speaker 1: we figure out whether the machine is able to recognize 273 00:17:10,400 --> 00:17:12,639 Speaker 1: if a picture has a bunny in it or not. 274 00:17:13,359 --> 00:17:18,439 Speaker 1: So you feed millions, tens of millions, hundreds of millions 275 00:17:18,480 --> 00:17:22,879 Speaker 1: of pictures to this system to train it. When you 276 00:17:22,920 --> 00:17:26,280 Speaker 1: start off, you might be doing this with a bunch 277 00:17:26,320 --> 00:17:29,080 Speaker 1: of images that you've already determined whether or not there 278 00:17:29,080 --> 00:17:33,360 Speaker 1: are bunnies in them. So you've got to control amount 279 00:17:33,359 --> 00:17:36,040 Speaker 1: of data that you're feeding just for the purposes of 280 00:17:36,080 --> 00:17:40,159 Speaker 1: training your system. And at the end, after it's sorted 281 00:17:40,200 --> 00:17:43,320 Speaker 1: through those images, you evaluate the system to see how 282 00:17:43,359 --> 00:17:46,639 Speaker 1: well it did in figuring out whether an image had 283 00:17:46,640 --> 00:17:48,359 Speaker 1: a bunny in it or not. Maybe in some of 284 00:17:48,400 --> 00:17:51,560 Speaker 1: the pictures it misses a bunny. Maybe in some pictures 285 00:17:51,640 --> 00:17:54,360 Speaker 1: it thinks there's a bunny there and there's not. And 286 00:17:54,400 --> 00:17:57,560 Speaker 1: then you might go in and start to adjust the 287 00:17:57,600 --> 00:18:01,520 Speaker 1: weights on those artificial neurons. This is the thing that 288 00:18:01,800 --> 00:18:06,520 Speaker 1: creates that transformational function. You might tweak those transformational functions 289 00:18:06,560 --> 00:18:09,680 Speaker 1: a little bit. You might start closest to the output 290 00:18:09,720 --> 00:18:12,760 Speaker 1: and work your way back. That's called back propagating. And 291 00:18:13,000 --> 00:18:16,920 Speaker 1: what you're trying to do is adjust all these settings 292 00:18:17,240 --> 00:18:19,800 Speaker 1: so that it is more accurate the next time you 293 00:18:19,840 --> 00:18:22,200 Speaker 1: feed all the images through, and you do it again, 294 00:18:22,840 --> 00:18:26,520 Speaker 1: and you might do this dozens or hundreds of times 295 00:18:26,600 --> 00:18:29,239 Speaker 1: in an effort to really refine your model so that 296 00:18:29,280 --> 00:18:32,560 Speaker 1: it gets better and better at identifying the pictures that 297 00:18:32,600 --> 00:18:36,000 Speaker 1: have bunnies in them. And then ideally you get the 298 00:18:36,040 --> 00:18:37,879 Speaker 1: system to a point where you could just feed it 299 00:18:38,080 --> 00:18:40,960 Speaker 1: raw data, like you haven't even looked at these images. 300 00:18:41,000 --> 00:18:45,080 Speaker 1: You're just dumping millions of images in and you're letting 301 00:18:45,160 --> 00:18:47,680 Speaker 1: it sort it through. And because it has reached the 302 00:18:47,800 --> 00:18:51,800 Speaker 1: level of accuracy that it's at, because you've trained it 303 00:18:51,840 --> 00:18:54,119 Speaker 1: for so long, you don't even have to worry so 304 00:18:54,280 --> 00:18:56,880 Speaker 1: much about whether or not it caught all the images 305 00:18:57,200 --> 00:19:00,359 Speaker 1: or if it misidentified some there's probably going to be 306 00:19:00,400 --> 00:19:02,920 Speaker 1: some error in there, but if your accuracy level is 307 00:19:02,960 --> 00:19:06,399 Speaker 1: high enough, then it's possibly good enough for whatever purpose 308 00:19:06,480 --> 00:19:09,679 Speaker 1: you've built it for. And yeah, the more data you 309 00:19:09,840 --> 00:19:13,439 Speaker 1: use to train your machine learning model in general, the 310 00:19:13,560 --> 00:19:17,240 Speaker 1: better it will perform, because it'll start to eliminate things 311 00:19:17,280 --> 00:19:20,879 Speaker 1: like outliers. And while image recognition is just one of 312 00:19:20,920 --> 00:19:23,880 Speaker 1: the more famous uses for deep neural networks in machine learning, 313 00:19:23,920 --> 00:19:26,760 Speaker 1: it is clearly not the only one. The one we've 314 00:19:26,760 --> 00:19:30,840 Speaker 1: been hearing about a lot lately involves large language models 315 00:19:30,960 --> 00:19:33,200 Speaker 1: or llms, like I mentioned at the top of the show. 316 00:19:33,280 --> 00:19:38,720 Speaker 1: So imagine feeding millions or even billions of documents to 317 00:19:38,800 --> 00:19:43,560 Speaker 1: a neural network that's trained to recognize patterns in language. 318 00:19:44,000 --> 00:19:46,679 Speaker 1: So you're feeding all sorts of stuff to this model, 319 00:19:47,240 --> 00:19:51,720 Speaker 1: and as you do, the system quote unquote learns how 320 00:19:52,000 --> 00:19:55,639 Speaker 1: words follow each other, like which words are likely to 321 00:19:55,680 --> 00:19:58,359 Speaker 1: follow other words. You probably wouldn't go so far as 322 00:19:58,359 --> 00:20:03,240 Speaker 1: to say the system under stands a language like English, 323 00:20:03,280 --> 00:20:06,320 Speaker 1: but it does have an incredibly sophisticated statistical model that 324 00:20:06,400 --> 00:20:09,680 Speaker 1: breaks down how likely one word is to follow another. 325 00:20:10,520 --> 00:20:12,240 Speaker 1: So you can think of it a little bit like 326 00:20:12,280 --> 00:20:15,520 Speaker 1: a word association game. You've probably played something like this 327 00:20:15,560 --> 00:20:17,760 Speaker 1: at some point or another. Someone gives you a word 328 00:20:18,160 --> 00:20:20,120 Speaker 1: and you're supposed to say the first word that comes 329 00:20:20,119 --> 00:20:22,480 Speaker 1: to your mind. So if I were to say the 330 00:20:22,520 --> 00:20:27,720 Speaker 1: word nuclear, you might think power or bomb or radiation. 331 00:20:28,760 --> 00:20:34,960 Speaker 1: You probably wouldn't think penguin or Chesterfield sofa just not 332 00:20:35,160 --> 00:20:38,000 Speaker 1: likely to pop up statistically, it's unlikely. Well, you can 333 00:20:38,080 --> 00:20:40,640 Speaker 1: kind of think of the large language model as being 334 00:20:41,200 --> 00:20:45,119 Speaker 1: an enormous version of that. So as these large language 335 00:20:45,119 --> 00:20:49,000 Speaker 1: models process increasing amounts of information, and as the neural 336 00:20:49,080 --> 00:20:54,879 Speaker 1: network experiences refinement over countless learning runs, you end up 337 00:20:54,880 --> 00:20:57,320 Speaker 1: with a system that is capable of doing some pretty 338 00:20:57,400 --> 00:21:00,760 Speaker 1: extraordinary things, at least on the surface life. It can 339 00:21:00,840 --> 00:21:06,000 Speaker 1: pull information together to answer questions about practically any topic. Unfortunately, 340 00:21:06,760 --> 00:21:11,000 Speaker 1: it can also invent answers by following a statistical probability 341 00:21:11,400 --> 00:21:14,720 Speaker 1: when it doesn't actually contain the answers to the question 342 00:21:14,760 --> 00:21:17,080 Speaker 1: you asked. This means you can end up with an 343 00:21:17,119 --> 00:21:21,000 Speaker 1: answer that isn't accurate at all, but it follows a 344 00:21:21,040 --> 00:21:24,919 Speaker 1: statistical model where each word is from a probability standpoint, 345 00:21:25,480 --> 00:21:29,800 Speaker 1: the perfect word to go in that point in a sentence, 346 00:21:29,840 --> 00:21:31,639 Speaker 1: which is a weird thing to think about, right, Like, 347 00:21:31,920 --> 00:21:35,560 Speaker 1: the answer you get isn't right, but each word is, 348 00:21:35,760 --> 00:21:40,040 Speaker 1: statistically speaking, the best one to put in that place 349 00:21:40,280 --> 00:21:45,680 Speaker 1: in lieu of any actual information. Now here's another thing 350 00:21:45,720 --> 00:21:49,159 Speaker 1: to consider. These tools can do stuff like build code. 351 00:21:49,880 --> 00:21:53,239 Speaker 1: This code isn't always reliable, it's not always right, but 352 00:21:53,640 --> 00:21:56,760 Speaker 1: sometimes it is. So maybe you use a tool like 353 00:21:56,880 --> 00:22:00,879 Speaker 1: GPT to look over the code that was made by 354 00:22:00,920 --> 00:22:03,119 Speaker 1: a group of engineers, and you do it to search 355 00:22:03,119 --> 00:22:06,680 Speaker 1: for errors, like you're using this to look for mistakes 356 00:22:06,680 --> 00:22:09,080 Speaker 1: that were made in the code. Or maybe you use 357 00:22:09,160 --> 00:22:11,280 Speaker 1: it to see if there's a way to make the 358 00:22:11,280 --> 00:22:16,679 Speaker 1: code that was written more elegant or efficient. Maybe you 359 00:22:16,720 --> 00:22:19,080 Speaker 1: figure you've reached a point where you don't even need 360 00:22:19,440 --> 00:22:23,520 Speaker 1: human engineers because the AI agent performs at a standard 361 00:22:23,600 --> 00:22:27,440 Speaker 1: that's high enough to replace them. Maybe you think it's 362 00:22:27,440 --> 00:22:30,359 Speaker 1: even better than what human engineers can do, and that 363 00:22:30,400 --> 00:22:33,520 Speaker 1: it's far faster, and that you can therefore develop and 364 00:22:33,600 --> 00:22:37,080 Speaker 1: deployee software at a pace that you couldn't before. So 365 00:22:37,160 --> 00:22:41,280 Speaker 1: the IT industry is in a particularly delicate place as 366 00:22:41,320 --> 00:22:45,880 Speaker 1: companies begin to explore how AI could augment or potentially 367 00:22:46,440 --> 00:22:51,200 Speaker 1: replace people. I go back to what IBM's CEO recently said. 368 00:22:51,480 --> 00:22:55,280 Speaker 1: He said that for nearly eight thousand job offerings that 369 00:22:55,320 --> 00:22:58,960 Speaker 1: the company has now put on a hiring freeze, he 370 00:22:59,119 --> 00:23:02,560 Speaker 1: might never hire a human to take one of those jobs. Instead, 371 00:23:02,600 --> 00:23:06,439 Speaker 1: he might rely on automation and AI to cover that job. 372 00:23:06,840 --> 00:23:10,240 Speaker 1: So it's not quite the same thing as firing someone 373 00:23:10,280 --> 00:23:12,720 Speaker 1: and then replacing him with a robot, but it is 374 00:23:12,800 --> 00:23:16,000 Speaker 1: given a robot a job instead of a human being. Okay, 375 00:23:16,640 --> 00:23:19,960 Speaker 1: let's switch gears. Let's talk about AI in the arts, 376 00:23:20,040 --> 00:23:24,119 Speaker 1: because that's also a really relevant conversation right now. So 377 00:23:24,359 --> 00:23:26,919 Speaker 1: last year we already started to see debates about the 378 00:23:27,000 --> 00:23:32,760 Speaker 1: validity of AI generated images. Should an AI generated image 379 00:23:32,800 --> 00:23:38,600 Speaker 1: be considered art? We saw people submit AI generated paintings 380 00:23:38,680 --> 00:23:44,040 Speaker 1: into competitions, some of which ended up receiving awards, and 381 00:23:44,080 --> 00:23:47,280 Speaker 1: then we're subsequently either stripped of those awards or you know, 382 00:23:47,320 --> 00:23:50,040 Speaker 1: people got in trouble for using AI even when they 383 00:23:50,040 --> 00:23:53,560 Speaker 1: were you know, admitting to it in an effort to say, hey, 384 00:23:53,640 --> 00:23:56,199 Speaker 1: we're trying to start a conversation about AI and its 385 00:23:56,280 --> 00:24:00,720 Speaker 1: role in arts. So is art actually art if the 386 00:24:00,800 --> 00:24:03,760 Speaker 1: image is a product of a complex series of decisions 387 00:24:04,080 --> 00:24:08,640 Speaker 1: that aren't driven by imagination or creativity, but rather some 388 00:24:09,880 --> 00:24:14,200 Speaker 1: really weird statistical model that's so complicated that no one 389 00:24:14,280 --> 00:24:18,920 Speaker 1: really understands it. Or is it just a meaningless image? 390 00:24:19,240 --> 00:24:23,080 Speaker 1: You know, maybe it's an image that mimics specific artists, 391 00:24:23,160 --> 00:24:26,119 Speaker 1: but in itself it's nothing more than just a picture. 392 00:24:26,520 --> 00:24:29,879 Speaker 1: I mean, you know, drawing a perfect circle freehand with 393 00:24:30,000 --> 00:24:32,560 Speaker 1: no tools is really really hard for a human to do, 394 00:24:32,800 --> 00:24:34,520 Speaker 1: but it's a piece of cake for a computer. So 395 00:24:34,520 --> 00:24:38,000 Speaker 1: should we be astounded by a computer's ability to generate 396 00:24:38,000 --> 00:24:41,800 Speaker 1: a perfect circle? What about a computer's ability to mimic 397 00:24:42,040 --> 00:24:47,840 Speaker 1: the style of say, you know, Picasso or Dali. Beyond 398 00:24:47,920 --> 00:24:51,200 Speaker 1: visual arts, there are examples like writing and music. There's 399 00:24:51,240 --> 00:24:53,520 Speaker 1: the case of the song Hard on My Sleeve that 400 00:24:53,720 --> 00:24:57,000 Speaker 1: features the deep fake voices of Drake and the Weekend, 401 00:24:57,720 --> 00:25:00,680 Speaker 1: so it sounds like Drake in the Weekend the song. 402 00:25:00,800 --> 00:25:04,480 Speaker 1: But these are just computer generated voices. So what happens 403 00:25:04,800 --> 00:25:07,920 Speaker 1: when people can create new songs that feature an imitation 404 00:25:08,080 --> 00:25:11,560 Speaker 1: of an established artist's voice or style. You could have 405 00:25:11,600 --> 00:25:13,520 Speaker 1: fun finding out what it would sound like if the 406 00:25:13,560 --> 00:25:16,040 Speaker 1: Beatles wrote a song in the style of the Ramones. 407 00:25:16,680 --> 00:25:20,720 Speaker 1: But this kind of distraction can become really harmful to 408 00:25:20,960 --> 00:25:26,040 Speaker 1: actual human artists. Honestly, what this illustrates is a need 409 00:25:26,080 --> 00:25:29,439 Speaker 1: to create more comprehensive right to publicity and right to 410 00:25:29,520 --> 00:25:34,640 Speaker 1: personality laws to protect people from being imitated without their consent. 411 00:25:35,440 --> 00:25:38,399 Speaker 1: Going a bit further, recently, Spotify had to purge a 412 00:25:38,440 --> 00:25:42,200 Speaker 1: whole bunch of songs from its streaming service because AI 413 00:25:42,480 --> 00:25:45,560 Speaker 1: was gaming the system. So there's this company called Boomy, 414 00:25:46,200 --> 00:25:49,240 Speaker 1: and Boomy lets you create a song based on a prompt, 415 00:25:49,400 --> 00:25:52,399 Speaker 1: kind of similar to how chat GPT will create a 416 00:25:52,440 --> 00:25:56,000 Speaker 1: text response to a prompt you type in a little 417 00:25:56,040 --> 00:25:59,520 Speaker 1: text field. So you could type something up like country 418 00:25:59,640 --> 00:26:03,320 Speaker 1: song in the style of Hank Williams with vocals like 419 00:26:03,440 --> 00:26:08,119 Speaker 1: Billie Eilish about going home after being away for many years, 420 00:26:08,640 --> 00:26:12,240 Speaker 1: and then Boomy would take this prompt and generate a 421 00:26:12,320 --> 00:26:15,679 Speaker 1: musical track for you, and then Boomy would actually release 422 00:26:15,760 --> 00:26:19,840 Speaker 1: that track on streaming services like Spotify. Now that's already 423 00:26:19,880 --> 00:26:23,560 Speaker 1: a bit sus because if you're using styles and voices 424 00:26:23,600 --> 00:26:29,200 Speaker 1: that actually originate with other people without their involvement or consent, 425 00:26:29,720 --> 00:26:32,400 Speaker 1: there's a problem with that. Even if there's not an 426 00:26:32,400 --> 00:26:36,680 Speaker 1: obvious law that you're violating, it's still an ethical issue. 427 00:26:37,000 --> 00:26:41,000 Speaker 1: But don't worry. It gets worse because someone maybe it 428 00:26:41,080 --> 00:26:43,679 Speaker 1: was Boomy, maybe it was one of Boomy's customers, I 429 00:26:43,720 --> 00:26:48,399 Speaker 1: don't know, but someone was trying to boost streams to 430 00:26:48,520 --> 00:26:54,280 Speaker 1: these AI generated songs because Boomy's business model was you 431 00:26:54,359 --> 00:26:57,840 Speaker 1: can create a song. You can use our AI to 432 00:26:57,880 --> 00:27:00,960 Speaker 1: create a song based on your prompts, and then we'll 433 00:27:01,000 --> 00:27:04,520 Speaker 1: post the song to streaming platforms and then we share 434 00:27:04,560 --> 00:27:07,760 Speaker 1: the royalties that are generated from the AI song. So 435 00:27:07,800 --> 00:27:10,680 Speaker 1: if your AI song is really popular, then you get 436 00:27:10,680 --> 00:27:14,080 Speaker 1: a payout, but Boomy takes a cut, So some money 437 00:27:14,080 --> 00:27:16,800 Speaker 1: goes to the user who's prompts served as the starting 438 00:27:16,840 --> 00:27:20,040 Speaker 1: point for that song, and the rest goes to Boomy. Well, 439 00:27:20,080 --> 00:27:24,320 Speaker 1: streaming royalties really don't amount to very much, so if 440 00:27:24,359 --> 00:27:27,280 Speaker 1: you want to start generating royalties, you need to get 441 00:27:27,600 --> 00:27:31,680 Speaker 1: like a crazy number of listens to a particular song. 442 00:27:32,200 --> 00:27:35,360 Speaker 1: So what better way to get a revenue bump than 443 00:27:35,440 --> 00:27:39,919 Speaker 1: to create a bot that artificially hits replay on a 444 00:27:40,000 --> 00:27:43,680 Speaker 1: track to get that number up into the stratosphere. And 445 00:27:44,160 --> 00:27:46,960 Speaker 1: if you think about it, it's a case of robots 446 00:27:47,320 --> 00:27:52,200 Speaker 1: making the music and robots listening to it. How insane 447 00:27:52,359 --> 00:27:56,119 Speaker 1: is that? Now? Artificially running up those numbers hurts everyone 448 00:27:56,160 --> 00:27:59,280 Speaker 1: in the long run, even if the streaming platforms didn't 449 00:27:59,280 --> 00:28:02,720 Speaker 1: pick up on it, and don't worry, they did. Well, 450 00:28:02,840 --> 00:28:05,919 Speaker 1: if you did this long enough the industry would have 451 00:28:05,960 --> 00:28:09,160 Speaker 1: to revisit how royalties are paid out. The whole business 452 00:28:09,160 --> 00:28:13,080 Speaker 1: would change, and ultimately that could hurt legitimate artists in 453 00:28:13,119 --> 00:28:16,359 Speaker 1: the process, you know, artists who aren't relying on bots 454 00:28:16,359 --> 00:28:20,680 Speaker 1: to artificially drive up the popularity of their music. There 455 00:28:20,720 --> 00:28:23,680 Speaker 1: are a lot of negative consequences to that kind of scheme. 456 00:28:24,200 --> 00:28:26,680 Speaker 1: But the platforms have already begun to remove those types 457 00:28:26,680 --> 00:28:29,600 Speaker 1: of tracks in response to suspicious playback numbers. Like there's 458 00:28:29,640 --> 00:28:33,520 Speaker 1: nothing inherently wrong with creating a track like that, at 459 00:28:33,600 --> 00:28:38,520 Speaker 1: least not on a legal standpoint, but you know, illegally 460 00:28:38,920 --> 00:28:43,520 Speaker 1: boosting the numbers, that is an issue. Okay, I've got 461 00:28:43,560 --> 00:28:45,160 Speaker 1: more to say about this, but we're going to take 462 00:28:45,160 --> 00:28:47,560 Speaker 1: another quick break and then we'll get back to doctor 463 00:28:47,640 --> 00:29:02,280 Speaker 1: Hinton's specific concerns with AI. Okay, so I mentioned AI 464 00:29:02,600 --> 00:29:04,640 Speaker 1: in the creative fields of things like, you know, the 465 00:29:04,720 --> 00:29:08,400 Speaker 1: visual arts and music. We've also got the current situation 466 00:29:08,520 --> 00:29:11,240 Speaker 1: here in the United States as I record this, it's 467 00:29:11,280 --> 00:29:13,720 Speaker 1: in May of twenty twenty three, I think I said 468 00:29:13,720 --> 00:29:16,320 Speaker 1: that at the top of the show, and the Writer's 469 00:29:16,360 --> 00:29:21,240 Speaker 1: Guild of America or WGA, is on strike. So this 470 00:29:21,440 --> 00:29:26,680 Speaker 1: union represents TV and film writers, and as they're on strike, 471 00:29:26,800 --> 00:29:29,920 Speaker 1: they cannot do any work in those fields. They can't 472 00:29:29,920 --> 00:29:34,800 Speaker 1: take any meetings, they can't discuss projects, nothing. One of 473 00:29:34,840 --> 00:29:37,320 Speaker 1: their many concerns, it's not the only one, but it 474 00:29:37,360 --> 00:29:40,440 Speaker 1: is one of them, is the role of AI in 475 00:29:40,520 --> 00:29:44,560 Speaker 1: the writing process in Hollywood. So the fear is that 476 00:29:44,640 --> 00:29:47,720 Speaker 1: studios will start to turn to AI in order to 477 00:29:47,760 --> 00:29:50,959 Speaker 1: do stuff like generate script ideas or maybe even a 478 00:29:51,000 --> 00:29:54,840 Speaker 1: first pass at a full story treatment, and then they 479 00:29:54,880 --> 00:29:58,160 Speaker 1: would turn to human writers to polish that idea up 480 00:29:58,240 --> 00:30:03,200 Speaker 1: into something that's con deceivably watchable. But see if you're 481 00:30:03,280 --> 00:30:05,080 Speaker 1: hired to do that, If you're hired to come in 482 00:30:05,120 --> 00:30:07,960 Speaker 1: and do a rewrite or a punch up on a script, 483 00:30:08,520 --> 00:30:11,400 Speaker 1: you make less than you would if you were writing 484 00:30:11,560 --> 00:30:16,440 Speaker 1: a new script from page one. So, in other words, studios, 485 00:30:16,600 --> 00:30:19,320 Speaker 1: the fear is that studios will lean on AI to 486 00:30:19,520 --> 00:30:22,400 Speaker 1: avoid having to pay people to come up with great ideas. 487 00:30:22,960 --> 00:30:25,600 Speaker 1: They'll just use the AI to create ideas, and then 488 00:30:25,680 --> 00:30:29,200 Speaker 1: the humans have to turn these AI generated ideas into 489 00:30:29,280 --> 00:30:33,200 Speaker 1: something that's theoretically going to be a hit, and those 490 00:30:33,240 --> 00:30:35,280 Speaker 1: ideas aren't always going to be great. Now. To be fair, 491 00:30:35,800 --> 00:30:39,440 Speaker 1: the stuff humans make is not a always great see 492 00:30:39,480 --> 00:30:43,120 Speaker 1: also pretty much anything coming out of asylum. That studio 493 00:30:43,160 --> 00:30:46,040 Speaker 1: seems to be run by committee, specifically for the purposes 494 00:30:46,080 --> 00:30:50,720 Speaker 1: of creating trash. But it's it's a real concern, right, 495 00:30:50,760 --> 00:30:54,440 Speaker 1: the worry that, oh, you're going to undercut writers, You're 496 00:30:54,480 --> 00:30:56,560 Speaker 1: going to make it even harder to make a living 497 00:30:56,600 --> 00:30:59,840 Speaker 1: to be a professional writer in Hollywood, because you're going 498 00:30:59,880 --> 00:31:03,760 Speaker 1: to take out one of the more lucrative parts of 499 00:31:03,800 --> 00:31:07,280 Speaker 1: the job by shifting that over to AI, and then 500 00:31:07,400 --> 00:31:11,840 Speaker 1: everyone ends up making even less money while cost of 501 00:31:11,840 --> 00:31:15,280 Speaker 1: living continues to go up, and the studio ends up 502 00:31:15,880 --> 00:31:18,440 Speaker 1: doing it all in the justification of cutting costs and 503 00:31:19,360 --> 00:31:25,160 Speaker 1: increasing profits. It's those kinds of concerns that partly led 504 00:31:25,200 --> 00:31:28,840 Speaker 1: to doctor Hinton to resign from Google. But again that's 505 00:31:28,960 --> 00:31:31,200 Speaker 1: just part of it. There is this concern about how 506 00:31:31,880 --> 00:31:35,600 Speaker 1: AI once was intended to be a thing to augment 507 00:31:36,200 --> 00:31:41,120 Speaker 1: a person's capabilities in their job, but there's this legit 508 00:31:41,360 --> 00:31:43,880 Speaker 1: fear that it could be more of a replacement than 509 00:31:43,880 --> 00:31:47,520 Speaker 1: an augmentation. But there's more to it. So imagine a 510 00:31:47,560 --> 00:31:51,760 Speaker 1: scenario in which an AI agent is not only able 511 00:31:51,840 --> 00:31:57,200 Speaker 1: to design code to build a program, imagine that it's 512 00:31:57,240 --> 00:32:00,800 Speaker 1: also able to execute that code. So it's not just 513 00:32:01,080 --> 00:32:05,280 Speaker 1: creating software, it's able to run that software. Now, imagine 514 00:32:05,280 --> 00:32:09,000 Speaker 1: an AI agent that creates code intended to improve the 515 00:32:09,040 --> 00:32:13,440 Speaker 1: AI itself. Now, this is one of those concepts that's 516 00:32:13,480 --> 00:32:15,719 Speaker 1: really popular in a lot of science fiction, and it 517 00:32:15,720 --> 00:32:20,680 Speaker 1: also shows up in variations of the Singularity. I mentioned 518 00:32:20,800 --> 00:32:23,680 Speaker 1: the singularity recently in an episode, but I didn't define 519 00:32:23,720 --> 00:32:25,880 Speaker 1: it in that episode. So let me do that right now, 520 00:32:25,920 --> 00:32:30,760 Speaker 1: because honestly, you don't hear the terms as frequently as 521 00:32:30,800 --> 00:32:34,880 Speaker 1: you did like a decade ago. But the idea of 522 00:32:34,920 --> 00:32:38,920 Speaker 1: the Singularity goes something like this, We eventually will reach 523 00:32:38,960 --> 00:32:44,160 Speaker 1: a point in technological development where there is a tipping point, 524 00:32:44,200 --> 00:32:47,840 Speaker 1: and after that tipping point, things will be evolving and 525 00:32:47,880 --> 00:32:52,040 Speaker 1: advancing so quickly that it becomes impossible to define the 526 00:32:52,080 --> 00:32:55,200 Speaker 1: present at any given moment, because from one minute to 527 00:32:55,240 --> 00:32:58,560 Speaker 1: the next, so much is advancing and changing that there's 528 00:32:58,680 --> 00:33:03,240 Speaker 1: no like the President is just change. That's it. It's 529 00:33:03,280 --> 00:33:09,880 Speaker 1: an era of incredibly, unimaginably rapid and constant evolution, and 530 00:33:09,920 --> 00:33:14,080 Speaker 1: it will encompass not just our technologies, but potentially even ourselves. 531 00:33:14,720 --> 00:33:19,240 Speaker 1: So some versions of the Singularity incorporate an idea where 532 00:33:19,280 --> 00:33:24,280 Speaker 1: humans integrate technology into themselves and augment their abilities, like 533 00:33:24,360 --> 00:33:31,080 Speaker 1: boosting their intelligence or giving them incredible skills, kind of 534 00:33:31,080 --> 00:33:33,880 Speaker 1: like the matrix idea of I know, kung fu, right, 535 00:33:34,000 --> 00:33:39,080 Speaker 1: like that sort of thing. Some versions of the Singularity 536 00:33:39,240 --> 00:33:42,800 Speaker 1: instead say, nah, humans just kind of get rid of 537 00:33:43,080 --> 00:33:48,160 Speaker 1: our fleshy, mortal bodies and we become digital beings. We 538 00:33:48,200 --> 00:33:53,479 Speaker 1: find a way to transport our consciousness into the digital realm, 539 00:33:53,600 --> 00:33:56,800 Speaker 1: and we become one with machines and potentially one with 540 00:33:56,880 --> 00:34:00,480 Speaker 1: each other. It gets really speculative fictiony when you start 541 00:34:00,480 --> 00:34:04,200 Speaker 1: talking about the Singularity, and I am not convinced that 542 00:34:04,200 --> 00:34:06,440 Speaker 1: that kind of thing is ever going to happen, But 543 00:34:06,560 --> 00:34:09,880 Speaker 1: I do see the potential danger of having machines design 544 00:34:09,960 --> 00:34:13,360 Speaker 1: their own code and then be able to execute that code. 545 00:34:14,239 --> 00:34:17,880 Speaker 1: That could include things like malware that is able to 546 00:34:18,040 --> 00:34:22,720 Speaker 1: bypass antivirus detection because it's built on a new model 547 00:34:22,920 --> 00:34:25,840 Speaker 1: and it's not based off some previous version of malware 548 00:34:25,880 --> 00:34:30,000 Speaker 1: that could potentially be detected. That's a real possibility. It's 549 00:34:30,040 --> 00:34:33,880 Speaker 1: something to really be concerned about. We've also seen already 550 00:34:34,160 --> 00:34:38,640 Speaker 1: with AI hallucinations, how these systems can present misinformation as 551 00:34:38,680 --> 00:34:42,720 Speaker 1: if it's the real deal, and with such unintended consequences 552 00:34:43,360 --> 00:34:48,479 Speaker 1: in a pretty innocent application of AI, you are left 553 00:34:48,520 --> 00:34:52,120 Speaker 1: to wonder what kind of problems would occur with coding? Right, 554 00:34:52,160 --> 00:34:54,359 Speaker 1: we've already seen a problem that can occur just with 555 00:34:54,840 --> 00:34:58,520 Speaker 1: simple text based interactions. What kind of problems could occur 556 00:34:58,600 --> 00:35:02,240 Speaker 1: if we start to depend upon A to build code? Now, 557 00:35:03,120 --> 00:35:04,640 Speaker 1: I think in a lot of cases we would just 558 00:35:04,800 --> 00:35:07,920 Speaker 1: end up with bad code. Like we'd have stuff that works, 559 00:35:07,960 --> 00:35:10,360 Speaker 1: but we'd also have stuff where, for some reason or another, 560 00:35:10,800 --> 00:35:13,640 Speaker 1: the AI introduced code that doesn't do anything, or it 561 00:35:13,760 --> 00:35:16,000 Speaker 1: causes the whole thing to crash, and so we just 562 00:35:16,080 --> 00:35:19,520 Speaker 1: end up with software that doesn't really work in those cases, 563 00:35:20,160 --> 00:35:23,880 Speaker 1: But there's enough doubt there to make us pause, Like 564 00:35:24,040 --> 00:35:26,960 Speaker 1: maybe the code would work, but maybe it would do 565 00:35:27,040 --> 00:35:32,520 Speaker 1: something malicious or ultimately harmful. I'm assuming that's how doctor 566 00:35:32,600 --> 00:35:36,319 Speaker 1: Hinton feels based on his statements post resignation. I don't 567 00:35:36,320 --> 00:35:38,520 Speaker 1: want to put words into his mouth, but this is 568 00:35:38,560 --> 00:35:41,920 Speaker 1: kind of the what I'm inferring based upon what he said. 569 00:35:42,560 --> 00:35:45,400 Speaker 1: Hinton is worried that deep neural networks and similar machine 570 00:35:45,480 --> 00:35:49,640 Speaker 1: learning techniques could be put to use in harmful, aggressive ways. 571 00:35:49,680 --> 00:35:52,319 Speaker 1: I mean this dates back to his decision to try 572 00:35:52,320 --> 00:35:55,799 Speaker 1: and avoid taking funding from the Department of Defense. Right, 573 00:35:56,239 --> 00:35:59,279 Speaker 1: he's worried about the stuff like AI controlled machines that 574 00:35:59,280 --> 00:36:02,360 Speaker 1: could be used in warfare, and we've already seen some 575 00:36:02,480 --> 00:36:06,279 Speaker 1: elements of that with things like drones. So it's a 576 00:36:06,360 --> 00:36:09,680 Speaker 1: reasonable fear to have, and a lot of experts in 577 00:36:09,719 --> 00:36:14,080 Speaker 1: AI have struggled with this and have you know, campaigned 578 00:36:14,640 --> 00:36:18,920 Speaker 1: to have kind of bands put on for AI controlled 579 00:36:19,239 --> 00:36:23,239 Speaker 1: weaponry and warfare materials. And there's a real fear that 580 00:36:23,360 --> 00:36:26,160 Speaker 1: in some countries, the push to create such tools will 581 00:36:26,160 --> 00:36:28,160 Speaker 1: be very hard to avoid, that there won't be these 582 00:36:28,280 --> 00:36:31,799 Speaker 1: checks in place or people concerned. It will be more 583 00:36:31,840 --> 00:36:34,520 Speaker 1: of a just an overall drive to develop those kinds 584 00:36:34,520 --> 00:36:38,520 Speaker 1: of tools and to thus dominate by having those tools 585 00:36:38,560 --> 00:36:41,600 Speaker 1: in your arsenal. That can lead to a situation where 586 00:36:41,640 --> 00:36:45,279 Speaker 1: everybody else rushes to weaponize AI because they're worried that 587 00:36:45,800 --> 00:36:48,200 Speaker 1: everyone else is already doing it and that they're going 588 00:36:48,239 --> 00:36:51,240 Speaker 1: to get left behind and thus be in a vulnerable position. 589 00:36:51,560 --> 00:36:54,560 Speaker 1: So it becomes kind of a self fulfilling prophecy. And 590 00:36:54,600 --> 00:36:57,000 Speaker 1: in that case, it's the AI experts that we have 591 00:36:57,040 --> 00:37:00,279 Speaker 1: to rely on to push back against that Trendeople who 592 00:37:00,320 --> 00:37:03,680 Speaker 1: are actually building the systems. We have to hope that 593 00:37:03,800 --> 00:37:08,440 Speaker 1: they will do so in a way that won't perpetuate harm. 594 00:37:08,800 --> 00:37:14,440 Speaker 1: But that's a big hope to place on that particular 595 00:37:14,480 --> 00:37:18,480 Speaker 1: group of people. Now, not everyone is as worried about AI, 596 00:37:18,600 --> 00:37:22,160 Speaker 1: at least not in the short term. Stanford researchers recently 597 00:37:22,200 --> 00:37:26,240 Speaker 1: published a paper titled are emergent abilities of large language 598 00:37:26,239 --> 00:37:30,160 Speaker 1: models a mirage? So, an emergent ability refers to a 599 00:37:30,200 --> 00:37:33,719 Speaker 1: system developing some sort of skill or function for which 600 00:37:33,880 --> 00:37:37,080 Speaker 1: it was not formally trained or programmed to do so. 601 00:37:37,200 --> 00:37:40,680 Speaker 1: For example, let's say you train a large language model 602 00:37:40,840 --> 00:37:44,040 Speaker 1: to answer questions that are posed in English, but then 603 00:37:44,080 --> 00:37:47,839 Speaker 1: you find out that it's also able to translate responses 604 00:37:47,880 --> 00:37:51,240 Speaker 1: into Spanish perfectly, even though you didn't design the system 605 00:37:51,320 --> 00:37:56,600 Speaker 1: to quote unquote understand Spanish. The researchers at Stanford concluded 606 00:37:57,000 --> 00:38:01,040 Speaker 1: that these apparent emergent abilities are in fact mirages. They 607 00:38:01,040 --> 00:38:04,600 Speaker 1: are not real, They are illusions. So the researchers are 608 00:38:04,600 --> 00:38:08,080 Speaker 1: saying that companies like Google and open ai might look 609 00:38:08,120 --> 00:38:10,440 Speaker 1: at the results of a model and then use a 610 00:38:10,480 --> 00:38:14,359 Speaker 1: metric that suggests the ability that was displayed was an 611 00:38:14,400 --> 00:38:18,080 Speaker 1: emergent one. But if they had chosen a different metric, 612 00:38:18,560 --> 00:38:20,640 Speaker 1: if they had looked at the output from a different 613 00:38:20,680 --> 00:38:24,440 Speaker 1: point of view. In other words, the illusion of emergent 614 00:38:24,560 --> 00:38:28,520 Speaker 1: behavior would fade. So, in other words, it just depends 615 00:38:28,520 --> 00:38:31,080 Speaker 1: on how you look at it, whether it looks like, oh, 616 00:38:31,200 --> 00:38:33,359 Speaker 1: this system is doing something it wasn't designed to do, 617 00:38:33,560 --> 00:38:35,680 Speaker 1: or oh, by looking at it this way, we see 618 00:38:35,680 --> 00:38:39,120 Speaker 1: it's performing exactly as it was designed to do. So 619 00:38:39,160 --> 00:38:42,080 Speaker 1: we're getting real obi wan kenobi here with a certain 620 00:38:42,120 --> 00:38:47,080 Speaker 1: point of view. Stuff, all right, But how worried should 621 00:38:47,120 --> 00:38:50,920 Speaker 1: we be about AI? Sadly, I think the answer to 622 00:38:50,960 --> 00:38:53,640 Speaker 1: that is really complicated. I wish I could just give 623 00:38:53,680 --> 00:38:59,239 Speaker 1: you a definitive answer from terrified to mift, But I 624 00:38:59,239 --> 00:39:01,719 Speaker 1: think it largely depends upon who you are and what 625 00:39:01,760 --> 00:39:04,600 Speaker 1: you do for a living, and how much you depend 626 00:39:04,880 --> 00:39:08,640 Speaker 1: upon automated technology. Honestly, I think, for example, that folks 627 00:39:08,680 --> 00:39:11,640 Speaker 1: who write code have a legitimate reason to be concerned, 628 00:39:12,040 --> 00:39:13,880 Speaker 1: not because I think AI is going to do their 629 00:39:13,960 --> 00:39:17,720 Speaker 1: job better, but rather because I have very little faith 630 00:39:18,200 --> 00:39:22,440 Speaker 1: in software company executives to avoid the temptation to push 631 00:39:22,480 --> 00:39:26,440 Speaker 1: their chips in on a big, long shot bet on AI. So, 632 00:39:26,480 --> 00:39:28,839 Speaker 1: in other words, I worry that business leaders are going 633 00:39:28,880 --> 00:39:31,440 Speaker 1: to make some poor decisions in an effort to cut 634 00:39:31,480 --> 00:39:36,080 Speaker 1: costs and maximize efficiency, and then get rid of human 635 00:39:36,239 --> 00:39:38,839 Speaker 1: engineers and rely on AI to build code, and then 636 00:39:38,840 --> 00:39:40,800 Speaker 1: we're going to end up with a really rough period 637 00:39:40,800 --> 00:39:45,839 Speaker 1: of subpar software. Now, in the long run, we might 638 00:39:45,960 --> 00:39:51,080 Speaker 1: either see companies that previously had discarded their human programmers 639 00:39:51,200 --> 00:39:53,839 Speaker 1: return to them and say, gosh, it turns out, yeah, 640 00:39:53,880 --> 00:39:56,880 Speaker 1: we need you because what the stuff that AI is 641 00:39:56,920 --> 00:40:00,520 Speaker 1: making is not consistent or good quality. But then again, 642 00:40:00,560 --> 00:40:03,240 Speaker 1: we might see AI generated code improved to a point 643 00:40:03,239 --> 00:40:07,560 Speaker 1: where you know, it is superior to what humans were making. 644 00:40:07,600 --> 00:40:10,919 Speaker 1: It's really impossible to say right now, we can't say 645 00:40:10,960 --> 00:40:12,840 Speaker 1: for sure which direction it's going to go in, and 646 00:40:12,880 --> 00:40:15,239 Speaker 1: it may even be more messy than that. Right It 647 00:40:15,280 --> 00:40:18,399 Speaker 1: might be in some cases the code generated by AI 648 00:40:18,560 --> 00:40:21,160 Speaker 1: is superior and in other cases it's inferior. It may 649 00:40:21,200 --> 00:40:25,760 Speaker 1: not be, you know, an industry wide thing we can 650 00:40:26,480 --> 00:40:29,640 Speaker 1: have a firm statement about, and maybe I should put 651 00:40:29,640 --> 00:40:32,040 Speaker 1: more faith in companies. I just know I've seen a 652 00:40:32,040 --> 00:40:35,319 Speaker 1: lot of decisions and a lot of different organizations over 653 00:40:35,320 --> 00:40:39,520 Speaker 1: the years that have proven to be really short sighted 654 00:40:39,719 --> 00:40:43,320 Speaker 1: strategies and ultimately harmful all in the name of returning 655 00:40:43,360 --> 00:40:45,720 Speaker 1: shareholder value. So I guess you could call me a cynic, 656 00:40:46,239 --> 00:40:48,000 Speaker 1: but I just feel like I've seen it a lot, 657 00:40:48,120 --> 00:40:50,720 Speaker 1: so I would not be surprised, and in fact, that 658 00:40:50,719 --> 00:40:57,120 Speaker 1: that IBM CEO statement of potentially filling around seveny eight 659 00:40:57,160 --> 00:41:00,400 Speaker 1: hundred jobs total, I think is the real estimate with 660 00:41:00,640 --> 00:41:05,000 Speaker 1: AI instead of with humans. That kind of speaks to 661 00:41:05,800 --> 00:41:10,520 Speaker 1: my worries. I do not think that we are on 662 00:41:10,560 --> 00:41:13,600 Speaker 1: the precipice of AI spiraling out of control and becoming 663 00:41:14,200 --> 00:41:19,560 Speaker 1: this malevolent superhuman intelligence that's going to ultimately decide to 664 00:41:19,560 --> 00:41:22,840 Speaker 1: get rid of all of us. However, that's just my opinion, 665 00:41:23,400 --> 00:41:25,880 Speaker 1: and Goodness knows, I don't have the experience or the 666 00:41:25,880 --> 00:41:29,440 Speaker 1: expertise of someone like doctor Henton, so I'm taking this 667 00:41:30,040 --> 00:41:35,040 Speaker 1: from what you could argue is a largely uninformed opinion. 668 00:41:35,320 --> 00:41:38,400 Speaker 1: I think that's a fair assessment. I do think AI 669 00:41:38,760 --> 00:41:42,080 Speaker 1: is posing real problems right here and now, and that 670 00:41:42,120 --> 00:41:44,560 Speaker 1: we have to consider those problems and we need to 671 00:41:44,600 --> 00:41:48,240 Speaker 1: address them, either in how we are developing and deploying AI, 672 00:41:48,800 --> 00:41:51,600 Speaker 1: or how we create legislation to protect humans who otherwise 673 00:41:51,640 --> 00:41:54,480 Speaker 1: might see their livelihood threatened. I do think we need 674 00:41:54,520 --> 00:41:58,200 Speaker 1: to revisit right to personality and right to publicity style 675 00:41:58,320 --> 00:42:01,480 Speaker 1: laws to make sure that the laws incorporate things like 676 00:42:01,880 --> 00:42:04,799 Speaker 1: deep fake video and audio and in the form you know. 677 00:42:04,920 --> 00:42:07,800 Speaker 1: Right now, we have protections in place if someone uses 678 00:42:08,040 --> 00:42:13,000 Speaker 1: your likeness without your permission. It's very specific rules about that. 679 00:42:13,160 --> 00:42:16,560 Speaker 1: It's not like if your image pops up, you know, 680 00:42:16,640 --> 00:42:18,840 Speaker 1: because someone took a photo and you happen to be 681 00:42:18,880 --> 00:42:21,680 Speaker 1: in the background. It's not like that's a case that 682 00:42:21,719 --> 00:42:25,000 Speaker 1: you're going to have a really strong, you know, legal 683 00:42:25,040 --> 00:42:27,839 Speaker 1: backing on if you want to protest the use of it. 684 00:42:28,440 --> 00:42:31,520 Speaker 1: But let's say you're a celebrity and someone runs your 685 00:42:31,560 --> 00:42:35,120 Speaker 1: image next to a product that you did not agree 686 00:42:35,160 --> 00:42:38,080 Speaker 1: to endorse. There are protections for that, but those protections 687 00:42:38,120 --> 00:42:41,720 Speaker 1: are largely for just likenesses like your image. It doesn't 688 00:42:41,760 --> 00:42:44,920 Speaker 1: necessarily cover things like the sound of your voice or 689 00:42:45,000 --> 00:42:49,080 Speaker 1: the style of music you create. The laws need to 690 00:42:49,080 --> 00:42:54,520 Speaker 1: be rewritten or tweaked in order to cover those cases, because, 691 00:42:54,560 --> 00:42:56,920 Speaker 1: I mean, it's a new world where that sort of 692 00:42:56,920 --> 00:43:00,839 Speaker 1: thing is possible. So right now, there's not like there's 693 00:43:00,840 --> 00:43:04,720 Speaker 1: no recourse for someone who hears a song that sounds 694 00:43:04,760 --> 00:43:08,440 Speaker 1: like they sang it, but they didn't there's nothing really 695 00:43:08,520 --> 00:43:10,560 Speaker 1: they can do, and you have to be careful with 696 00:43:10,600 --> 00:43:12,920 Speaker 1: how you word such laws because there are things like 697 00:43:13,440 --> 00:43:17,400 Speaker 1: you know, parody being protected by fair use. So if 698 00:43:17,440 --> 00:43:20,600 Speaker 1: you wanted to create a parody of a song and 699 00:43:20,680 --> 00:43:24,320 Speaker 1: you hire someone who sounds kind of like the musical 700 00:43:24,400 --> 00:43:27,799 Speaker 1: artist you're parodying, that shouldn't necessarily be illegal. But if 701 00:43:27,800 --> 00:43:29,640 Speaker 1: you're trying to pass it off as if it were 702 00:43:30,200 --> 00:43:34,960 Speaker 1: the artists themselves, that's a different story. So yeah, it's complicated, 703 00:43:35,000 --> 00:43:37,880 Speaker 1: it's messy. It's complicated not just on the tech side, 704 00:43:38,120 --> 00:43:43,480 Speaker 1: but on the legislative side, the cultural side, military as well. 705 00:43:44,400 --> 00:43:48,600 Speaker 1: I do think that doctor Hinton has some legitimate concerns. 706 00:43:48,880 --> 00:43:51,520 Speaker 1: I think some of them, at least I hope anyway, 707 00:43:51,880 --> 00:43:56,520 Speaker 1: are a little premature. I hope it's the things he's 708 00:43:56,520 --> 00:43:58,960 Speaker 1: worried about are far enough out into the future that 709 00:43:59,000 --> 00:44:05,760 Speaker 1: we can actually steps to prevent bad outcomes and negative consequences. 710 00:44:06,200 --> 00:44:08,440 Speaker 1: We'll never prevent all of them, because some of them 711 00:44:08,440 --> 00:44:11,480 Speaker 1: will be completely unintended, but I would like to see 712 00:44:11,480 --> 00:44:14,319 Speaker 1: them minimized at the very least in the meantime. I'm 713 00:44:14,360 --> 00:44:18,120 Speaker 1: not going to panic about AI, but I'm giving it 714 00:44:18,160 --> 00:44:22,480 Speaker 1: a lot of side eye so it knows it needs 715 00:44:22,520 --> 00:44:26,719 Speaker 1: to stay in line. That's it. I hope you are 716 00:44:26,840 --> 00:44:30,799 Speaker 1: all well, and I'll talk to you again really soon. 717 00:44:36,719 --> 00:44:41,360 Speaker 1: Tech Stuff is an iHeartRadio production. For more podcasts from iHeartRadio, 718 00:44:41,680 --> 00:44:45,400 Speaker 1: visit the iHeartRadio app, Apple Podcasts, or wherever you listen 719 00:44:45,440 --> 00:44:50,279 Speaker 1: to your favorite shows.